The hero of the dystopian thriller “Soylent Green” learns at the end that the mysterious food that keeps New Yorkers alive—and the name of the film—is actually made of human beings. Another wrinkle in the story is the rumor that Meta Platforms may pay $15 billion to purchase 49% of the artificial intelligence startup Scale AI. AI is a human, too.
Thus far, creating artificial intelligence has primarily involved managing limited inanimate resources. There is a shortage of high-end chips, for instance, and power limitations are a factor. This year, Meta is expected to spend up to $72 billion on capital expenditures, the majority of which has been allocated to servers and data centers, the technological counterpart of physical buildings.
However, other elements are also getting scarce. One is good data, which is required for effective AI model training. Ilya Sutskever, co-founder of OpenAI, has cautioned that we are approaching “peak data” in the same manner that we are approaching “peak oil”.
Scale AI is a firm that aims to postpone the point when data runs out. It names, cleans, categorizes, and corrects so-called “tokens” and the models that rely on them. Founder Alexandr Wang claims that openly available data has reached its limit, and what follows next will be more difficult to transfer into a format that these algorithms can exploit.
Mark Zuckerberg, the CEO of Meta, is also concerned about this. Privacy is one of the issues the company is addressing: European users, for example, can deny the corporation the permission to utilize their public Instagram photos to train models, although their US counterparts cannot. Reddit, an online forum, is suing AI model provider Anthropic for “scraping” user messages without authorization.
Meta and its counterparts are aware that more individuals are part of the answer. Scale AI advocates for “reinforcement learning from human feedback,” or RLHF, in which actual personnel course-correct and coach models. Wang believes algorithms perform best when accompanied by a person, and will continue to do so in the foreseeable future — a hybrid method called as “centaur AI”.
That could explain why Meta would contemplate not only acquiring half of Scale AI, but also hiring its founder to lead a new “superintelligence” branch. The label would be appropriate for both the product and the process. To elevate AI to commanding heights, it must be paired with not just humans, but really brilliant ones — leading scientists and great thinkers — who can teach it not only what brainiacs know but also how they reason.
Granted, $15 billion is a lot of money to entice some intelligent individuals. It was reported that Scale AI was expected to generate $2 billion in sales this year. According to reports, Meta’s bid would put the entire company at 15 times its current worth.
However, this is little compared to the $44 billion in dividends and share buybacks projected by LSEG this year. If Zuckerberg is the first to achieve AI supremacy, opening a market worth trillions of dollars, $15 billion will seem like a trifle. AI was already a bidding war for processors, power, and data; now human neurons have joined the mix.
