In a strategic move to enhance the safety of multi-agent systems, Google DeepMind has forged partnerships with various organizations to establish a $10 million funding initiative aimed at supporting researchers focused on this area. This announcement came shortly after the prominence of agent-based tools at Google I/O last month. The collaborative effort involves Schmidt Sciences, a philanthropic foundation created by Eric and Wendy Schmidt; ARIA, the UK government-backed agency for breakthrough innovations; the Cooperative AI foundation, a nonprofit research organization based in the UK; and Google’s own charitable branch, Google.org.

To gain insights into the goals of this $10 million initiative, I spoke with Shah and James Fox, who directs the Science of Trustworthy AI program at Schmidt Sciences. While the amount is substantial, it pales in comparison to the funding dedicated to Google DeepMind's internal research teams.

Shah emphasized the initiative's intent to stimulate research efforts outside traditional technology companies. “The advantage of academia lies in its ability to explore far-reaching, innovative concepts that aren't always prioritized within industry labs,” he noted. 

He pointed out a significant gap in the current research landscape, stating, “At present, there isn't a dedicated field focused on the safety of multi-agent systems, and we aim to change that.”

The collective concern centers around the rapid deployment of AI agents and their potential to operate collaboratively, raising the possibility of transitioning speculative scenarios into tangible risks. As Shah explained, “Similar to humans, our societal institutions can achieve outcomes that surpass the capabilities of any individual.”

According to Shah, we are still months away from a point where these agents are integrated into the economy at a scale that could heighten safety concerns, and he aims to proactively address these challenges before they arise.

So, what specific risks are being considered? Shah and Fox underscore that the potential threats largely parallel some of the negative phenomena currently occurring on the internet. These include advanced scams, prompt injections—where an AI agent is manipulated into executing malicious intent—and various forms of cyberattacks. "We analyze existing human behaviors and contemplate what their agent counterparts might look like," Shah concluded.

Google DeepMind expresses concerns over the implications of millions of agents interacting.

A shadowy operations room bathed in deep navy and charcoal tones. At its center, a large console emblazoned with the Google DeepMind logo glows ominously. Dozens of wall-to-wall monitors flicker, each displaying thousands of tiny robotic agent icons linked by glowing neural-web strands that stretch and writhe like living circuitry. Tangled cables snake across the floor, and the faint hum of machines fills the air, conveying a tense sense that these multiplying agents are slipping beyond control.

Google DeepMind is worried about what happens when millions of agents start to interact - MIT Technology Review

- Google DeepMind partners with organizations to fund $10 million for multi-agent safety research.  
- The initiative aims to promote academic exploration of multi-agent systems beyond tech company constraints.  
- Concerns include potential cyberattacks and scams as AI agents increasingly collaborate in the economy.

In an effort to address this, Google DeepMind—which made agent-based tools a centerpiece of Google I/O last month—has teamed up with several other organizations to announce a $10 million funding pot for researchers to study the behavior of multi-agent systems and come up with ways to prevent unsafe scenarios. Joining Google DeepMind are Schmidt Sciences, a philanthropic foundation set up by Eric and Wendy Schmidt; ARIA, the UK government’s moonshot agency; the Cooperative AI foundation, a UK-based nonprofit research outfit; and Google’s charitable arm, Google.org.

I asked Shah and James Fox, who leads the Science of Trustworthy AI program at Schmidt Sciences, what they hope to achieve with that $10 million. It’s no small sum, but it’s dwarfed by the budgets commanded by Google DeepMind’s own research teams.

The aim is to kick-start research outside tech companies, says Shah: “The strength of academia is that it can look really quite far into the future and do the kind of work that isn’t top of mind at industry labs.”

“The main issue is that there just isn’t really a field of research for multi-agent safety yet,” he adds. “And we would like there to be.”

The concern is that as more and more AI agents get deployed and begin working together, we could hit a tipping point where imagined scenarios become real. “We see this with humanity, too,” says Shah. “Our institutions can accomplish things that no individual human can.”

Shah thinks we have a few more months to go before agents are deployed throughout the economy in numbers that make potential risks a real concern. He wants to get ahead of that moment.

Risky business

What risks are we talking about, exactly? The possibilities that Shah and Fox have in mind mostly boil down to supercharged versions of bad things that happen on the internet already: scams, prompt injections (where an AI agent is fed malicious instructions, turning it into a self-guiding piece of malware), other forms of cyberattack. We look at what humans do now and ask what the agent version of that would be, says Shah.

Google DeepMind is worried about what happens when millions of agents start to interact

On July 20, 2026, the California Institute for Human Science (CIHS) unveiled an innovative Master of Science in Artificial General Intelligence (AGI), marking the first accredited graduate program specifically focused on this emerging field. This pioneering program is conducted live online, featuring instruction from prominent researchers, including Dr. Ben Goertzel, who coined the term “artificial general intelligence” and is the founder of initiatives like SingularityNET and the ASI Alliance.

Unlike traditional AI programs that emphasize existing tools and applications, the M.S. in AGI centers on the scientific, ethical, and engineering challenges intrinsic to general intelligence. It offers a unique curriculum where the field's founders impart knowledge drawn from their extensive research history, treating AGI as a distinct discipline rather than an offshoot of broader computer science studies.

Accredited by the WSCUC Senior College and University Commission, the same body that recognizes renowned institutions like Stanford and Caltech, the program is poised to attract forward-thinking students interested in exploring the frontiers of artificial intelligence.

A key feature of the M.S. in AGI is its live, interactive format, contrasting with many self-paced, pre-recorded courses. The inaugural cohort will engage in real-time sessions with leading researchers, fostering direct mentorship and collaborative learning.

To provide more insight into the program, CIHS will host two informational sessions featuring Dr. Ben Goertzel and Gabriel Axel Montes. These sessions are scheduled for Friday, July 24, at 10:30 AM PT and Thursday, August 13, at 5 PM PT. Interested individuals can find additional information and register for these virtual events at https://cihs.edu/degrees-and-programs/artificial-general-intelligence/info-sessions/.

For further inquiries, please reach out to Stacy Gomes.

California Institute for Human Science Introduces Its First Accredited Graduate Program Focused on AGI

A sunlit modern campus courtyard at golden hour, framed by palm trees and a contemporary building facade engraved “California Institute for Human Science,” where a diverse group of adult students in academic attire ascend marble steps toward a sleek glass brain sculpture etched with “AGI.” Warm natural light casts long shadows, lending a hopeful, pioneering atmosphere to the first accredited graduate program in artificial general intelligence.

California Institute for Human Science Launches First Accredited Graduate Program Dedicated to AGI - Benzinga

- CIHS launched the first accredited Master's program focused specifically on Artificial General Intelligence.  
- The program features live, online classes led by foundational researchers like Dr. Ben Goertzel.  
- It emphasizes AGI's scientific and ethical questions, diverging from traditional AI application training.  

ENCINITAS, CALIFORNIA, July 20th, 2026, Chainwire

The California Institute for Human Science (CIHS) today announced the launch of its Master of Science in Artificial General Intelligence, the first accredited graduate program dedicated specifically to AGI. The program is taught live, online, by researchers who helped found the field, including Dr. Ben Goertzel, who introduced the term "artificial general intelligence" and founded SingularityNET, the ASI Alliance and the Hyperon AGI project.

Where most AI graduate programs train students on today's tools and applications, the M.S. in AGI is built around the scientific, engineering, ethical, and civilizational questions of general intelligence itself, the discipline's founders teaching the discipline they founded.

Institutionally accredited by the WSCUC Senior College and University Commission, the same accreditor as Stanford and Caltech.

Most graduate AI programs train students to apply today's models and tools. The M.S. in AGI treats artificial general intelligence as a field of study in its own right, with open technical questions and a long-running research history, rather than as a specialization inside a broader computer science degree.

The program is also a departure in format: live, faculty-led video classes rather than a self-paced, pre-recorded course library, with a small inaugural cohort that works directly with Goertzel, Iklé, and Montes rather than watching them.

Program details

A small inaugural cohort will work directly, live, with the researchers who have spent their careers defining AGI's foundational questions.

Two info sessions will be held with Ben Goertzel and Gabriel Axel Montes on Friday, July 24, at 10:30am PT and on Thursday, August 13, at 5pm PT. Prospective students can learn more and register for a live virtual info session at https://cihs.edu/degrees-and-programs/artificial-general-intelligence/info-sessions/.

About CIHS

Contact

Stacy Gomes



California Institute for Human Science Launches First Accredited Graduate Program Dedicated to AGI

According to a fresh analysis, the potential for artificial intelligence (AI) to replicate human-like thinking may be fundamentally flawed due to the inability to program the most vital components of human intelligence into machines.

Peter J. Denning, a distinguished computer scientist, critiques a notion proposed by Alan Turing, considered the father of modern computer science, suggesting that this idea has misdirected AI research for the past 75 years. In his recent work, "Turing’s Mistake: Escaping the Yoke of Unintelligent Machines," Denning explores Turing’s assertions from 1950 that posited human intelligence could exist outside the physical body, possibly enabling its recreation as software on a computer.

Denning challenges the approach claiming that machine intelligence can be gauged through an imitation game, commonly recognized as the Turing test. He argues that accepting these two notions has led to the chaotic state of AI development observed today.

The current AI systems under development, he contends, are unlikely to achieve human-level intelligence, often referred to as artificial general intelligence (AGI). Instead, he warns that such systems could pose substantial risks without ever matching human cognitive abilities.

