Support me this summer in the Clarion Write-A-Thon and help raise money for the Clarion Science Fiction and Fantasy Writers' Workshop! This summer, I'm writing The Reverse-Centaur's Guide to AI, a short book for Farrar, Straus and Giroux that explains how to be an effective AI critic.
It's been more than a year since I gave up on Google Search (I switched to Kagi.com and never looked back). I don't miss it. It had gotten terrible. It's gotten worse since, thanks to AI (of course):
Google's a very bad company, of course. I mean, the company has lost three federal antitrust trials in the past 18 months. But that's not why I quit Google Search: I stopped searching with Google because Google Search suuuucked.
In the spring of 2024, it was clear that Google had lost the spam wars. Its search results were full of spammy garbage content whose creators' SEO was a million times better than their content. Every kind of Google Search result was bad, and results that contained the names of products were the worst, an endless cesspit of affiliate link-strewn puffery and scam sites.
It's not that the internet lacks for high-quality, reliable reviews. There are plenty of experts out there who subject a wide range of products to careful assessment, laboratory tests, and extensive comparisons. The sites where these reviews appear are instantly recognizable, and it's a great relief to find them.
One such site is Housefresh.com, whose proprietor, Giselle Navarro, runs a team that produces extremely detailed, objective, high-quality reviews of air purifiers. This is an important product category: if you're someone with bad allergies or an immunocompromising condition, finding the right air purifier can exert enormous influence on your health outcomes.
As good as Housefresh are at reviewing air purifiers, they are far less skilled at tricking Google. The world champions of this are spammers, content farms that produce garbage summaries of Amazon reviews and shovel them into massive, hidden sections of once-reputable websites like Forbes.com and Better Homes and Gardens, and thus dominate the Google results for product review searches:
Google calls this "site reputation abuse" and has repeatedly vowed to put a stop to it, and has repeatedly, totally failed to do so. What's more, Google has laid off more than 10,000 workers, including "core teams," even while spending tens of billions of dollars on stock manipulation through "buyback" schemes:
Of course, the Housefresh team are smart cookies – hence the high caliber of their air purifier reviews – and they could apply that intelligence to figuring out how to use SEO to trick Google's algorithm. Rather than doing so, they took the high road: they applied all that prodigious analytical talent to researching and publishing on Google's systematic failures – and even collusion – with the spammers who are destroying the web.
This month, Housefresh released its latest report on Google's enshittification, this time with an emphasis on the "AI Overviews" that now surmount every search results page. Google has widely touted these as the future of search, a way to bypass the ad-strewn, popup-obscured, AI-sloppified (!) pages that it is seemingly powerless to filter out of its search corpus:
Rather than hunting through these SEO-winning garbage pages, you can simply refer to Google's AI Overview, which will summarize the best the internet has to offer, in hyperlegibile black sans-serif type on a white background, with key phrases helpfully highlighted in bold.
Most critiques of AI Overview have focused on how these AI Overviews are a betrayal of the underlying bargain between the web and its monopoly search engine, whereby we all write the web and let Google index it for free, and in exchange, Google will send us traffic in proportion to the quality of our work:
This is true, as far as it goes, but it doesn't go far enough. Google is a platform, which is to say, a two-sided marketplace that brings together readers and publishers (along with advertisers). The bargain with publishers is that Google will send them traffic in exchange for access to their content. But the deal with readers is that Google will help them answer their questions quickly and accurately.
If Google's marketing pitch for AI Overviews is to be believed, then Google is only shafting publishers in order to double down on its bargain with readers: to give us faster, better access to high-quality information (recall Google's mission statement, "To organize the world's information and make it useful"). If that's true, then Google is the champion of readers in their long battle with publishers, a battle in which they are nearly helpless before publishers' abusive excesses.
This is a very canny move on Google's part. Publishers and advertisers have more concentrated money than readers, but the dominant theory of antitrust since the Reagan administration is something called "consumer welfare," which holds that monopolistic conduct is only to be condemned if it makes consumers worse off. If a company screws its workers or suppliers in order to deliver better products and/or better prices, then "consumer welfare" holds that the government should celebrate and protect the monopolist for improving "efficiency."
