Meta’s face recognition AI may yet metastasize

No man is an island, said John Donne, a metaphysical poet whose words survived the industrial age even if they didn’t all that age well, gender wise. Now four decades into today’s information age, Meta Inc would have us all web-linked as our virtual avatars in a so-called metaverse, a big-bang creation of software that could expand forever and engage all five senses in its quest to blur our cognitive calls on reality. Meta is the name of Facebook Inc’s new parent firm, as unveiled on 28 October by its chief Mark Zuckerberg, and it signals a daring pivot in its business as it envisions itself as the host of a life-like online economy, complete with its own digital currency. To enrapture us by the billion, however, this concept would have to race the restraints of regulation that may close in. Amid public pressure on it to respect our privacy, last week Facebook said it would drop its face-recognition system: “People who’ve opted in will no longer be automatically recognized in photos and videos and we will delete more than a billion people’s individual facial recognition templates.”

The tool in question was adopted years ago by the social network to let users ‘tag’ friends easily, with prompts identified by its DeepFace artificial intelligence (AI) that would scan uploaded images to work out who’s who and enrich its facial database. It was explicitly made an opt-in feature only in 2019, had a run-in with a local biometric privacy law in the US last year, and “growing societal concerns” have since been found to outweigh its utility. Indeed, AI-based face scanners can be put to various ends, from unlocking phones and doors to nabbing absconders and dissenters under mass surveillance, as in China. Meanwhile, the fallibility of such tools has given ‘AI’ the ring of an oxymoron. Their reliability has been found to vary by skin tone, for example. This could reflect either the science of light reflection or the folly of human biases; either way, the failure so far of this technology on diversity argues for it not to be thrust upon us. On privacy, the policy rankle has been this: While law enforcers and spy agencies can claim a public purpose for a global database of profiles tagged with face IDs, what justifies bio-invasive data being deployed by a social-media monopoly for private profit? As a default setting of law, the right to retain or sign away personal data ought to be ours.

Despite its deletion of facial templates due this year-end, the issue could return to haunt Meta as it bets its future on the ‘network effects’ of viral enrolment for its metaverse, the mass appeal of whose matrix might help it outpace regulators. Nation-states have been tardy in their response to an internet pivot from the dispersal of data to its concentration (think “new oil”) and what this implies for private power. Of specific concern is that a metaverse full of avatars as economic agents will likely need bio-verifiable faces to function best. What little we know of Meta’s own project—there are others—augurs an augmented role for DeepFace-like AI in the creation of digital body clones that will not just capture and relay our eye movements, facial expressions, gestures, etc, but perhaps also store all this stuff to monetize. If so, people who sign up for avatarhood must not just be armed with legal rights to their personal data, but duly alerted to the implications of being watched. As it is, Big Tech knows too much about too many of us.