The Biggest Danger of AI May Be One That No One is Talking About (Premium)

Microsoft’s dueling announcements about the cost of AI to its customers can serve as a wake-up call for the true danger of this technology: it will only exacerbate one of the worst abuses of Big Tech by expanding and refining how it violates our privacy and tracks us online and then sells that information to advertisers.

A few recent events triggered this realization.

Last week, Microsoft confirmed what two sources had told me in confidence: Because AI is so expensive, it will charge Microsoft 365 commercial customers—that is, businesses—an additional $30 per user per month on top of existing monthly charges for access to the AI-powered Microsoft 365 Copilot. (And it’s worse than it sounds because only customers on the more expensive Microsoft 365 tiers will even be offered this upgrade.)

But Microsoft also confirmed to an enthusiast blog that its AI-powered and consumer-focused Bing chatbot “will remain free via Bing.com, Microsoft Edge side panel, Windows Copilot and other places.”

These two offerings seem at odds with each other. And given the stark reality of the economics of paid subscriptions vs. ad-supported services, this should raise some questions if not alarm bells. If it doesn’t, I can explain, using a well-worn but timely example.

The music streaming service Spotify has both paid subscribers and ad-supported users, and it just reported its most recent quarterly earnings. It now has 551 million total monthly users, of which 220 million of them are paid subscribers and 317 million are ad supported. But paid subscribers, which represent 40 percent of the user base, represent 88 percent (€2.8 billion) of the firm’s total revenues (€3.2 billion), while ad-supported subscribers, which represent 60 percent of the user base, only contributed 12 percent (€404 million) of the revenues. Long story short, paid subscribers bring in much more revenue than ad-supported revenues, even when it’s a small group of users.

Looking back at Microsoft and AI, it’s hard to come up with exact numbers when it comes to the relative user bases of Microsoft 365 commercial and Bing, of course. But it’s important to remember that Microsoft aspires to turn Bing into a Google-like ad-based powerhouse, and that its push to AI, damn the current costs, is all about chipping away at Google’s lead in this market. (That is, we as individuals think of this as “search engines,” but the business is really selling ads.)

We know that AI is expensive right now—each AI interaction costs an estimated 10 times that of a typical web search—and that AI providers like Microsoft and Google are doing everything they can—including designing and using their own custom-designed datacenter chipsets—to lower costs.

This won’t happen overnight, and so these companies are incurring huge quarterly costs related to AI until and unless some combination of workable business models and reduced operating costs arrives to save the day. It will be interesting to see when—not if, but when—these companies start reporting these costs, as they will have to. And it could happen as soon as today, as both companies will report their most recent quarter earnings later this afternoon.

Anyway, Microsoft’s decision to charge its corporate customers for AI makes sense in some ways, because this is a proven business model for those customers: Microsoft has long offered monthly per-user subscriptions to these customers, and monthly add-on services on top of them, the most recent example being Teams Premium (which also seems very expensive to those outside this world). The economics make sense, as the decision-makers at the companies that use Microsoft 365 can justify the expense: That additional $30 per user per month, which might only be applied to some users, represents some number of employees they will not have to hire, pay, and provide benefits for, at an even greater cost.

But Microsoft can’t apply this thinking to its consumer customers, which are both smaller in number and more likely to jump ship to competing services because doing so is almost completely frictionless and easy. If a consumer doesn’t like Bing Chat, for example, they can simply move to Google, ChatGPT or any number of other free AI services. And so it will monetize these users like many other online services do, via advertising.

But there is a big difference between Microsoft’s ad business and that of most other online services. That is, Microsoft isn’t using a service for ads, Microsoft is the service. And as with Amazon, Google, Meta, and the small number of other exclusively Big Tech companies with huge ad businesses (or at least aspirational ad businesses), Microsoft is double-dipping. It sells the ads you see when you visit its services to third parties, and it’s making this market lucrative for those companies by tracking all your online activities and selling them to others. Like its competitors, Microsoft knows a lot about you, and that information is very valuable to advertisers. If Bing takes off like Microsoft hopes because of AI, perhaps the additional advertising revenue that follows will help offset and then cover its AI costs over time.

I write this now because of a recent event that drove home how this system already works today with startling clarity.