But all that is true only if Google AI Overviews are good. And they are very, very bad.
In the Housefresh report, titled "Beware of the Google AI salesman and its cronies," Navarro documents how Google's AI Overview is wildly bad at surfacing high-quality information. Indeed, Google's Gemini chatbot seems to prefer the lowest-quality sources of information on the web, and to actively suppress negative information about products, even when that negative information comes from its favorite information source.
Indeed, Navarro identifies a kind of madlibs template that Gemini uses to assemble an AI overview in response to the query "Is the [name of air purifier] worth it?"
The [model] air purifier is [a worthwhile investment/generally considered a good value for its price/a worthwhile purchase]. It's [praised/well-regarded] for its ability to [clean the air/remove particles/clean large rooms]. Whether the [product] is worth it depends on individual needs and priorities.
This is the shape of the response that Google's AI Overview shits out when you ask about any air purifier, including a model that Wirecutter called "the worst air purifier ever tested":
What's more, AI Overview will produce a response like this one even when you ask it about air purifiers that don't exist, like the "Levoit Core 5510," the "Winnix Airmega" and the "Coy Mega 700."
It gets worse, though. Even when you ask Google "What are the cons of [model of air purifier]?" AI Overview simply ignores them. If you persist, AI Overview will give you a result couched in sleazy sales patter, like "While it excels at removing viruses and bacteria, it is not as effective with dust, pet hair, pollen or other common allergens." Sometimes, AI Overview "hallucinates" imaginary cons that don't appear on the pages it cites, like warnings about the dangers of UV lights in purifiers that don't actually have UV lights.
Google argues that AI Overview won't displace traffic to the sites it summarizes. The company points to the fact that the statements in an AI Overview are each linked to the web-page they come from. This is a dubious proposition, predicated on the idea that people looking up a quick answer on a search engine will go on to follow all the footnotes and compare them to the results (this is something that peer reviewers for major scientific journals often fail at, after all).
But the existence of these citations allowed Navarro to compile statistics about the sources that Google relies on most heavily for information about product quality:
43.1% of these statements come from product manufacturers' marketing materials;
19.5% of these statements are sourced from pages that contain no information about the product.
Much of the remainder comes from the same "site reputation abuse" that Google said it would stop prioritizing two years ago. An alarming amount of this material is also AI generated: this is the "coprophagic AI" problem in which an AI ingests another AI's output, producing ever-more nonsensical results:
Adding "reddit" to a Google query is a well-known and still-useful way to get higher quality results out of Google. Redditors is full of real people giving their real opinions about products and services. No wonder that Reddit appears in 97.5% of product review queries:
https://detailed.com/forum-serps/
Obviously, the same SEO scumbags who have been running circles around Google for years are perfecctly capable of colonizing and compromising Reddit, which has been rocked by a series of payola scandals in which the volunteer moderators of huge, reputable subreddit were caught taking bribes to allow SEO scumbags to spam their forums and steal their valor:
When it comes to product reviews, Google's AI Overviews consist of irrelevancies, PR nonsense, and affiliate spammer hype – all at the expense of genuine, high-quality information, which is still out there, on the web, waiting for you to find it.
Google CEO Sundar Pichai is unapologetic about the way that AI Overviews blurs the line between commercial pitches and neutral information, telling Bloomberg, "commercial information is information, too":
Which raises the question: why is Pichai so eager to enshittify his own service? After all, AI isn't a revenue center for Google – it's a cost center. Every day, Google's AI division takes a blowtorch to the company's balance sheet, incinerating mountains of money while bringing in nothing (less than nothing, if you count all the users who are finding ways to de-Google their lives to escape the endless AI slop):
It's true that AI loses money for Google, but AI earns something far more important (at least from Pichai's perspective): a story about how Google can continue to grow.