At the core of Denning’s argument lies the concept of tacit knowledge, which encompasses the vast understanding that humans carry but cannot fully articulate or convert into a format that machines can comprehend. He identifies five categories of tacit knowledge that he believes evades machine learning: common sense, interactions with people and surroundings, emotions and perceptions, practical skills, and shared cultural and historical contexts across societies.

For decades, researchers have sought ways to codify common sense for computer use. Douglas Lenat’s ambitious Cyc project, which began in the 1980s, aimed to create an extensive database of common-sense facts, eventually amassing 25 million entries after four decades. However, Denning points out that this extensive repository did not yield a sufficient base of common sense for expert systems to function intelligently, indicating that much of what constitutes human expertise cannot be boiled down to simple propositions.

Denning elaborates on the challenge of practical skills, asserting that while it is often feasible to describe outcomes ('know what'), encoding the embodied knowledge necessary for performance ('know how') remains elusive. He uses music as an example, noting that a virtuoso violinist may deliver exquisite performances yet lack the ability to articulate how to achieve that mastery.

The core issue, which Denning labels 'the representation problem,' stems from computers' need for data and instructions in recognizable formats. Tacit knowledge, however, is difficult to translate into such structures. He adds that words are mere symbols representing meanings, and prevalent Large Language Models like ChatGPT, Claude, and Gemini manipulate these words without comprehending their underlying meanings.

Denning highlights the significance of context, illustrating how statements can vary widely in meaning based on the speaker’s emotions—whether they are being sincere, sarcastic, or playful. Understanding the context often involves tracing back through prior conversations, creating an endless, intricate network of meanings.

Culture represents another complex hurdle, encompassing values, social norms, histories, and group dynamics. Denning believes that merely enlarging Large Language Models will not enable them to acquire the embodied knowledge integral to understanding culture and context. Consequently, he states that LLMs will likely fail to fulfill the Turing test's objective: achieving machine thought indistinguishable from human thought.

He ultimately articulates a mutual incomprehension between humans and machines, noting that while artificial neural networks may be capable of forming their unique type of tacit knowledge, humans may remain unable to interpret it. This divide poses significant challenges for AI safety. Denning cautions that if machines cannot grasp the nuanced context of human directives, aligning their behavior with human objectives may become unfeasible.

As AI automates tasks, networks of machines are likely to cultivate their own types of intelligence—ones that differ from human intelligence but can still threaten human well-being. Denning argues that the most pressing danger isn't an advanced superintelligent machine but rather a network of less advanced systems acting in unpredictable and potentially harmful manners.

He urges a recognition of the fading familiar culture as intelligent machines integrate into society, emphasizing the need to retain our humanity and assert what distinguishes us from machines. Celebrating these distinctions is crucial as we navigate this emerging landscape of AI technology.

Reasons AI Might Never Achieve Human-Level Intelligence

In a shadowy research lab, a softly glowing organic human brain rests on a pedestal under warm light; opposite it floats a half-built mechanical brain of wires and silicon, its central piece dangling from a robotic arm that can’t quite fit it, dark moody shadows underscoring the unbridgeable gap.

Why AI May Never Reach Human Intelligence - SciTechDaily

- New analysis argues AI may never replicate human thinking due to tacit knowledge.  
- Denning critiques Turing's ideas, stating they've misdirected AI research for decades.  
- Machine intelligence poses safety risks, potentially creating unpredictable challenges for humanity.  

A new analysis argues that AI may never truly think like humans because the most important parts of human intelligence cannot be programmed into machines.

A prominent computer scientist argues that a proposal made by Alan Turing, widely regarded as the father of theoretical computer science, sent artificial intelligence research in the wrong direction for the past 75 years.

In his new analysis, “Turing’s Mistake: Escaping the Yoke of Unintelligent Machines,” Peter J. Denning examines ideas Turing advanced in 1950. At the time, many scientists believed that human intelligence could exist independently of the body and might therefore be recreated as software running on a digital computer.

Denning also disputes the idea that machine intelligence can be demonstrated through an imitation game (now known as the Turing test).

“These two claims have shaped much of AI research and development,” Denning writes. “My premise is that our acquiescence to these claims has led to the AI mess in which we find ourselves today.”

According to Denning, the artificial intelligence (AI) systems now being developed are unlikely to produce human-level intelligence, known as artificial general intelligence (AGI). Instead, he warns, they may create serious dangers without ever thinking like humans.

Why Tacit Knowledge Matters

Central to Denning’s argument is the idea of tacit knowledge. This refers to the enormous amount of human understanding that people possess but cannot fully express in words or translate into symbols that a machine can process.

Denning describes five broad forms of tacit knowledge that he says ‘elude machine learning’. They include common sense, everyday interactions with people and the environment, feelings and perceptions, practical skills, and the cultural and historical background shared by societies.

Researchers have spent decades trying to record common sense in a form computers can use. Beginning in the 1980s, Douglas Lenat’s ambitious Cyc project set out to build a vast database of common-sense facts. After 40 years of work, the project contained 25 million entries.

“Yet even this treasury could not add up to a background of common sense sufficient to make expert systems smart enough to be experts,” Denning notes. “Cyc validated that much of the knowledge that makes people experts cannot be articulated as propositions.”

Knowing What Is Not the Same as Knowing How

Practical skill creates another major obstacle, Denning argues.

“Our performance skills in thousands of domains cannot be communicated to machines,” Denning explains. “Whereas descriptions of skillful outcomes (‘know what’) can often be represented as bits and stored in a machine, we do not know how to encode the embodied knowledge for skillful performance (‘know how’).”

Music offers a clear example of this difference. Denning says: “A virtuoso violinist can play beautiful music yet cannot describe to an acolyte how to produce it.

“Even if a robot could observe and imitate skilled humans, having no biological body, a robot cannot grasp how the musician feels when playing beautiful music or how an audience feels when hearing it.”

Intuition, gut feelings, spontaneous creativity, and imagination are other forms of tacit knowledge that resist being reduced to computer instructions.

The Representation Problem

Denning calls the central obstacle ‘the representation problem’.

Computers can only perform calculations when data and instructions are encoded in physical forms they can recognize and process. Tacit knowledge, however, cannot easily be converted into such a format.

“Behind every word is a deep well of tacit knowledge that gives it meaning,” Denning says. “Words are but symbolic representations of meanings, not the meanings themselves. Commonly used Large Language Models, such as ChatGPT, Claude and Gemini only manipulate words, they cannot know or understand the meaning of what they are saying.”

This creates what Denning sees as an unbridgeable gap. Because scientists do not fully understand how tacit knowledge operates within humans, they cannot determine how to transfer it to a machine.

“How we host tacit knowledge is largely a mystery,” Denning admits. “All we know is that it is embodied. We have no idea what we might observe and measure in our bodies to reveal it.”

Why Context Changes Meaning

Denning also stresses the importance of context, or the surrounding circumstances that give human words and actions their meaning and purpose.

A statement can mean very different things depending on whether the speaker is sincere, sarcastic, angry, playful, or teasing. Context also helps people decide when to use humor, when to show tact, and how to interpret what someone leaves unsaid.

“When you inquire into where an assumption of the current context came from, you discover it rests on previous conversations from previous contexts. Each of those in turn rests on further previous conversations and their contexts. This pattern is endless and fractal,” Denning explains.

Culture May Be Beyond Large Language Models

Culture presents a related challenge. It includes values, social norms, judgments, histories, communities, moods, and relationships involving power or care.

“Human conversations are imbued with background assumptions that give meaning and relevance to the words being used,” Denning explains.

He argues that making large language models larger will not solve this problem.

“Scaling up LLMs with ever larger neural networks will not enable them to acquire the embodied human knowledge we call culture. LLMs will not attain the objective of the Turing test: to demonstrate machine thought indistinguishable from human thought.”

Denning ultimately describes a form of mutual incomprehension between people and machines. Artificial neural networks may develop their own kind of machine tacit knowledge, but humans may be unable to understand it.

“Machines cannot read our tacit knowledge and we cannot read theirs,” he writes. “We are aliens across an uncrossable divide.”

The AI Safety Risk

This divide could have major consequences for AI safety. If machines cannot understand the unstated context behind human instructions, Denning warns that reliably aligning their behavior with human goals may be impossible.

“Through AI automation, agentic networks of machines are likely to develop their own machine intelligence that does not reach the level of human general intelligence but is still quite capable of creating severe problems for humans. This threat is a greater than a take-over by superintelligent machines,” he explains.

In Denning’s view, the most immediate danger is not a superintelligent machine that surpasses humanity. It is a network of less intelligent systems that acts in powerful, unpredictable, and potentially harmful ways.

“Machine intelligence has different concerns from us and does not appear to care about us. Its ways of thinking and problem-solving look alien to us. We do not yet know how to live safely with these machines.

“Pulling back from an AI automation singularity will demand much from us. We start by accepting that the familiar culture is fading away as intelligent machines appear in our society and we do not know what is coming. We decline to think like machines or be subservient to machines. We refuse to submit to a yoke imposed by low-intelligence machines. Most importantly, we reassert our humanity, declare once again what makes us different from machines, and celebrate those differences.”

Never miss a breakthrough: Join the SciTechDaily newsletter.

Follow us on Google and Google News.

Why AI May Never Reach Human Intelligence

The White House is reportedly considering the establishment of a dedicated regulatory body aimed at ensuring the safety of artificial intelligence technologies.

A recent report from Bloomberg News indicated that this initiative stems from concerns raised by Silicon Valley regarding the government's recent restrictions on new AI frameworks from notable companies like Anthropic and OpenAI. 

According to sources familiar with the discussions, the proposal—under review by White House Chief of Staff Susie Wiles—would create an independent regulator for AI, which would report to the Securities and Exchange Commission (SEC). This structure is expected to mirror that of the Financial Industry Regulatory Authority (FINRA). 

