Time Plus News

Breaking News, Latest News, World News, Headlines and Videos

Google and TikTok give Meta an AI lesson

Facebook (or Meta, as it prefers to call itself these days) last week said it was building what will soon be the world’s biggest supercomputer for handling artificial intelligence.

For anyone wary of so much computing power in the hands of a company that has admitted to devastating slips in the past, it felt about as ominous as watching a Bond super-villain unveil their latest invention.

Not long ago, an exaflop of computing power (about the equivalent of 10m laptops all firing away at once on the same problem) was still a dream among supercomputer aficionados. Now Meta says its machine will have five times that horsepower by the middle of this year.

Rather than planet-threatening, Meta insists its technology will be world-building. Specifically, it will conjure up the digitally generated 3D habitats that will one day make up the metaverse.

This week, however, the social media giant has been humbled by an altogether more mundane AI. TikTok’s core engagement algorithm, which works to maximise the time people spend watching its catchy short videos, has been touted as the most effective consumer use of AI. It now appears that TikTok’s technology is starting to bite. Mark Zuckerberg singled out his Chinese rival’s headlong growth while explaining an ominous erosion in his own company’s prospects, wiping nearly a quarter of a trillion dollars from its stock market value.

Zuckerberg may dream of distant metaverses, but TikTok’s AI is transforming the here-and-now. Its algorithm looks for signals that show how popular each video is, then tests how effective these signals are in increasing the number of times users return to the app, or in getting them to spend more time there. Armed with enough data and applied to mass market digital services, AI like this is already having a sweeping impact.

A day before Facebook’s ominous warning about the threat from TikTok, Google provided one of the most powerful demonstrations of this technology in action. Its advertising revenue surged ahead unexpectedly in the latest quarter — in good part a result, the company claimed, of applied AI.

As with TikTok, it is about enhancing a service that is already highly popular with a particular set of customers to make it even more valuable. Google’s AI analyses how ad campaigns are working and helps customers devise the most effective bidding strategies to use in its real-time auctions. It also looks through all the different text, images and video that advertisers have at their disposal, and recommends how to assemble this into the most effective message for each audience.

As with TikTok, what makes this powerful is the sheer volume of data the algorithms can be trained on. More than $250bn of advertising flowed through Google’s systems last year — a huge part of the entire global ad industry’s estimated $700bn or so of revenue for 2021.

Companies that apply techniques like these hand in hand with the latest algorithmic advances should be in the best position — though the highly opaque nature of the technology makes it hard to tell, and Google’s critics claim its recent success owes more to predatory business practices than technology.

The steady drip of announcements about the latest AI breakthroughs suggests that, if anything, the pace is picking up. They include Pathways, a Google system that combines several different AI techniques to come up with a more flexible way of analysing problems. Meanwhile, sister company DeepMind showed off what may be the best-performing language-generation system since the field was revolutionised by OpenAI’s GPT-3. And last month, Google published a paper claiming that its own language system, LaMDA, had made big strides in tackling pervasive inaccuracies — the glaring weakness for all such systems — by tapping into authoritative sources to check its facts.

Advances like these would once have been greeted by breathless media coverage. Now, they come with almost routine frequency — itself a sign of how rapidly the field is advancing.

AI breakthroughs may one day bring the artificially generated worlds that Meta promises. But for the foreseeable future their main impact is more likely to come from their ability to supercharge digital services, operating at vast scale, that are already devouring the world.

richard.waters@ft.com

Source link