The future of Big AI may lie with product-development teams.
What’s new: Meta reorganized its AI division. Henceforth, AI teams will report to departments that develop key products.
How it works: Prior to the reshuffle, the company’s Responsible AI, AI for Products, AI4AR (that is, for augmented reality), and Facebook AI Research teams were managed by a single division called Meta AI. This structure made it difficult to translate machine learning into marketable applications, according to chief technology officer Andrew Bosworth.
- Responsible AI, which aims to mitigate bias in the company’s models, will report to Social Impact, which develops tools to help nonprofits use Meta’s social media platform.
- AI for Product, which develops applications for advertising and recommendation, will join the product engineering team.
- AI4AR, which develops augmented- and virtual-reality tools like Builder Bot, will join Meta’s Reality Labs as part of the XR (an acronym for extended reality) team, which oversees technologies like Spark AR and Oculus headsets.
- Facebook AI Research, led by Antoine Borges, Joelle Pineau, and Yann LeCun, will also report to Reality Labs. In addition, Pineau will lead a new team that assesses company-wide progress on AI.
- Jerome Pesenti, Facebook and Meta’s vice president of AI since 2018, will depart the company in mid-June.
Shaky platform: AI teams who work for Meta’s flagship Facebook social platform have had a rocky few years.
- Last year, a former product manager leaked documents to the press showing that the company knowingly tweaked its recommendation algorithm in ways that harmed both individuals and society at large.
- In 2020, reports surfaced that company leadership had blocked internal efforts to reduce the amount of extreme content the algorithm promoted over concerns that doing so would drive down profits.
- In 2018, Joaquin Quiñonero Candela, an architect of Facebook’s recommendation algorithm, took charge of the Responsible AI division to mitigate the algorithm’s propensity to promote disinformation, hate speech, and other divisive content. (Candela departed in 2021.)
Trend in the making? Meta isn’t the first large company to move AI teams closer to its product groups.
- Last year, Microsoft moved its data and AI units under the umbrella of its Industries and Business Applications group. In 2018, Microsoft had integrated AI research more closely with its cloud computing business.
- In 2018, Google absorbed its DeepMind division’s healthcare unit with the goal of translating applications, such as the Streams app that alerts caregivers to concerning test results, into clinical practice.
Why it matters: In 2019, 37 percent of large AI companies maintained a central AI group, The Wall Street Journal reported. Reorgs by Meta and others suggest that centralization hindered their ability to capitalize on AI innovations.
We’re thinking: In a corporate setting, when a technology is new, a centralized team can make it easier to share learnings, set standards, and build company-wide platforms. As it matures, individual business units often gain the ability to manage the technology themselves and absorb experienced developers. Apparently this pattern — which we describe in AI For Everyone — is playing out in some leading AI companies.