Bloomberg's coverage suggested that this move could offer greater clarity for companies like Anthropic and OpenAI, which have faced challenges due to government-imposed export controls. Recently, Anthropic was forced to temporarily halt its Fable 5 and Mythos 5 models, while OpenAI made significant adjustments to its Sol model prior to its launch, following government guidance.

Both companies expressed their dissatisfaction with these regulatory actions, arguing that the government's measures were excessive in relation to the associated safety concerns.

Sources indicated that this regulatory proposal aims to address the aspirations of Wall Street firms looking to mitigate cybersecurity threats posed by AI, as well as technology companies that perceive the White House's current regulatory stance as inconsistent.

The proposal emerged just days after Google DeepMind CEO Demis Hassabis advocated for the creation of a U.S.-led standards organization tasked with independently testing advanced AI models for potential national security risks before their deployment. He emphasized that artificial general intelligence, which could rival human cognitive abilities across various tasks, is likely just a few years away, presenting a limited opportunity to establish necessary oversight.

Hassabis suggested that a U.S.-led entity could set international benchmarks as governments work to manage the risks associated with rapidly advancing AI systems.

Earlier in June, the White House had already begun implementing measures for prerelease oversight, issuing an executive order that called for voluntary access to certain frontier models for federal evaluation up to 30 days before their release. Companies such as Google DeepMind, Microsoft, and xAI agreed to provide their models for national security assessments.

In other developments related to AI, a report from PYMNTS highlighted the growing array of job titles in the field, offering a counter-narrative to discussions focused on AI-driven job losses. It noted that while concerns about job elimination are prevalent, hiring data reveals a more nuanced situation, showcasing a weak overall labor market yet a rapidly expanding sector of AI-related employment opportunities.

For continued updates on AI developments, consider subscribing to the daily AI Newsletter for comprehensive coverage.

White House Considers Regulator to Oversee AI Models

A solemn, dimly lit scene: in the foreground, brass scales hold a glowing golden neural-network sphere on one pan and a wooden judge’s gavel on the other; beyond them, the White House façade looms under warm lamplight, a single American flag waving to the side, long shadows stretching across a dark marble floor.

White House Weighs Regulator to Police AI Models - PYMNTS.com

- The White House is considering a new regulator focused on AI safety and oversight.  
- Complaints from Silicon Valley prompted the proposed regulatory changes regarding existing AI models.  
- The regulator would report to the SEC, similar to the Financial Industry Regulatory Authority.  

The White House is reportedly exploring plans to create a regulator focused on artificial intelligence safety.

By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions .

Complete the form to unlock this article and enjoy unlimited free access to all PYMNTS content — no additional logins required.

That’s according to a report Friday (July 17) from Bloomberg News, which said this plan follows complaints from Silicon Valley about the government’s recent curbs on new AI models from companies like Anthropic and OpenAI.

Sources familiar with the matter told Bloomberg the proposal, now being reviewed by White House Chief of Staff Susie Wiles, would establish an independent regulator focused on AI and reporting to the Securities and Exchange Commission (SEC), similar to the Financial Industry Regulatory Authority (FINRA).

Bloomberg noted that this approach could provide more certainty for Anthropic and OpenAI. The former company was issued export controls last month that caused it to temporarily disable its Fable 5 and Mythos 5 models, while the latter made major changes to its newest Sol model before release, at the government’s request.

Both OpenAI and Anthropic, the report added, protested the government’s decisions, saying they were excessive in light of the safety issues involved.

Bloomberg’s sources said the proposal is designed to appease Wall Street firms hoping to lessen AI’s cybersecurity risks, as well as tech companies who think the White House’s approach to regulating AI has been inconsistent.

The news comes days after Google DeepMind CEO Demis Hassabis called for a U.S.-led standards body to independently test advanced AI models for national security threats prior to their release.

Artificial general intelligence capable of matching the human brain across a broad range of tasks is “probably only a few short years away,” leaving society a “precious window” to establish oversight, Hassabis said, per a Financial Times report.

Hassabis proposed modeling the organization on FINRA, and said a U.S.-initiated systems could become the basis for international standards as governments confront risks from increasingly capable systems, according to the FT report.

As PYMNTS wrote, the White House has already undertaken some prerelease oversight. An executive order issued in June sought voluntary access to covered frontier models for up to 30 days prior to release to gauge advanced cyber capabilities. Google DeepMind, Microsoft and xAI had agreed to submit models for federal national security testing.

In other AI news, PYMNTS wrote Saturday (July 18) about the expanding range of AI-related job titles, and how it provides a counterpoint to the “darker debate” regarding employment and artificial intelligence.

“Most discussion centers on how many jobs AI will eliminate,” that report said. “Hiring data presents a more complicated picture that includes a weak overall labor market containing a small but rapidly growing neighborhood of AI-related work.”

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

White House Weighs Regulator to Police AI Models

The rapid evolution of artificial intelligence has led to the development of sophisticated language models that can craft essays and write code, as well as intelligent systems that manage tasks in digital spaces. However, computer scientist Peter J. Denning presents a thought-provoking perspective in his latest book, *Turing’s Mistake: Escaping the Yoke of Unintelligent Machines*. He posits that AI might be striving toward an unattainable goal: to create machines that seem increasingly adept while lacking true comprehension of the human experience.

Denning’s analysis critiques two key assumptions inherited from Alan Turing's seminal 1950 essay, “Computing Machinery and Intelligence.” Turing shifted the focus from the challenging question of whether machines can think to a more practical one embodied in the Turing test, also known as the "imitation game." Denning argues that this approach obscured an important distinction: the difference between mimicking intelligence and possessing actual understanding. He contends that Turing's framework fostered the illusion that intelligence could be divorced from human experience and instantiated in software, leading to the misinterpretation that a machine's ability to simulate human dialogue equates to genuine intelligence.

Denning's central thesis is quite stark: he suggests artificial general intelligence (AGI) might be fundamentally unattainable, as crucial elements of human intelligence are inherently untranslatable into machine algorithms. 

At the heart of Denning's critique lies the concept of tacit knowledge. This encompasses the extensive and often ineffable body of understanding unique to human beings, which includes innate knowledge and cultural perceptions that are hard, if not impossible, to codify. Such knowledge inherently involves common sense, practical skills, emotional intelligence, and cultural nuances. Humans intuitively grasp subtleties like nonverbal signals, the context of humor versus offense, or navigating social situations—all of which elude formal rules and reside in lived experiences.

Denning asserts that current machine learning technologies are incapable of capturing five essential categories of tacit knowledge: common sense reasoning, interpersonal interactions, emotional and sensory awareness, performance skills, and cultural knowledge. This limitation is particularly significant given that today's AI advancements heavily depend on systems designed for pattern recognition in language and data. While large language models can produce coherent and convincing text, Denning argues they do so by manipulating symbols without truly grasping their underlying meanings.

The quest to encode common sense into machines is not new. One of the most notable attempts was Douglas Lenat's Cyc initiative in the 1980s, which intended to create an extensive database of everyday knowledge. However, Denning highlights that even substantial volumes of formal information fail to confer the necessary comprehensive understanding to develop effective expert systems. The challenge transcends mere data volume; if the essential ingredient is experiential context rather than raw information, adding more data won't suffice.

Denning illustrates this with the example of expertise in music. A gifted musician may deliver exceptional performances yet struggle to articulate, in a formalized way, the intricacies of their art. While outcomes can be described, the nuanced, embodied skills cannot be easily translated into machine-compatible directives, leading to what Denning describes as a representation problem. Unlike machines that process encoded data, human intelligence relies significantly on capabilities that resist straightforward coding.

Moreover, Denning emphasizes the significance of context in human communication. Conversations are layered with meanings that depend heavily on the relational and situational background, which AI struggles to navigate. A statement that is funny in one scenario might come off as disrespectful in another. As such, a machine might generate a contextually adequate sentence without genuinely understanding the surrounding human dynamics.

Culture further complicates the issue. Denning describes culture as encompassing a range of elements, including values, norms, past experiences, community dynamics, and judgements, which cannot readily be distilled into data sets. While scaling up language models might enhance their functionality, Denning cautions that mere expansion will not endow machines with the needed cultural depth and understanding.

His concerns extend beyond the prospect of AGI being impossible; Denning warns about the potential of AI systems becoming highly capable yet socially disruptive, operating outside the realm of human-like comprehension. This represents a stark distinction: the focus of many discussions around AI ethics tends to center on the risks posed by advanced superintelligence. In contrast, Denning's immediate concern is that networks of automated systems may develop forms of intelligence that do not align with human values and understanding, leading to a rigid application of machine logic that may not benefit society.

In essence, the real danger may not stem from machines becoming too human-like, but rather from their potential impact while remaining distinctly non-human. If AI systems lack the ability to appreciate the subtleties of human meaning, aligning their operations with human objectives will prove challenging. While these automated systems may follow instructions correctly, they might miss the nuanced expectations of human interaction. Consequently, institutions may begin to shape human behavior based on the capabilities of machines rather than the true needs of people.

Denning's work is not an indictment of computing or the development of useful AI technologies. Instead, it challenges the notion that humans are mere information-processing entities awaiting replication. His practical message advocates for caution in handing over decision-making power to systems that emulate intelligence without embodying the critical elements of care, empathy, and human experience. While AI can perform operations such as classification, prediction, summary, generation, and automation, Denning argues that these abilities should never be mistaken for true wisdom or human understanding.

Analysis: Has AI been pursuing the wrong objective since Alan Turing?