Google's current price-to-earnings (PE) ratio is 20:1. That means that for every dollar Google brings in, investors are willing to spend $20 on Google's stock. This is a very high PE ratio, characteristic of "growth stocks" (companies that are growing every year). A high PE ratio tells you that investors anticipate that the company will get (much) bigger in the foreseeable future, and they are "pricing in" that future growth when they trade the company's shares.
Companies with high PE ratios can use their stock in place of money – for example, they can acquire other companies with stock, or with a mix of cash and stock. This lets high PE companies outbid mature companies – companies whose growth phase has ended – because stock is endogeous (it is produced within the company, by typing zeroes into a spreadsheet) and therefore abundant, while dollars are exogenous (produced by the central bank – again, by typing zeroes into a spreadsheet! – and then traded to the company by its customers) and thus scarce.
Google's status as a growth stock has allowed it to buy its way to dominance. After all, Google has repeatedly, continuously failed to create new products in-house, relying on acquisitions of other people's companies for its mobile technology, ad-tech, server management, maps, document collaboration…virtually every successful product the company has (except Search).
For so long as investors believe Google is growing, it can buy other companies with its abundant stock rather than its scarce dollars. It can also use that stock to hire key personnel, which especially important for AI teams, where compensation has blasted through the stratosphere:
But that just brings us back to the original question: why build an AI division at all?
Because Google needs to keep up the story that it is growing. Once Google stops growing, it becames a "mature" company and its PE ratio will fall from 20:1 to something more like 4:1, meaning an 80% collapse in the company's share price. This would be very bad news for Googlers (whose personal wealth is disproportionately tied up in Google stock) and for Google itself (because many of its key personnel will depart when the shares they've banked for retirement collapse, and new hires will expect to be paid in scarce dollars, not abundant stock). For a company like Google, "maturity" is unlikely to be a steady state – rather, it's likely to be a prelude to collapse.
Which is why Google is so desperately sweaty to maintain the narrative about its growth. That's a difficult narrative to maintain, though. Google has 90% Search market-share, and nothing short of raising a billion humans to maturity and training them to be Google users (AKA "Google Classroom") will produce any growth in its Search market-share. Google is so desperate to juice its search revenue that it actually made search worse on purpose so that you would have to run multiple searches (and see multiple rounds of ads) before you got the information you were seeking:
Investors have metabolized the story that AI will be a gigantic growth area, and so all the tech giants are in a battle to prove to investors that they will dominate AI as they dominated their own niches. You aren't the target for AI, investors are: if they can be convinced that Google's 90% Search market share will soon be joined by a 90% AI market share, they will continue to treat this decidedly tired and run-down company like a prize racehorse at the starting-gate.
This is why you are so often tricked into using AI, by accidentally grazing a part of your screen with a fingertip, summoning up a pestersome chatbot that requires six taps and ten seconds to banish: companies like Google have made their product teams' bonuses contingent on getting normies to "use" AI and "use" is defined as "interact with AI for at least ten seconds." Goodhart's Law ("any metric becomes a target") has turned every product you use into a trap for the unwary:
There's a cringe army of AI bros who are seemingly convinced that AI is going to become superintelligent and save us from ourselves – they think that AI companies are creating god. But the hundreds of billions being pumped into AI are not driven by this bizarre ideology. Rather, they are the product of material conditions, a system that sends high-flying companies into a nosedive the instant they stop climbing. AI's merits and demerits are irrelevant to this: they pump AI because they must pump. It's why they pumped metaverse and cryptocurrency and every other absurd fad.
None of that changes the fact that Google Search has been terminally enshittified and it is misleading billions of people in service to this perverse narrative adventure. Google Search isn't fit for purpose, and it's hard to see how it ever will be again.
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
I'm on tour with my new, nationally bestselling novel The Bezzle! Catch me in TORONTO on Mar 22, then with LAURA POITRAS in NYC on Mar 24, then Anaheim, and more!
A key requirement for being a science fiction writer without losing your mind is the ability to distinguish between science fiction (futuristic thought experiments) and predictions. SF writers who lack this trait come to fancy themselves fortune-tellers who SEE! THE! FUTURE!
