Monday, December 23, 2024

Sure, Google’s AI overviews could be useful – if you like eating rocks | John Naughton

Must read

Once upon a time, Google was great. For those who were online in 1998, history’s timeline bifurcated into two eras: BG (Before Google), and AG. It was elegant and clean: elegant because it was driven by a semi-objective algorithm called PageRank, which ranked websites according to how many other websites linked to them; and clean because it had no advertising, which of course also meant that it had no business model and accordingly was burning its way through its investors’ money.

It was too good to last, and of course it didn’t. Two of its biggest investors showed up one day, demanding a return on their investments. The company’s co-founders had an idea. One of the reasons theirs was such a good search engine was that they intensively monitored what people searched for, and then used that information continually to improve the engine’s performance. Their big idea was that the information thus derived had a commercial value; it indicated what people were interested in and might therefore be of value to advertisers who wanted to sell them stuff. Thus was born what Shoshana Zuboff christened “surveillance capitalism”, the dominant money machine of the networked world.

As the inventor of this machine, Google has prospered mightily from it ever since. For as long as most of us can remember, it has been the world’s favourite search engine, deriving colossal revenues from that dominance. In the process, though, it also created a new global economy – the link economy – consisting of the countless millions of organisations and businesses that derive revenues and attention from traffic generated by searches on Google.

Its dominance, though, also triggered a perpetual arms race between so-called “search-engine optimisers”, seeking ways of gaming the PageRank algorithm, and Google, tweaking it to defeat these ingenious tricksters. Every so often, a change in the algorithm would send seismic ripples through the link economy, generating heartbreaking stories of people who had built profitable business out of niches only to wake up one morning to find that their websites no longer attracted any visits. These ripples provided vivid evidence of Google’s central role in the ecosystem.

Which brings us to the here and now. The launch of generative AIs such as ChatGPT clearly took Google by surprise, which is odd given that the company had for years been working on the technology, and indeed was instrumental in two significant breakthroughs: the “transformer” model and the deployment of neural networks generally. Whatever the explanation, it was suddenly clear to the company that generative AI posed a strategic threat to its great cash cow – especially since, for a decade or so, its comfortable monopoly had led to a complacency that was manifested in the deterioration of its service as it prioritised advertising over objectivity in search results. For those of us who watch the industry, therefore, the question became: how will Google respond to the threat?

Now we know: it’s something called AI overviews, in which an increasing number of search queries are initially answered by AI-generated responses. “Sometimes,” the company burbles, “you want a quick answer, but you don’t have time to piece together all the information you need. Search will do the work for you with AI overviews.” Or, to put it more succinctly: “Let Google do the searching for you.”

To date, some of this searching suggests subhuman capabilities, or perhaps just human-level gullibility. At any rate, users have been told that glue is useful for ensuring that cheese sticks to pizza, that they could stare at the sun for for up to 30 minutes, and that geologists suggest eating one rock per day (presumably to combat iron deficiency). Memo to Google: do not train your AI on Reddit or the Onion.

There’s a quaint air of desperation in the publicity for this sudden pivot from search engine to answerbot. One hears it in the claim that it is “reimagining what a search engine can do”, that it can “unlock entirely new types of questions you never thought Search could answer”, or that it will “transform the way information is organised, to help you sort through and make sense of what’s out there”. Translation: Google is catching up with what a cheeky startup such as Perplexity has been doing perfectly well for ages.

The really big question about the pivot, though, is what its systemic impact on the link economy will be. Already, the news is not great. Gartner, a market-research consultancy, for example, predicts that search engine volume will drop 25% by 2026 owing to AI chatbots and other virtual agents, which will replace user queries in traditional search engines. Another report says that publishers could see a decline of up to 60% in organic search traffic, translating to a loss of an estimated $2bn (£1.6bn) in ad revenue.” Compared with the seismic ripples of PageRank tweaks, could this could be like the asteroid strike that did for the dinosaurs? Or is Google also a dinosaur?

What I’ve been reading

A pinch of salt
Prof Rasmus Nielsen, director of the Reuters Institute, has written an insightful essay on inadequate media coverage of AI, and how often-uncritical news articles help build up the hype.

Hello, world
There is an interesting potted history of online public messaging, from bulletin boards to Discord, by Jeremy Reiner on the Ars Technica site.

At what cost?
Molly White has written a nice acerbic essay, “AI isn’t useless. But is it worth it?”, for her newsletter Citation Needed, weighing benefits against harms.

Latest article