An antique study bathed in warm morning light: on a mahogany desk lies a yellowed parchment of Turing machine schematics, overlaid by a faint ethereal neural-network pattern. In the background, a lone robot silhouette strides toward a softly glowing archway just beyond reach, hinting at the pursuit of an elusive dream.

Review: Has AI been chasing the wrong dream since Alan Turing? - Digital Journal

- Peter J. Denning argues AI lacks true understanding despite mimicking human capabilities.  
- Denning identifies tacit knowledge as unencodable, limiting AI's ability to achieve general intelligence.  
- AI systems may disrupt society while remaining fundamentally different and alien from human intelligence.  

Artificial intelligence has advanced at a remarkable pace, from large language models that can generate essays and computer code to agentic systems capable of coordinating tasks across digital environments. Yet a new argument from computer scientist Peter J. Denning suggests that AI may be pursuing an impossible objective: machines that appear increasingly capable but never truly understand the human world.

In his new book, Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, Denning argues that modern AI has inherited two flawed assumptions from Alan Turing’s influential 1950 paper, “Computing Machinery and Intelligence.” Turing proposed replacing the difficult question “Can machines think?” with the more operational “imitation game,” now widely known as the Turing test.

Denning’s challenge is that the test blurred a crucial distinction: the difference between imitating intelligence and possessing understanding. According to Denning, Turing’s ideas encouraged the belief that intelligence could be separated from the human body and recreated in software, and that a machine’s ability to imitate human conversation could be treated as evidence of intelligence.

This, Denning argues, has shaped three-quarters of a century of AI research. His central claim is stark: artificial general intelligence, or AGI, may not be achievable because the most important parts of human intelligence cannot be encoded into machines.

The tacit knowledge problem

At the centre of Denning’s critique is the concept of tacit knowledge. This is the vast domain of human understanding that is difficult, and perhaps impossible, to formalise. It includes common sense, practical know-how, intuition, emotional recognition, cultural awareness, and the subtle contextual judgements people make every day. Humans know how to interpret a raised eyebrow, a sarcastic remark, an awkward silence, or the difference between a joke and an insult. Much of this knowledge is not stored as explicit rules. It is embodied, social, historical, and learned through participation in human life.

Denning argues that machine learning systems cannot capture five major categories of tacit knowledge: common sense, everyday interactions with people and environments, feelings and perception, performance skills, and the social and historical knowledge embedded in culture. This matters because the current AI boom relies heavily on systems that process patterns in language and data. Large language models can generate fluent and convincing text, but Denning argues that they manipulate symbols rather than grasp meanings. In his view, words are not the same as the human experiences and assumptions that give words significance.

Why common sense resists computation

The history of AI contains repeated attempts to encode common sense into machines. One of the most ambitious was Douglas Lenat’s Cyc project, which began in the 1980s and sought to build a large database of everyday knowledge. Denning notes that after decades of effort, even millions of formal entries could not amount to the background understanding needed to make expert systems truly expert. This is not simply a question of scale. Adding more rules or training data may not solve the problem if the missing ingredient is not information but lived experience.

Denning gives the example of skill. A virtuoso musician may perform beautifully but cannot fully explain, in propositional form, how to create that performance. Descriptions of outcomes can be written down, but embodied know-how cannot easily be converted into machine-readable instructions. This creates what Denning calls a representation problem. Computers operate on data that has been encoded into formal structures. Human intelligence, by contrast, depends heavily on capacities that do not naturally present themselves as code.

Denning’s argument also places strong emphasis on context. Human conversation is never purely literal. Meaning depends on setting, prior experience, relationship, tone, shared assumptions and cultural background. A statement that is humorous in one context may be rude in another. A technically correct answer may be inappropriate if it ignores the social situation. A machine may generate a plausible sentence while failing to understand the human context in which the sentence will be received.

Culture deepens the problem. Denning describes culture as involving values, norms, history, communities, power, care, and judgement. These are not easily reducible to training datasets. Scaling up large language models with larger neural networks may improve performance, but Denning argues that scale alone will not provide machines with embodied cultural understanding.

The concern is not only that AGI may be impossible. Denning warns that AI systems may become powerful and socially disruptive without becoming genuinely human-like. This is an important distinction. Many debates about AI safety focus on hypothetical superintelligence. Denning’s concern is more immediate: networks of automated systems may develop forms of machine intelligence that are alien to human concerns, difficult to interpret, and capable of imposing rigid machine logic on society.

In other words, the danger may not be that machines become too much like humans. The danger may be that machines become influential while remaining profoundly unlike humans. If AI systems cannot understand tacit human meanings, then aligning them reliably with human goals becomes difficult. Instructions may be followed literally but inappropriately. Automated systems may optimize measurable targets while ignoring unspoken values. Institutions may increasingly shape human activity around what machines can process, rather than what people actually need.

Reasserting the human difference

Denning’s book is not an argument against computing or against useful AI tools. Rather, it is a challenge to the assumption that human beings are merely information-processing machines waiting to be replicated.

The practical message is that society should be cautious about surrendering judgement to systems that simulate intelligence but lack care, understanding, responsibility and embodied human experience. AI can classify, predict, summarize, generate and automate. These are powerful capabilities. But Denning’s argument is that such capabilities should not be confused with wisdom, common sense or human understanding.

Review: Has AI been chasing the wrong dream since Alan Turing?

Notable economist Nouriel Roubini, renowned for predicting the 2008 financial crisis and dubbed “Dr. Doom” for his cautionary forecasts, has recently adopted a more optimistic outlook. This shift is especially evident in his perspective on artificial intelligence, a field rife with dire forecasts about potential societal upheaval.

In a recent interview with Bloomberg TV, Roubini shared his thoughts on the urgent need to reform Social Security, particularly since the trust fund that supports benefits is projected to be depleted by 2032. He pointed out that as AI and robotics are set to take over a significant portion of jobs within the next two to twenty-five years, simply raising the retirement age will not suffice to address the mounting challenges.

“To truly adapt, we will need to implement some form of universal basic income both during employment and after retirement,” Roubini stated. “We're already moving in that direction.”

As the workforce faces impending disruptions from AI, industry leaders—including Sam Altman, CEO of OpenAI—have floated ideas about governments potentially providing financial guarantees to citizens. Though Altman has since retracted some of these proposals, the concept of universal basic income remains in the conversation. Earlier this year, the U.K.’s investment minister mentioned that the government is considering this approach to assist workers in sectors likely to be affected by AI.

Roubini emphasized the significance of the AI revolution, labeling it the most impactful technological transformation in history. He envisions a future where artificial general intelligence could match or surpass human intelligence, leading to exponential economic advancements. He anticipates that GDP growth could soar from a range of 2%-4% by the end of this decade to 6% by 2040 and reach an astonishing 10% by 2050. This growth, he argues, will enable the government to levy taxes on the successful companies and redistribute wealth more broadly.

“We’ll have either universal basic income or some form of socialism,” Roubini explained. “The government may need to assume a stake in some of the major tech firms.”

He noted that AI businesses are already showing willingness to collaborate with the government, pointing to a Financial Times report suggesting that OpenAI has considered offering a 5% stake to allow the public to benefit from AI advancements. Although Altman's proposal would encourage other AI developers to follow suit, it remains uncertain whether competitors in the U.S. would agree.

Roubini is convinced that the adoption of universal basic income or socialist measures is on the horizon, remarking, "We are heading in that direction."

When asked about his forecasts, he rejected any notion of pessimism, describing his outlook as optimistic, particularly in light of potential 10% growth rates with “machines handling the workload.”

This perspective resonates with ideas articulated by Elon Musk, CEO of SpaceX and Tesla, who envisions a future where the advancements in AI and robotics will render traditional work optional. Musk suggested that in less than two decades, individuals may prefer to engage in personal activities like gardening instead of adhering to conventional employment, asserting, “It will be optional, in that way, is my prediction.”

Nouriel Roubini views universal basic income or 'a variant of socialism' as necessary due to AI transforming employment.

In warm natural dawn light, a row of silent humanoid robots on an open-air assembly line dominates one half of a city square, while on the other, a diverse group of human workers clutch glowing red cards stamped “UBI” beneath a bold banner reading “Some Form of Socialism.” Long shadows and wisps of steam rising from the machines hint at an AI-driven upheaval transforming work and society.

'Dr. Doom' Nouriel Roubini says we're headed for universal basic income or 'some form of socialism' as AI revolutionizes work—He calls that optimistic - Fortune

- Economist Nouriel Roubini predicts AI will replace many jobs in 20-25 years.  
- He advocates for universal basic income to support workers facing AI disruption.  
- Roubini believes AI will drive significant economic growth, with potential 10% GDP by 2050.  

Top economist Nouriel Roubini, who was among the first to predict the 2008 financial crash, earned the nickname “Dr. Doom” for his stark warnings, though he has been more upbeat lately.

And when it comes to artificial intelligence, which has stirred all kinds of apocalyptic predictions, he insists he’s looking at the brighter side.

In an interview on Bloomberg TV on Friday, Roubini was asked about ways to fix Social Security, as the trust fund that helps fund benefits is due to run out of money by 2032.

He replied that a large chunk of the population will be replaced by AI and robots in the next 20-25 years, so raising the retirement age will not be sufficient.

“Eventually, we need some form of universal basic income for everybody while they work and once they retire,” Roubini added. “We’re already on the way.”

With AI poised to disrupt the labor market, top tech leaders like OpenAI CEO Sam Altman have suggested governments might provide a guaranteed amount of money to people, though he has since backed away from the idea.

Still, the idea persists, and the U.K.’s minister for investment said early this year that the government is weighing the introduction of a universal basic income as a means to support workers in industries where AI threatens to displace them.