As part of a coming “What I Use” article I’m writing for our current trip to Mexico City, I looked up the Bose QuietComfort Earbuds II that I purchased from Amazon back in May and was able to wear on a plane for the first time when we flew here. This is one of those weird areas where I use a particular tech product, but haven’t reviewed or otherwise really written about it at all (for reasons I can’t explain). And in this case—it’s literally about what I use—I just wanted to mention the change and write a bit about how they differ from my previous ANC earbuds. It’s the type of thing I do all the time and don’t really think much about.

But over the next several days, I noticed something odd, albeit something all of you have noticed, too, from time to time. Reviews of the Bose wireless earbuds I looked up on Amazon, which are now 8 or 9 months old, given when the product was originally released, started popping up in my Google Discovery feed. Each time, it was denoted by an “in case you missed it…” subheading. And there were no other older links like that. It was just the Bose earbuds, again and again and again.

Coincidence? Come on.

Look, I’m about as anti-conspiracy theory as you can get, which is especially notable if you understand how hard-wired we are as humans to make connections that aren’t there was a natural result of millennia of evolution. (The canonical example is that the proto humans who thought they saw a predator in the long grass were the ones who survived and kept breeding, while those who did not were killed off; the conspiracy theory bit comes from the fact that, most times, there was no predator and we’re now wired to see it anyway.)

Even with my privacy radar up, I had never considered the ramifications of what I had done: I navigated to Amazon using Brave, a very safe and private browser, but I was also signed into my Amazon account so I could look up my previous purchases to more easily find the Bose earbuds. Gotcha. Had I done this in what Brave calls Private mode, and other browsers call Incognito or InPrivate mode, maybe I could have avoided Amazon’s ad network from commingling with Google’s and using their combined knowledge of my activities to start showing me things it assumed I was interested in. Maybe. But I didn’t.

So here’s my conspiracy theory.

Amazon and Google are both among the biggest AI-capable firms in the world. We know that AI queries cost about 10 times that of normal search queries, and we know that only the biggest companies in the world can afford this. We also know that online advertising isn’t a great business unless you’re also one of the bigger user-tracking and ad-supplying companies in the world too, as both Amazon and Google are. Hell, over 80 percent of Google’s revenues come from ads. If this ad business didn’t exist, Google wouldn’t exist for all intents and purposes. It certainly wouldn’t be a Big Tech company. It might not even be a Fortune 500 company.

Put simply, advertising is incredibly lucrative for this tiny handful of companies, Microsoft included, though it wants a much bigger piece of that pie. And given one of AI’s biggest strength, its ability to tie together disparate information that’s already out there in the world, it is perfectly reasonable to believe that AI can be and maybe already is being used to improve advertising. These companies are already selling our personal data. But now they can put that together in far more meaningful ways, and then develop even more accurate and invasive profiles of us all. One conjures up an image of AI as Hannibal Lector, able to derive intensively private information from what seems like just a little bit of disclosure.

This, I think, is the real aim of Bing Chat and all other public-facing AIs that will be ad-supported and not paid subscriptions (and, likely, even those that are paid). AI will pray on most people’s sad indifference to the tracking and privacy violations they know are already occurring. But it will do so much more with that data. And as the costs of AI to companies like Google and Microsoft comes down, as they will, their profits will only go up. This may literally be the next great technology gold rush, the next lucrative business model, with the caveat that it will only benefit Big Tech, the few companies that can afford to invest now in a meaningful way, when the costs are so high. And so the big gets bigger, as always, this time at the expense of our privacy.

Given all this, it’s no wonder that all the major players in AI raced to adopt the safeguards recommended by the Biden administration. Those safeguards are all about the largely imagined AI dangers everyone is talking about, the hallucinations, and so on, the issues Bill Gates summarily dismissed in a recent editorial. Left unguarded and unmentioned is how AI vendors will do what they’re already doing without AI, by taking the personal information they scrape and steal to violate our privacy and track us online, and then sell our personal information to advertisers. (Don’t believe me? The word “privacy” comes up only once in the White House announcement about this agreement, and that’s only as part of AI research, not implementation.) This business model isn’t just ignored, it’s implicitly protected because everyone is fake-hustling to address the many other AI concerns. Keep everyone busy, and pay no attention to the man behind the curtain. Or the AI, in this case.

And that’s why I feel comfortable discussing my conspiracy theory with you. It’s already happening, so it’s real. And there is literally no reason to believe that any Big Tech company that handles advertising today will not use AI to extend and improve today’s successful business model. It’s just common sense. So say what you will about the dangers of AI. Its biggest danger, perhaps, is the one no one is talking about.