The thing is, sf writers cheat. We palm cards in order to set up pulp adventure stories that let us indulge our thought experiments. These palmed cards – say, faster-than-light drives or time-machines – are narrative devices, not scientifically grounded proposals.
Historically, the fact that some people – both writers and readers – couldn't tell the difference wasn't all that important, because people who fell prey to the sf-as-prophecy delusion didn't have the power to re-orient our society around their mistaken beliefs. But with the rise and rise of sf-obsessed tech billionaires who keep trying to invent the torment nexus, sf writers are starting to be more vocal about distinguishing between our made-up funny stories and predictions (AKA "cyberpunk is a warning, not a suggestion"):
In that spirit, I'd like to point to how one of sf's most frequently palmed cards has become a commonplace of the AI crowd. That sleight of hand is: "add enough compute and the computer will wake up." This is a shopworn cliche of sf, the idea that once a computer matches the human brain for "complexity" or "power" (or some other simple-seeming but profoundly nebulous metric), the computer will become conscious. Think of "Mike" in Heinlein's *The Moon Is a Harsh Mistress":
For people inflating the current AI hype bubble, this idea that making the AI "more powerful" will correct its defects is key. Whenever an AI "hallucinates" in a way that seems to disqualify it from the high-value applications that justify the torrent of investment in the field, boosters say, "Sure, the AI isn't good enough…yet. But once we shovel an order of magnitude more training data into the hopper, we'll solve that, because (as everyone knows) making the computer 'more powerful' solves the AI problem":
As the lawyers say, this "cites facts not in evidence." But let's stipulate that it's true for a moment. If all we need to make the AI better is more training data, is that something we can count on? Consider the problem of "botshit," Andre Spicer and co's very useful coinage describing "inaccurate or fabricated content" shat out at scale by AIs:
"Botshit" was coined last December, but the internet is already drowning in it. Desperate people, confronted with an economy modeled on a high-speed game of musical chairs in which the opportunities for a decent livelihood grow ever scarcer, are being scammed into generating mountains of botshit in the hopes of securing the elusive "passive income":
Botshit can be produced at a scale and velocity that beggars the imagination. Consider that Amazon has had to cap the number of self-published "books" an author can submit to a mere three books per day:
As the web becomes an anaerobic lagoon for botshit, the quantum of human-generated "content" in any internet core sample is dwindling to homeopathic levels. Even sources considered to be nominally high-quality, from Cnet articles to legal briefs, are contaminated with botshit:
Ironically, AI companies are setting themselves up for this problem. Google and Microsoft's full-court press for "AI powered search" imagines a future for the web in which search-engines stop returning links to web-pages, and instead summarize their content. The question is, why the fuck would anyone write the web if the only "person" who can find what they write is an AI's crawler, which ingests the writing for its own training, but has no interest in steering readers to see what you've written? If AI search ever becomes a thing, the open web will become an AI CAFO and search crawlers will increasingly end up imbibing the contents of its manure lagoon.
This problem has been a long time coming. Just over a year ago, Jathan Sadowski coined the term "Habsburg AI" to describe a model trained on the output of another model:
There's a certain intuitive case for this being a bad idea, akin to feeding cows a slurry made of the diseased brains of other cows:
https://www.cdc.gov/prions/bse/index.html
But "The Curse of Recursion: Training on Generated Data Makes Models Forget," a recent paper, goes beyond the ick factor of AI that is fed on botshit and delves into the mathematical consequences of AI coprophagia:
https://arxiv.org/abs/2305.17493
Co-author Ross Anderson summarizes the finding neatly: "using model-generated content in training causes irreversible defects":
Which is all to say: even if you accept the mystical proposition that more training data "solves" the AI problems that constitute total unsuitability for high-value applications that justify the trillions in valuation analysts are touting, that training data is going to be ever-more elusive.
What's more, while the proposition that "more training data will linearly improve the quality of AI predictions" is a mere article of faith, "training an AI on the output of another AI makes it exponentially worse" is a matter of fact.
Name your price for 18 of my DRM-free ebooks and support the Electronic Frontier Foundation with the Humble Cory Doctorow Bundle.
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