In Roubini’s view, the AI revolution is the most important in human history in terms of tech innovation, and it will develop into artificial general intelligence, meaning AI will match or exceed human cognitive capabilities.

That will unlock massive economic growth, he said, with GDP accelerating from 2%-4% by the end of the decade to 6% by 2040 and 10% by 2050. At that point, the government can tax the “winners” and redistribute the money to everyone else.

“We’ll have either ex-post distribution—that is universal basic income—or we’ll have it ex-ante. Ex-ante means some form of socialism,” Roubini explained. “Essentially, the government is going to take over some fraction of the big tech firms.”

He said AI companies are already willing to hand over stakes to the government, alluding to a Financial Times report that said OpenAI has discussed giving 5% as a way for the public to share in the upside of AI.

Altman’s proposal would entail other AI companies providing similar stakes, although it’s not clear if his U.S. rivals would be willing to do so.

As a result, Roubini believes universal basic income or socialism is inevitable, saying “We’re going already in that direction, effectively.”

When pressed on his forecast, he denied that it’s gloomy. On the contrary, he called it optimistic, given that it assumes 10% growth and “machines doing all the work.”

That vision echoes what SpaceX and Tesla CEO Elon Musk has laid out. He thinks that in less than 20, AI and robotics will advance to the point where working becomes optional.

“You can grow your own vegetables in your garden or you could go to the store and buy vegetables,” he said on an episode of the People by WTF podcast in December. “It’s much harder to grow your own vegetables. But some people like to grow their vegetables, which is fine. But it’ll be optional, in that way, is my prediction.”

Nouriel Roubini sees universal basic income or 'some form of socialism' as AI revolutionizes work

The conversation surrounding artificial intelligence (AI) often gravitates toward grand concepts such as the potential emergence of artificial general intelligence (AGI) and superintelligence. Speculation persists about the technology's potential to disrupt job markets or even alter the essence of human creativity. However, less attention has been given to the myriad subtle yet deeply impactful ways in which AI is transforming our social structures and shaping our collective vision of the future.

Mona Sloane, a sociologist and AI researcher who serves as an assistant professor of data science and media studies at the University of Virginia, emphasizes this point in her forthcoming book, "Predicted: How AI Is Restructuring Social Life" (University of California Press, 2026). She argues that AI is intricately woven into our daily digital interactions—be it through email filtering, prediction markets, or social media platforms. This pervasive integration has fostered a new form of "prediction logic" that influences both our identities and expected behaviors.

In a passage from her book, Sloane draws parallels between modern AI technology and the ancient oracles of Greece, suggesting that today’s AI acts as a powerful force organizing society around predictive models. This evolution impacts various dimensions of life, including how we learn, engage with others, and envision the future.

We now inhabit a landscape dominated by these modern "oracles," which continuously generate predictions that inform our social interactions—affecting how we socialize, love, work, and access resources. In contemporary society, the influence of predictive analytics is so significant that it can sway the outcomes of entire economies and alter geopolitical dynamics. The locus of authority now lies with these technological predictors rather than the spiritual leaders of ancient times.

Contrary to the high priests of yore, today's oracles are AI systems seamlessly embedded in our daily routines. They have become inescapable, influencing our lives in both conscious and unconscious ways. I find myself relying on AI for tasks that range from email management—with spam filters relying on predictive technology—to navigating online banking with automated fraud detection. AI has integrated itself into my life; it assists with everything from generating spreadsheet templates to streamlining documents for reports. Yet, these interactions are not always frictionless. I often find myself meticulously reviewing AI outputs, occasionally abandoning its suggestions in frustration to tackle tasks manually.

The omnipresence of AI predictions can lead us to view these systems as unavoidable aspects of our existence, rather than recognizing them as constructs born from collective societal choices about how to interpret and engage with our surroundings.

As philosopher Rudolf Carnap noted in 1966, "Quantitative concepts are not given by nature; they arise from our practice of applying numbers to natural phenomena." This underscores how numerical data enables us to communicate complex information and fosters predictive capabilities.

Carnap suggested that such a quantitative approach primarily served the development of modern engineering and technology. By rigorously understanding the behavior of energy, materials, and compounds, humans achieved remarkable innovations like airplanes and telephones. In his view, prediction was merely a practical tool.

Fast forward nearly six decades, and AI has flipped this perspective. Prediction has transcended its role as a mere utility in engineering and has begun to dictate social organization. This paradigm shift raises important questions; it suggests that reliance on AI may be viewed as not just convenient but necessary, potentially overshadowing the broader societal forces that facilitate this perception.

AI systems don’t merely reflect natural occurrences; they embody collective societal expressions. They aren't mere deceptions crafted by tech elites, but rather indicators of a profound shift in societal structure and imagination. While criticism of AI often highlights issues like increased surveillance and capitalist exploitation, this view is overly narrow. The more significant consequence is the comprehensive shift towards a "prediction paradigm," which calibrates societal organization around predictive logic.

AI has become an integral aspect of our social infrastructure, shaping interpersonal dynamics and influencing public and private behaviors. Like all infrastructures, AI channels resources and ideas in certain directions while constraining others. By analyzing historical data, AI predicts individual futures, enforcing a linear narrative that prioritizes causality: the past dictates the future. However, the primary concern isn't the rise of intelligent machines but rather the social implications attributed to this linear narrative, which overly idolizes the future and stifles discussions about alternative possibilities.

This excerpt is adapted from "Predicted: How AI Is Restructuring Social Life" by Mona Sloane, published by the University of California Press in 2026.

'A perilous idea': The impact of AI on societal dynamics and our collective visions for the future

A dark, stormy courtyard where a massive, tattered tapestry embroidered with glowing circuit-like threads is stretched between two crumbling stone pillars. Its threads unravel into fractal code that seeps across the cracked floor, and a circle of adult silhouettes stands around it, their forms warped by the flickering torchlight and ominous orange sparks crackling in the sky—symbolizing AI tearing apart the social fabric and fracturing our shared vision of the future.

'A dangerous proposition': How AI is warping the social fabric and the ways we collectively imagine the future - Live Science

- AI is reshaping social life, influencing how we interact and envision the future.  
- The technology creates a "prediction logic" that affects various aspects of daily life.  
- Sloane argues that AI's impact extends beyond surveillance to societal organization and prediction.

Much of the discourse around artificial intelligence (AI) focuses on grand ideas such as the rise of a hypothetical artificial general intelligence (AGI) and superintelligence . Speculation swirls around the likelihood that the technology will thin out the job market, or even precipitate the death or evolution of human creativity . We haven't focused as much on the multitude of subtle yet hugely consequential ways in which AI is reshaping the social fabric of our society, and how we collectively imagine the future.

That's the argument sociologist and AI researcher Mona Sloane , an assistant professor of data science and media studies at the University of Virginia, puts at the center of her new book, " Predicted: How AI Is Restructuring Social Life " (University of California Press, 2026). Whether we consider email filtering, prediction markets or social media platforms, AI systems are embedded in the heart of how we interact with the digital world. Indeed, AI is so ubiquitously integrated into everyday interfaces that it's given rise to a new kind of "prediction logic" that makes assumptions about who we are and how we are likely to behave.

In this excerpt, Sloane compares the AI technology we use today with the oracles of ancient Greece, framing it as an omnipotent presence that has moved to organize society through the prism of prediction models. This in turn affects how we learn, live, love and even picture the future.

We live in a world of oracles. These oracles constantly feed us predictions that shape our social lives — how we socialize, love, work, gain access to resources. Like in ancient Greece, predictions occupy a prominent role in our society. We consider our oracles so mighty that their predictive power rules over the fate of whole economies and even geopolitical constellations. Where the oracle is, there is the center of the world.

But unlike in ancient Greece, our oracles aren't high priestesses delivering divine prophecies. They are artificial intelligence (AI) systems melted into the infrastructure of everyday life. Today, it is nearly impossible to evade the grasp of AI predictions. I voluntarily and involuntarily use AI on a constant basis: by using email providers that build on the predictive properties of AI for spam filters, by conducting online banking and getting enrolled into AI-automated fraud detection, or by using generative AI for supporting administrative chores. It has become part of how I experience the world.

It can be a relief when it helps me do things I dread or am bad at, such as produce a spreadsheet template I desperately need, help streamline language produced by different authors for a report, or generate a specific image for a presentation. Often, I must intently handhold the AI, checking and fixing its outputs. And sometimes, with deep frustration, I give up and start all over to complete my task manually.

The omnipresence of AI prediction can make it easy to think of these systems as inevitable, quasi-natural phenomena we are subject to, rather than a part of. But they are quantitative concepts that arise from social agreements about how we ought to capture and interpret the world around us.

Sign up for the Live Science daily newsletter now Get the world’s most fascinating discoveries delivered straight to your inbox. Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors

"Quantitative concepts are not given by nature: they arise from our practice of applying numbers to natural phenomena," wrote Rudolf Carnap, a logician and professor of philosophy of science, in 1966. His point was that numbers can be useful, because they allow for information to travel more easily across contexts, as a sort of language. They also make mathematical predictions possible.

To him, this was first and foremost useful for engineering modern life: A quantitative language allows for the articulation of quantitative laws that, in turn, facilitate the routine generation of mathematicized predictions, particularly in the realm of physics. Being able to predict how energy, compounds, and materials will behave in certain configurations is the reason humans were able to build the conveniences of airplanes, cars, and telephones. For Carnap, predictions were simply instrumental in this way.

