This afternoon has been eventful for Pip, the AI assistant I've been exploring within Meta's Muse application. Since its debut on September 8, this agent has efficiently scheduled a Friday night out for me and devised a packing timeline for my imminent relocation. It even reached out to a friend via email regarding our trip planned for next month and compiled a document featuring potential destinations we want to check out.
At the top of my screen, a charming Pixar-inspired avatar tapped away on a laptop, reflecting the ongoing work of my chosen AI companion.
The introduction of chatbots like ChatGPT has significantly transformed how users access information and obtain answers. Now, major tech players are striving to create advanced AI assistants capable of managing multifaceted tasks, enabling users to issue broad directives— such as “order my groceries for the week”—while they go about their daily routines.
Muse has quickly become the most downloaded free app in the iPhone App Store, surpassing 2.5 million downloads since its launch, as reported by market analytics firm Sensor Tower. This achievement signals Meta's latest ambitious effort to compete with established chatbots like OpenAI’s ChatGPT and Google’s Gemini.
Mark Zuckerberg, the CEO of Meta, is anticipated to unveil further insights into his vision for Muse and the company's broader AI strategy during the upcoming Connect conference on Wednesday.
Nevertheless, a crucial consideration remains: Will users feel comfortable sharing personal data with Meta? The app’s surge in popularity arrives in the wake of privacy concerns surrounding their smart glasses project.
According to analysts from Bank of America, while general feedback for Muse has been encouraging, privacy and trust are essential factors that will influence its widespread adoption.
The advantages of using Muse
Unlike ChatGPT or Gemini, which can feel somewhat open-ended, Muse provides users with clearer guidance through its structure. The app features five tabs: a primary chat area for interacting with the agent, a feed that can be tailored to reflect user interests, an Ideas section with example prompts, a goal-tracking list based on previous conversations, and a compartment for files and media.
For my impending move to a new apartment, I shared my packing needs with Muse, specifying how much time I had until the moving day and my preference for packing in short sessions.
The AI agent promptly generated a daily packing checklist for the upcoming week and recommended time allocations for each day, also advising me on setting aside a designated bag for immediate essentials for the first night in my new place.
Much like other AI assistants, Muse excels at executing planning-related tasks that often involve a mix of research, communication with family or friends, and reservation-making.
Upon linking my Google apps to Muse, I tasked it with emailing a friend to discuss our travel plans for a trip to Cape May, New Jersey next month. After her response, Muse efficiently organized her restaurant suggestions into a refined Google Doc, along with alternative choices.
One notable aspect of Muse is its ability to identify and communicate any obstacles encountered during task execution. For example, when I attempted to buy a specific shampoo, the agent investigated the issue and discovered that the online store required a professional license number for purchase.
This capability is facilitated by Muse's operation via a virtual cloud-based system, allowing it to browse websites to fulfill its assignments. Users can observe the process and intervene whenever necessary.
However, other AI solutions like ChatGPT and Gemini can similarly manage these tasks and occasionally deliver more detailed insights.
Limitations of Muse
Despite Muse's capabilities, it also illustrated that AI isn't yet equipped to handle every task seamlessly. For instance, when I requested a list of date night venues, it suggested two places in my vicinity that had been out of business for years.
In another instance, Muse seemed to overstate its abilities. It claimed in its prompt ideas that it could negotiate prices on Facebook Marketplace, stating that it would directly message sellers to haggle on my behalf when the right opportunity arose. However, when I attempted this, I discovered Muse could only draft messages for me to send, rather than negotiating directly.
A Meta representative confirmed that the prompt idea's description is indeed accurate, and that Muse can negotiate within the parameters set by users.
Shopping tasks often prove easier to tackle manually due to restrictions on bot activity. For example, Amazon does not allow Muse to navigate its online marketplace, although it can still obtain product links through searches. Similarly, while Muse successfully browsed the Target website, I was required to input personal information manually, as the retailer obstructed automated checkout processes. (Muse utilizes Stripe’s Link service for transactions, creating a one-time-use card to safeguard credit card information).
Consumer sentiment toward AI agents
As Meta and other technology companies place significant bets on the future of AI agents, consumers exhibit a growing skepticism toward this technology. A Pew Research Center report from June indicated that about two-thirds of American adults believe AI is progressing too rapidly, and more people foresee a negative societal impact from AI rather than a positive one.
Concerns about privacy remain paramount, particularly regarding users' willingness to let Meta manage payment details and app access, especially given the fallout from the Cambridge Analytica scandal. Notably, a Gallup survey published in July revealed that 41% of Americans lack confidence in large tech firms.
Miranda Bogen, director of the Center for Democracy & Technology’s AI Governance Lab, highlighted that while AI agents are marketed as helpful personal assistants, their functionality often hinges on extensive private data, necessitating stringent safeguards to protect user information.
Meta asserts that Muse has been designed to prioritize safety and privacy, running on a dedicated cloud server inaccessible to other agents. The company also emphasizes that Muse cannot view users' passwords or payment details.
While AI companies profess that AI agents excel at enhancing online shopping experiences, a study by Gartner found that 64% of consumers believe such technology benefits corporations more than the end users.
Despite Muse’s early success in the App Store, research from eMarketer indicates that ChatGPT and Gemini are the most commonly used AI assistants in the U.S. Respondents reported their preferences, with 33.7% opting for ChatGPT, 18.2% for Gemini, and only 4.7% for Meta AI.
Nevertheless, Muse presents Meta's best chance yet to emerge as a significant contender in the competitive AI landscape, at least for the time being.
