A senior editorial director at Vox is responsible for leading the Future Perfect and climate teams, as well as crafting the Good News newsletter. This individual previously dedicated 15 years to Time magazine, where they served as a foreign correspondent in Asia and later as an international editor and climate writer. They have also authored a book focusing on existential risks.
But how can we determine when a significant transformation has occurred in the world?
On June 1, a research group unveiled a preprint paper claiming a groundbreaking advancement in human embryonic DNA editing, surpassing previous attempts in precision. This achievement stands out as it largely sidesteps the errors seen in earlier gene modification endeavors. With continued advancements, this form of embryonic editing may eventually eliminate devastating genetic disorders from future generations. However, as renowned science journalist Carl Zimmer noted in the New York Times that same week, the major takeaway is that this work could pave the way for the creation of “designer babies,” adding a new dimension to ethical discussions.
That same day, the AI firm Anthropic released a post arguing that artificial intelligence is already accelerating its own development, suggesting we may be on the brink of recursive self-improvement (RSI). This refers to AI systems that could design and construct their own successors more rapidly. Notably, much of the code driving Anthropic's Claude was produced by Claude itself, allowing the company to increase its output eightfold compared to two years prior. While this doesn’t automatically equate to better outcomes, and Claude is not yet self-sufficient, the emergence of self-improving AI could arrive sooner than institutions are ready to handle, as emphasized by Anthropic co-founder Jack Clark and head of the Anthropic Institute, Marina Favaro.
These two publications, though emanating from disparate fields, signify a potential future that could look very different from today’s landscape.
Both developments are monumental steps toward introducing unprecedented capabilities — each carrying its own risks and ethics. The genetic precision that holds the promise of curing hereditary diseases like Huntington's disease also threatens to lead to societal structures based on genetic divisions. Similarly, the advancements in AI that could propel scientific research forward may also lead to our disempowerment as creators.
The recent advancements could mark a crucial juncture in history, though it remains uncertain what kind of future they will usher in.
Let’s first consider the biological advancement. Beyond the sensational headlines crafted by the media, the underlying research is quite focused.
Utilizing base editors, which introduce minor modifications to genetic strands rather than deleting entire sections like the CRISPR method, Columbia University geneticist Dieter Egli and his team performed edits on two specific genes: PCSK9 and HBG. The first gene, PCSK9, produces proteins that help regulate cholesterol levels; certain mutations can dangerously elevate LDL cholesterol. The second gene, HBG, is responsible for producing a type of hemoglobin crucial during fetal development but inactive afterward. Controlling these genes could help mitigate risks associated with heart disease (via PCSK9) and reactivate fetal hemoglobin in adults, offering potential relief for conditions like sickle cell disease and beta-thalassemia (via HBG).
The research team successfully delivered these base editors into fertilized eggs and two-cell human embryos, achieving edits with fewer chromosomal errors compared to past attempts using CRISPR.
While the paper is still awaiting peer review, it represents a significant step in the quest for precise gene editing in human embryos. Nevertheless, 'impressive' does not imply 'perfect' or even 'safe' — some modifications improperly targeted the genome, and the number of embryos that developed normally was relatively low (the embryos were donated from IVF patients and were only allowed to develop to early stages without being implanted). The authors cautioned that the notion of using this technique for clinical purposes remains “premature.” However, it demonstrates that such editing is achievable without causing widespread chromosomal damage.
When Chinese scientist He Jiankui used standard CRISPR methods in 2018 to edit human embryos and produce three children, his work faced widespread condemnation for both ethical and technical failures, as it caused significant genetic harm. Should the latest findings be validated, the technical barriers to embryo modification could diminish.
What lies ahead is uncertain. While certain genetic conditions, such as sickle-cell anemia, can be rectified with a single edit, preventing more intricate health issues or engineering desirable traits like height or intelligence would necessitate countless edits across numerous genes — combinations that we do not yet fully comprehend. If the technical challenges continue to decrease, we will have to confront ethical implications, which have historically not impeded technological progress.
Turning to the AI development aspect, the rapid progress in this field contrasts with the slower evolution of biological innovations.
Anthropic's detailed post highlighted a singular trend: the share of human effort dedicated to AI development is diminishing at every juncture. Developers who once wrote code now predominantly review what Claude generates. Experimentation that was traditionally crafted by hand is now increasingly initiated and executed by the AI model itself. Although human oversight remains crucial, the authors assert that even decision-making processes are beginning to shift toward the model’s proposals.
This emerging research ecosystem, increasingly dominated by AI, has the potential for exponential acceleration. Technological advancements have previously been paced by human thought and action; an AI capable of self-enhancement would obliterate these constraints, potentially evolving at a rate that outpaces human governance and oversight. Consequently, intelligence could reach a critical inflection point — each iteration of AI generating an even more sophisticated successor in a compounding cycle.
While this assessment stems from a few months of coding data from a tech company keen to present its models favorably, independent studies corroborate that AI models are spending longer on specific tasks, leading to more profound understanding and efficiency. Thus, while we can debate the pace of these changes, the consensus is that AI is indeed advancing rapidly.
Robust and swift AI capabilities could fuel significant strides in economic, scientific, and medical realms — embodying the aspirations outlined by Anthropic CEO Dario Amodei.
However, this rapid progress also carries existential risks, mirroring the societal consequences posed by genetic enhancements for those who might be marginalized. The swift rate of change is so alarming that Anthropic has even suggested that AI developers consider a collective slowdown or temporary halt in the advancement of cutting-edge AI, enabling society and research on AI alignment to catch up. The authors referenced historical frameworks designed to manage past dangerous technologies, such as nuclear arms, which, despite their flaws, have helped avoid catastrophic outcomes. Yet, the institutions capable of managing such technology took decades to establish. As noted by the leaders at Anthropic, with self-enhancing AI, “We don’t have that long.”
How do we know when a pivotal shift has transpired in the world?
At times, it’s instantly recognizable. When Otto Hahn and Fritz Strassmann achieved nuclear fission in December 1938, experts were quick to grasp its implications: the potential for nuclear weapons became clear. Other instances see scientists recognizing the significance while the public remains oblivious. Conversely, the critical cases often challenge even experts to fully grasp the ramifications. Developments such as fission hinted at destructive potential, with those knowledgeable enough to understand it powerless to avert disaster.
Last week’s advances are multifaceted, each possessing both beneficial and perilous implications. The same technology offering a remedy for a terminal illness could facilitate the emergence of societal genetic hierarchies. Concurrently, AI promising intellectual achievements may also lead to human obsolescence.
Ultimately, we find ourselves at a crossroads shrouded in uncertainty. The concern lies not only in having ventured through the wrong portal but in having crossed it without realizing no choice was made.