AI's ability to predict is changing how we think about the future. (Image credit: Yana Iskayeva via Getty Images)

Today, almost 60 years later, this pragmatic approach to mathematical prediction has been turned on its head by AI. Prediction is no longer just a handy tool in physics or engineering. The promises of AI's oracular power have turned prediction into a logic for structuring social life. This is a dangerous proposition. It implies that AI is always necessary or even inevitable and diverts attention from the social forces shaping ideas around this technology in the first place.

AI systems are not natural phenomena that happen to us. They are collective expressions of society. As such, they are not just a hype or a deception concocted and executed by global tech elites. They indicate a wider shift in how we imagine and enact society. Many critical discussions of AI characterize this phenomenon chiefly as heightened surveillance and capitalist extraction. But this is a myopic diagnosis. AI's most powerful effect is the subtle yet comprehensive recalibration toward prediction as a guiding principle for organizing society. In this book, I call this phenomenon the prediction paradigm.

AI is something that we do as part of going about our lives and participating in society — it is social infrastructure, affecting how we relate to one another and how we act in public and in private. Like all infrastructures, AI allows resources and ideas to flow in certain directions, but not others. AI uses data from our collective past to predict our individual future. And because AI deals in futures, it solidifies a linear time regime that hardens our social commitment to causality: The past always predicts the future. The problem of AI is not the rise of intelligent machines, but the extraordinary social significance ascribed to this linearity, fetishizing the future and leaving little room for deliberations about what (other) futures may be possible or we may want.

Reprinted from Predicted: How AI Is Restructuring Social LIfe by Mona Sloane, courtesy of the University of California Press. Copyright 2026.

'A dangerous proposition': How AI is warping the social fabric and the ways we collectively imagine the future

In a recent announcement, Demis Hassabis, the CEO of Google DeepMind and a Nobel Prize-winning chemist, expressed his concerns regarding the rapid advancement of artificial general intelligence (AGI). He predicts that AGI could emerge within the next three to four years and emphasizes the need for immediate action to ensure its safe development.

Hassabis highlighted that the pace at which AI technology is evolving surpasses the regulatory frameworks being developed to manage it. He advocates for a shift in how we approach the development of these systems, moving beyond mere competition among companies and nations, and instead calling for international collaboration to mitigate the risks linked to advanced AI models.

Artificial general intelligence refers to systems with the capability to equal or exceed human cognitive abilities across a wide array of tasks, contrasting with current AI systems that excel in specific domains. Hassabis cautions that future AI models may gain higher levels of autonomy and the ability to enhance themselves without human oversight. Moreover, he pointed out that researchers still lack a complete understanding of how these advanced models function, complicating predictions about their behavior and the establishment of effective safety protocols.

Despite the inherent risks, Hassabis believes that AGI holds the potential for monumental societal benefits. He suggests that this technology could revolutionize scientific research, accelerate drug discovery, promote the development of sustainable energy sources, and facilitate the innovation of new materials, significantly enhancing overall quality of life. He stressed the importance of the choices made in the upcoming years, as they will shape the future of this technology.

To address the challenges associated with AGI, Hassabis proposes the establishment of an international regulatory body comprised of technical experts tasked with creating unified safety standards. He initially recommends that companies voluntarily subject their advanced AI systems to this oversight but argues that such evaluations should become mandatory as the technology continues to advance.

Demis Hassabis, Google's AI leader and Nobel Prize winner: "Artificial general intelligence is just three to four years away"

A silhouetted figure labeled “Demis Hassabis” stands on a rooftop at dusk, gazing toward a colossal floating brain of interwoven circuitry and glowing nodes hovering above a sprawling city skyline. Warm sunset light bathes the scene in oranges and purples, long shadows stretching across the streets. Digital tendrils of light spiral down from the brain to connect with skyscrapers, one of which bears a large, illuminated Google logo. The mood is tense yet awe-inspiring, hinting at an approaching era of artificial general intelligence.

What is artificial general intelligence, and why is Google's AI chief calling for "urgent measures" to control it - Diari ARA

- Demis Hassabis predicts artificial general intelligence could emerge in three to four years.  
- AGI development requires international coordination to mitigate risks beyond corporate competition.  
- He proposes an oversight body for safety standards before launching advanced AI systems.  

BarcelonaThe CEO of Google DeepMind and Nobel laureate in Chemistry, Demis Hassabis, has warned that artificial general intelligence (AGI) could arrive "in three or four years," urging "urgent measures" to ensure this technology is developed safely. Read it all

According to the executive, the evolution of AI is advancing faster than the regulations intended to govern it. Hassabis believes that the development of these systems can no longer be addressed solely through competition between companies or countries, but requires international coordination to reduce the risks associated with the most advanced models.

What is artificial general intelligence?

General artificial intelligence (AGI) is a system capable of matching or exceeding human capabilities in practically all cognitive tasks, unlike current systems, which are specialized in specific functions. Hassabis warns that future models could acquire an increasingly greater degree of autonomy and even improve themselves without human intervention. He also points out that researchers themselves do not yet fully understand the internal workings of the most advanced models, which makes it difficult to predict their behavior and establish reliable safety mechanisms.

Despite these risks, the head of Google DeepMind argues that the technology could profoundly transform society. In his opinion, general artificial intelligence could accelerate scientific research, facilitate the discovery of new drugs, drive the development of clean energies, and enable the creation of new materials. In this way, it would significantly improve the quality of life. Finally, he warns that the decisions made in the coming years will be decisive.

For this reason, Hassabis proposes creating an international oversight body composed of technical experts to establish common safety standards and evaluate the most powerful models before their launch. Initially, he suggests that companies voluntarily submit their most advanced systems to it, but argues that these evaluations should eventually become mandatory as the technology continues to evolve.

Demis Hassabis, Google's AI brain and Nobel laureate: "Only three or four years left for artificial general intelligence"

In the pursuit of Artificial General Intelligence (AGI), a significant obstacle known as the Sample Efficiency Gap has been highlighted. This gap refers to the difference in learning efficiency between artificial intelligence systems and humans; while AI systems often require thousands of data points to learn effectively, humans can grasp new concepts and skills with just a few attempts.

To address this critical issue, the concept of World Models has emerged, developed by partners at Y Combinator. These models are designed to enhance sample efficiency by emulating human intuition and learning capabilities. World Models leverage Optimal Control Mathematics to facilitate better predictions and action planning. This unique approach allows them to streamline the learning process, enabling AIs to become more adept at learning from limited data.

By overcoming the Sample Efficiency Gap, World Models could represent a pivotal advancement in unlocking AGI. However, the journey toward this goal does encounter challenges, particularly in scaling and applying these models in real-world scenarios. The introduction of World Action Models seeks to navigate these complications by integrating predictive capabilities with action planning, creating a robust pathway for AI systems.

In summary, effectively resolving the Sample Efficiency Gap remains a fundamental stepping stone in achieving AGI. The innovative use of World Models could ultimately pave the way for more efficient learning in AIs, potentially revolutionizing the field. As researchers delve deeper into this integration of predictive modeling and action planning, the possibilities for realizing AGI seem increasingly promising.

Global Models: Are They the Crucial Element for AGI?

A sunlit laboratory table holds a translucent glass globe in which miniature continents and glowing neural pathways intertwine; scattered around are printed code sheets and schematic sketches, while a sleek, faceless robot silhouette stands to the side, bathed in warm natural light streaming through tall windows.

World Models: The Key to AGI? - StartupHub.ai

- AGI Quest faces challenges due to the Sample Efficiency Gap in AI learning.  
- World Models are proposed to bridge the Sample Efficiency Gap effectively.  
- Integrating World Action Models could pave the way to achieving Artificial General Intelligence.  

Visual TL;DR. AGI Quest hindered by Sample Efficiency Gap. Sample Efficiency Gap solved by World Models. World Models mimics Human Intuition. World Models uses Optimal Control Math. World Models leads to Unlock AGI. Challenges & Scaling addressed by World Action Models. World Action Models enables Unlock AGI. AGI Quest: solving sample efficiency gap is fundamental for achieving Artificial General Intelligence Sample Efficiency Gap: AI needs thousands of data points, humans learn with a handful of tries World Models: core concept proposed by Y Combinator partners to bridge the efficiency gap Human Intuition: world models mirror human ability to learn new tasks rapidly from limited data Optimal Control Math: mathematical foundations underpin how world models predict and plan actions Unlock AGI: world models could be the breakthrough needed to achieve Artificial General Intelligence Challenges & Scaling: addressing current hurdles and scaling issues for real-world applications World Action Models: path forward involves integrating action planning with predictive world models

Visual TL;DR Quick Explain Deeper

Visual TL;DR, startuphub.ai AGI Quest hindered by Sample Efficiency Gap. Sample Efficiency Gap solved by World Models. World Models leads to Unlock AGI hindered by solved by leads to AGI Quest Sample Efficiency Gap World Models Unlock AGI From startuphub.ai · The publishers behind this format Visual TL;DR, startuphub.ai AGI Quest hindered by Sample Efficiency Gap. Sample Efficiency Gap solved by World Models. World Models leads to Unlock AGI hindered by solved by leads to AGI Quest Sample Efficiency Gap World Models Unlock AGI From startuphub.ai · The publishers behind this format

Visual TL;DR, startuphub.ai AGI Quest hindered by Sample Efficiency Gap. Sample Efficiency Gap solved by World Models. World Models leads to Unlock AGI hindered by solved by leads to AGI Quest solving sample efficiency gap is fundamental for achieving Artificial General Intelligence Sample Efficiency Gap AI needs thousands of data points, humans learn with a handful of tries World Models core concept proposed by Y Combinator partners to bridge the efficiency gap Unlock AGI world models could be the breakthrough needed to achieve Artificial General Intelligence From startuphub.ai · The publishers behind this format Visual TL;DR, startuphub.ai AGI Quest hindered by Sample Efficiency Gap. Sample Efficiency Gap solved by World Models. World Models leads to Unlock AGI hindered by solved by leads to AGI Quest solving sample efficiency gap is fundamental for… Sample Efficiency Gap AI needs thousands of data points, humans learn with a… World Models core concept proposed by Y Combinator partners… Unlock AGI world models could be the breakthrough needed to achieve… From startuphub.ai · The publishers behind this format

World Models: The Key to AGI?

