Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
The maths behind artificial intelligence is complicated but, when it comes to creating “personal superintelligence”, Mark Zuckerberg appears to be operating on the basis of a simple equation: if hiring one star AI researcher is good for his clandestine project, it will be even better if he recruits dozens.
The founder of Meta is spending tens of billions of dollars on new facilities, equipment and talent. The group has hired some 50 scientists so far for its superintelligence lab, reportedly dangling nine-figure sign-on bonuses in front of the best AI minds.
Money does talk. On Friday, Meta announced the hiring of Shengjia Zhao, co-creator of arch-rival OpenAI’s ChatGPT, as the lab’s chief AI scientist. He will report to another recent high-profile hire, maths prodigy Alexandr Wang. But Zuckerberg has also been rebuffed by many of his targets. Scientists and engineers are motivated by multiple considerations. Some members of the AI elite have preferred to stay put at Meta’s rivals, driven by their ethical positions on the use of the technology, loyalty to teammates or, perhaps, the greater rewards they think will come from keeping an equity stake in a successful smaller company.
But Zuckerberg’s mission is risky for other reasons, because it can be dangerous to stack a team with stars and expect them automatically to make magic together. Grumbles can already be heard from Meta’s headquarters, where existing AI staff worry about being sidelined by the expensive new superstars. OpenAI co-founder Sam Altman has obliquely accused Meta of recruiting “mercenaries”. Studies of the movie business, equity analysts and sports teams, such as Real Madrid’s galácticos in the 2000s, suggest teams can succumb to a “too-much-talent” effect. In many cases, stars work better when supported by a network of highly competent teammates.
One study by Harvard Business School’s Boris Groysberg and others found that once the proportion of high-status equity analysts in a team rose above a certain level, their research ratings started to fade. Stars also often find it hard to transfer their skills to a new company or culture. One head of research at an investment bank told Groysberg that hiring a star analyst “resembles an organ transplant”, with the same risk of painful rejection.
Research published last year in Management Science, by Denisa Mindruta of HEC Paris and others, looked specifically at science and found that collaboration between top scientists and a “constellation” of less starry researchers usually yielded greater innovation than individual efforts. In addition, if the expertise of stars and their teams overlapped too much, their collaboration was less successful.
The superintelligence project could be the exception that proves the rule that it is perilous to pack your team with hotshots. With all due respect to stock analysts, the “creation and discovery of new things that aren’t imaginable today”, to quote Zuckerberg, is not the same as deciding whether to tip BP or Shell. In setting up a secret development unit detached from head office, the Meta chief is also applying a template that worked for the atomic Manhattan Project and Lockheed Martin’s “Skunk Works” jet-fighter design programme.
One other large hurdle looms for Meta. From football coaches to chief executives, leadership matters in ensuring high-octane teams are productive and avoid damaging internal clashes. Presumably if Meta’s researchers make the breakthrough to personal superintelligence they can simply hand the reins to the bots. Until then, Zuckerberg would be well advised to spend some of his budget hiring more star managers to keep his galácticos on course.
Leave a Reply