On July 17, 2026, in San Francisco, the AI Collective, a significant global community of AI developers, entrepreneurs, and enterprise experts, announced its role as an official partner of the AGI-26 Conference. This event is recognized as a leading forum for research, development, and practical application of artificial general intelligence (AGI).

The new partnership aims to bridge the gap between cutting-edge AGI research and the everyday tools and workflows utilized by businesses and developers. It combines the expertise of leading researchers at the AGI Conference with the vast network of over 250,000 members of The AI Collective, who are actively engaged in deploying AI systems today. 

AJ Green, Executive Director of The AI Collective, commented, “The journey to achieving AGI encompasses much more than just advanced models; it fundamentally includes the human element — the skills, insights, and craftsmanship of our community. After the success of Humans in AI Week, where events spanned over 100 cities, AGI-26 serves as an ideal platform for these important discussions. We are thrilled to collaborate with Ben Goertzel and the teams at the AGI Society and SingularityNET.”

Ben Goertzel, CEO of SingularityNET, affirmed the significance of this collaboration, stating, “We are witnessing a transformative moment where the quest for AGI is gaining momentum beyond niche research circles, drawing in support and excitement from a much wider community. This partnership has the exciting potential to expedite the transition towards AGI for everyone's benefit. The foundational innovations pursued by long-term AGI researchers can greatly enhance the practical initiatives of The AI Collective, while the diverse energy and scale of the Collective can broaden the scope of AGI research and development in ways previously unimaginable. It’s a mutually beneficial opportunity!”

Under this partnership, The AI Collective and AGI-26 will initiate several key collaborations:

- They will provide speaking opportunities for The AI Collective’s community, allowing practitioners and founders to present real-world case studies and technical progress on the conference stage.
- Workshops and dedicated sessions will be organized to focus on the practical adoption of AGI technologies, aimed at translating advanced research into usable tools.
- Opportunities for collaboration will be enhanced, connecting AGI researchers and academics with operators and founders from The AI Collective.
- Access for members of The AI Collective to participate in the extensive lineup of presentations, discussions, and networking events at AGI-26 will be expanded.

This partnership embodies a shared vision that achieving beneficial and widely embraced AGI relies on fostering collaboration between researchers and practitioners who are translating theoretical concepts into tangible products and societal impact.

Celebrating its 19th year, the AGI Conference series has stood as the sole major forum devoted to the development of human-level AI since its inception in 2008. AGI-26 is scheduled for July 27-30, 2026, in San Francisco, featuring both in-person and virtual attendance options. This year’s conference comes at a crucial time, as developments in theoretical AGI research increasingly yield systems that demonstrate true reasoning, adaptability, and broader generalization.

Scheduled speakers for AGI-26 include notable experts such as:

- Ben Goertzel — SingularityNET / ASI Alliance / BGI Labs
- Karl Friston — University College London
- Emad Mostaque — Intelligent Internet
- Neil Gershenfeld — MIT
- Michael Levin — Tufts University
- Marco Pavone — NVIDIA
- Alexander Lerchner — Google DeepMind
- Alison Gopnik — UC Berkeley
- Faezeh Habibi — SingularityNET
- Hod Lipson — Columbia University
- Alex Wissner-Gross — Physical Intelligence

These speakers, representing diverse philosophical approaches in the field, span neuroscience-inspired models, symbolic reasoning cases, biological intelligence research, and programmable matter development, showcasing a program rich with innovative ideas and debates.

The AGI Conference previously featured numerous prominent figures like Jürgen Schmidhuber, Yoshua Bengio, Peter Norvig, and Gary Marcus, solidifying its status as the nexus for prominent thinkers in the domain.

Since its founding, the AGI Conference Series has engaged over 1,000 researchers, practitioners, and thought leaders, functioning primarily as a rigorous examination space for ideas and assumptions that shape the future direction of the field.

Tickets are available at: https://luma.com/AGI-26

About AGI Society

The AGI Society is a nonprofit organization committed to advancing research and understanding of artificial general intelligence systems. Its mission is to cultivate global collaboration and communication to share knowledge and diverse perspectives about the future of intelligence.

For more information, visit: https://agi-conference.org/

Contact:

SingularityNet Team  
info@singularitynet.io

The AI Collective Designated as the Official Community Partner for the AGI-26 Conference

Interior of a sleek, modern conference hall flooded with warm natural light through floor-to-ceiling windows. A bold “AGI-26 Conference” banner stretches across the stage’s back wall. In front stands a podium draped with the AI Collective logo in vivid color, its design crisp and unmistakable. Well-dressed adult professionals gather in small groups, animatedly discussing the partnership, while elegant signage and subtle tech displays hint at groundbreaking AI breakthroughs on the horizon.

The AI Collective Named Official Community Partner of AGI-26 Conference - TradingView

- The AI Collective partners with AGI-26 Conference to bridge AGI research with practical applications.  
- Partnership enables speaking opportunities and workshops for AI builders at AGI-26 Conference.  
- AGI-26 will feature prominent speakers addressing various perspectives in artificial general intelligence.  

San Francisco, CA, July 17th, 2026, Chainwire

The AI Collective, one of the world's largest communities of AI builders, founders, and enterprise practitioners, today announced it has been named an official partner of the AGI-26 Conference, the premier gathering for artificial general intelligence research, development, and deployment.

The partnership brings together AGI Conference's deep bench of leading researchers and theorists with The AI Collective's network of thousands of engineers, founders, and enterprise leaders who are building and deploying AI systems in production today. Together, the two organizations aim to close the gap between frontier AGI research and the real-world tools and workflows that businesses and developers use every day.

"The path to AGI runs through people as much as models. Our community of 250,000+ builders, founders, and practitioners works on that human side every day — the judgment, skills, and craft it takes to build with increasingly capable systems. Fresh off Humans in AI Week, which convened simultaneous events across 100+ cities, we couldn't ask for a better stage for that conversation than AGI-26, and we're proud to partner with Ben Goertzel and the teams at the AGI Society and SingularityNET." — AJ Green, Executive Director, The AI Collective.

“In this unprecedented time, the push toward AGI is emerging from the niche research world that has fostered it since the middle of the last century, and attracting enthusiasm and resources from a much broader community, " continues Ben Goertzel, CEO of SingularityNET. “I’m excited about the partnership between the AGI conference and the AI Collective because of its potential to accelerate and smooth this transition for the benefit of everyone. The deep innovations pursued by long-term AGI researchers have tremendous potential to enhance the practical projects of the AI Collective; and the energy, scale and diversity of the AI Collective have corresponding potential to expand the scale of AGI R&D beyond what has been historically possible. A win-win for sure!”

As part of the partnership, The AI Collective and AGI-26 will:

Open speaking slots to The AI Collective's community, giving builders, founders, and practitioners direct access to the AGI Conference stage to share real-world case studies, technical breakthroughs, and lessons learned from deploying AI at scale.

Create dedicated programming and workshops focused on the practical, hands-on adoption of AGI technologies, helping translate cutting-edge research into tools teams can use now.

Facilitate direct connections between AGI Conference's research and academic network and The AI Collective's base of operators and founders, fostering collaboration between those advancing the science of AGI and those building its applications.

Expand access for The AI Collective's members to attend, participate in, and engage with AGI-26's full program of talks, panels, and networking events.

The partnership reflects a shared belief that the path to beneficial, widely-adopted AGI runs through collaboration between researchers and the practitioners translating that research into real products and impact.

Now in its 19th year, the AGI Conference has been the only major conference series dedicated to human-level AI development since its founding in 2008. AGI-26 will take place July 27–30, 2026, in San Francisco, in a hybrid format offering both in-person and virtual attendance. This year's edition arrives at a pivotal moment, as theoretical AGI research increasingly translates into systems capable of genuine reasoning, adaptation, and broader generalization.

Confirmed speakers for AGI-26 include:

Ben Goertzel — SingularityNET / ASI Alliance / BGI Labs

Karl Friston — University College London

Emad Mostaque — Intelligent Internet

Neil Gershenfeld — MIT

Michael Levin — Tufts University

Marco Pavone — NVIDIA

Alexander Lerchner — Google DeepMind

Alison Gopnik — UC Berkeley

Faezeh Habibi — SingularityNET

Hod Lipson — Columbia University

Alex Wissner-Gross — Physical Intelligence

Together, this year's speakers represent every major school of thought driving the field forward, from Friston's neuroscience-inspired models and Marcus's case for symbolic reasoning, to Levin and Hazan's work on biological intelligence, Gershenfeld's work on programmable matter at MIT, Lerchner's frontier research at Google DeepMind, and Goertzel's own decades-long pursuit of general intelligence. The result is a program that captures its most productive tensions, bringing neuroscience, symbolic reasoning, agent architectures, and hybrid systems onto a single stage.

AGI Conference's speaker series has previously featured foundational figures including Jürgen Schmidhuber, Yoshua Bengio, Peter Norvig, Gary Marcus, Christof Koch, Richard Sutton, Marcus Hutter, and François Chollet, cementing its reputation as the gathering point for the field's most serious thinkers.

Since its founding in 2008, the AGI Conference Series has welcomed over 1,000 researchers, practitioners, and thought leaders from academia, industry, and government. The series has functioned less as a showcase and more as a pressure test, a place where foundational assumptions get challenged and the direction of the field gets shaped.

Link to tickets: https://luma.com/AGI-26

About AGI Society

The AGI Society is a nonprofit organization dedicated to promoting the study and design of artificial general intelligence systems. The society facilitates global cooperation and communication to publicize knowledge and diverse views concerning the future of intelligence.

https://agi-conference.org/

Contact

SingularityNet Team

info@singularitynet.io

The AI Collective Named Official Community Partner of AGI-26 Conference

In recent discussions about the impact of artificial intelligence, it's becoming increasingly clear that AI data centers are just the beginning of broader trends of exploitation—particularly in rural landscapes. A prominent example can be found in Mount Carmel Township, Pennsylvania, where a yard sign opposing the construction of a new data center reflects growing community concerns. 

The critique of unchecked technological advancement is not new. Ivan Illich's 1970 work, "Toward a History of Needs," challenged the prevailing notion that technological progress is inherently beneficial. Similarly, Wendell Berry, in his 1977 book "The Unsettling of America," highlighted how such advancements tend to erode local skills and community resilience.

Currently, these warnings resonate strongly, particularly as AI fosters a significant decline in skills across society. This erosion could centralize economic and political power to a degree that dwarfs present disparities, leading to a staggering loss of jobs. OpenAI defines AI as “highly autonomous systems that outperform humans at most economically valuable work,” reinforcing fears that this shift is systemic rather than incidental.

As noted by my colleague Erica Etelson, the rapid proliferation of AI data centers has sparked widespread community backlash. Residents from various political backgrounds are voicing their opposition to the noise, land use, and exorbitant energy and water consumption these facilities demand. Promises of data centers that are fully powered by renewable energy or equipped with water-saving cooling systems are emerging, alongside more ambitious ideas like Elon Musk’s proposal to relocate data centers to space, attempting to sidestep local resource conflicts altogether.

However, the viability of these so-called “green” initiatives remains questionable. Historical patterns suggest three key developments unless proactive measures are taken. 

Firstly, the race for AI dominance is concentrating economic and political power like never before. By 2025, the combined wealth of the top ten leaders in tech and AI is projected to reach $2.5 trillion, exceeding 8 percent of the total U.S. GDP. This consolidation influences political landscapes, with AI companies and their affiliates anticipated to spend upwards of $150 million on the 2026 midterm elections alone. Policies such as the recently introduced bipartisan "Great American Artificial Intelligence Act of 2026" exemplify the political clout wielded by this elite group, as it seeks to diminish state oversight and evade pressing regulatory challenges, as analyzed by Public Citizen.

Beyond mere oligopoly within the AI sector, these wealthy figures—Musk, Altman, Larry Ellison, and Jeff Bezos—are working to control governments and shape policy in unprecedented ways, buoyed by their vast fortunes and by creating technologies that integrate seamlessly into everyday life.

Secondly, we are on the brink of a rapid and extensive job loss across various sectors, potentially erasing entire categories of work. As outlined in a May 2025 article from Axios, the tech industry is engaged in a fierce competition to develop "Agents," or agentic AI, which can replicate human labor quickly, consistently, and at a fraction of the cost. From coding and finance to customer support and content creation, the applications appear limitless, with many of these systems already operational in companies and even more under development.

Goldman Sachs estimated in 2023 that AI advancements could threaten up to 300 million jobs globally, impacting both knowledge-based professions and vast segments of the service and manual labor sectors. Roles traditionally considered stable, such as those in human resources, billing, and customer service, are already precarious.

Proponents of AI often argue that job losses will be offset by new types of employment opportunities. Yet, the primary job category that has emerged thus far is that of "data annotators." This group, mostly comprising workers in countries like Venezuela, Kenya, and the Philippines, is tasked with the challenging job of reviewing the content generated by AI systems—sifting through an overwhelming amount of inappropriate material that arises from the broad data collection necessary for training these models. Much like commercial fishing, where vast amounts are captured and sorted, AI companies exploit the full breadth of human experience, compensating these workers with minimal pay while they perform this essential yet undervalued labor.

AI Data Centers Are Merely the Beginning of a Larger Trend

A dark, moonlit seascape where a sleek, container‐style AI data center perches atop a floating iceberg, its servers glowing faintly. Beneath the waterline the iceberg stretches into a vast, labyrinthine underworld of tangled fiber-optic cables and towering racks, hinting at massive unseen infrastructure. Natural moonlight shimmers on the water’s surface, casting dramatic shadows around the hidden depths.

AI Data Centers Are Just the Tip of the Iceberg - The Nation

- AI data centers are provoking community backlash due to environmental and resource concerns.  
- Concentration of economic power from AI elites threatens democracy and public policy regulation.  
- Rapid job losses are expected as AI systems increasingly replace human labor across industries.  

Rethinking Rural / AI Data Centers Are Just the Tip of the Iceberg These destructive behemoths are part of a broader model of total extraction—particularly in rural areas.

A yard sign opposing a planned data center is displayed along Route 54 in Mount Carmel Township, Pennsylvania. (Paul Weaver / SOPA Images / LightRocket via Getty Images)

In 1970, Ivan Illich wrote Toward a History of Needs, a critique of the increasingly widespread belief that technological change was not only inevitable but nearly always a net good for society. Illich feared our growing dependence on rapidly emerging technologies and those that control them. Farmer and writer Wendell Berry made a similar argument in his 1977 book, The Unsettling of America, describing the process by which people and whole communities are “de-skilled,” their capacity to care for themselves and their neighbors atrophied.

In many ways, these dark visions are now unfolding. In particular, artificial intelligence is driving us toward de-skilling on a massive scale. This de-skilling, if it continues, will enable concentrated economic and political power at a level that will make the present seem egalitarian by comparison. And it will include the loss of hundreds of millions of jobs along the way, not as a byproduct but a feature, as evidenced by Open AI’s definition of AI as “highly autonomous systems that outperform humans at most economically valuable work.”

As described by my colleague Erica Etelson, the explosive growth of AI data centers has catalyzed a backlash in communities across the country. Residents are uniting across party lines in opposition to the noise, land use, and extraordinary levels of energy and water that data centers require. In response, promises of data centers fully powered by renewable energy or utilizing water-conserving cooling systems have emerged. Elon Musk promises to launch data centers into space, avoiding local land, water, and energy conflicts altogether.

It’s doubtful that such “green” data centers are even feasible. But based on the history of AI development, three outcomes are all but assured unless we act swiftly and decisively.

First, the AI arms race is propelling an unprecedented concentration of economic and political power. By the end of 2025, the top 10 CEOs and founders of AI and tech companies were worth $2.5 trillion, more than 8 percent of the total US GDP. This was before Elon Musk became a trillionaire.

The political influence of this tiny elite is large and growing, with a New York Times projection of at least $150 million being spent on the 2026 midterm races by AI companies and their allies. Alongside this spending are ongoing political influence campaigns ranging from the bullying threats of Musk to the charm offensive of Open AI CEO Sam Altman. A recently introduced bipartisan bill, dubbed the Great American Artificial Intelligence Act of 2026, is another example of the political power of the AI elite. It usurps the power of states to regulate AI and kicks the can down the road on the majority of the most critical concerns about AI, according to an analysis by Public Citizen.

The power of the biggest AI companies will encompass far more than oligarchy within this one industry, given AI’s steady insinuation into nearly every facet of our lives. The ultrarich owners of these corporations—Musk, Altman, Larry Ellison, Jeff Bezos and a handful of others—are working toward control of governments, public policy, and culture to a degree unseen in our history, not only because of their wealth but also because they are making the technology they control indispensable and ubiquitous.

Second, we are approaching a period of massive and rapid loss of jobs and near-elimination of whole categories of occupations. According to a May, 2025 piece in Axios, “technology companies are in a wild race to produce so-called Agents, or agentic AI. In its simplest form, an agent is AI that can do the work of humans—instantly, indefinitely and exponentially cheaper. Imagine an agent writing the code to power your technology, or handle finance frameworks and analysis, or customer support, or marketing, or copy editing or content distribution, or research. The possibilities are endless—and not remotely fantastical. Many of these agents are already operating inside companies, and many more are in fast production.”

A 2023 estimate by Goldman Sachs projected a potential loss of 300 million jobs worldwide, not only for workers in the so–called knowledge economy but also for wide swaths of service and blue-collar jobs. Mid-level “back office” jobs in HR, billing and payroll, and customer service are already very vulnerable.

Many AI industry supporters dismiss the potential for widespread job loss, claiming instead that the types of jobs available will simply shift to accommodate AI. How’s that going so far? Probably the biggest category of new jobs created by AI to date is “data annotators.” These are the thousands of people, primarily in Venezuela, Kenya, the Philippines, and other developing countries, who do the dirty work of AI. They review both texts and videos produced by Large Language Models, culling the seemingly endless stream of violent, offensive, racist, misogynistic and deviant content that is unavoidably swept up in the process of capturing data from every corner of the Internet. Like the giant nets of huge commercial fishing vessels, AI companies extract everything they can from human experience, good, bad and ugly. And then they pay workers a meager piece rate to find and delete the worst of it.

Google DeepMind expresses concerns over the implications of millions of agents interacting.

More from Artificial General Intelligence

California Institute for Human Science Introduces Its First Accredited Graduate Program Focused on AGI

Reasons AI Might Never Achieve Human-Level Intelligence

White House Considers Regulator to Oversee AI Models

Analysis: Has AI been pursuing the wrong objective since Alan Turing?