Recall the dogs out

Wednesday, June 5, 2024

Comments: 39   (latest 2 days later)

Tagged: recall, microsoft, windows, ai, llms, copilot, stross

You can't sneeze right now without blowing over a "Microsoft's Recall is terrible" headline. The most positive take I've seen is "you might find this useful, but there are privacy concerns". The median take is less than positive. For details, check Kevin Beaumont's post which is the substantive analysis behind the current shouting.

I was struck, though, by Charlie Stross's post this morning, titled "Is Microsoft trying to commit suicide?"

Stross is a sharp writer and a sharp tech observer, and I entirely agree with the body of his missive: that Recall is terrible for many reasons. (Although he misstates that the feature is "impossible to disable". As commenters note, Recall installs default-enabled, but you can disable it in settings.)

But as a thesis statement, I am really not on board. Because the obvious answer is "No." Microsoft is not trying to commit suicide. Microsoft is trying to survive. And it's worth considering what this calamitous self-own tells us about Microsoft's mindset.

Microsoft's dedication to AI-as-a-feature began with the original Github Copilot, announced in 2021. I think it was the first big Github move after Microsoft acquired it. That is, Copilot was a clear signal of Microsoft's priorities as OpenAI's investor and Github's owner.

Since then we've gotten a thundering cascade of "you have to use AI" announcements, either from Microsoft or from the bubble-wave of startups committed to using OpenAI compute time. The rest of the equiv-tech stratum are cranking away on their own AI roadmaps, but they aren't really driving the hype. Apple wants to improve Siri. Google wants to improve Google search (and they're not fixated on AI as the only answer). Facebook has no actual ideas so why not open-source. None of them really seems interested in beating Microsoft to any particular punch.

But Microsoft is behaving like they are. MS acts like they have to be first to market, whatever market it is that Recall serves, and GoogAppleFace are hot on their heels.

Why? Spoiler: I don't know. I don't sneak into Microsoft's board meetings.

Stross suggests "to sell Windows on ARM", but I don't like that answer. ARM-based laptops sell themselves: vastly improved performance, vastly improved battery life, or any tradeoff you like between the two. Apple has been smirking about it since 2020 and Microsoft just needed to catch up, which they now have.

But the point is, it doesn't matter what catastrophe Microsoft sees in the future where they come in second. They think they're doomed without AI. Conventional wisdom is "Azure is doing fine", but maybe that's not true. Maybe it's only true if you factor in sky-high demand for AI compute. Maybe Satya Nadella sits on a Silicon Throne that stabs him in the butt every day that Microsoft isn't a market leader in something. (Windows no longer counts -- OSes are loss leaders now.)

"They think they're doomed" explains a lot about this whole situation. Notably, nobody at Microsoft is allowed to think about the downsides. AI is a bet they cannot hedge. Feature not ready for prime time? We're not delaying it, because this plan cannot fail. AI spews lies and hallucinations? Customers will love it anyway, because this feature cannot fail. Market not materializing? Shut up, yes it will.

Anybody who says otherwise gets ignored or steamrolled. La la la, "...when his salary depends on his not understanding it."

This is not to say that they cannot course-correct. Microsoft wants this stuff to succeed, not to destroy them. I suspect they will improve the security story around Recall. Maybe not by June 18th (the ship date of "Copilot+ PCs"), but over the course of this year.

Will they improve it enough? I have no idea. They have lots of smart engineers. Maybe they'll flex enough to consider a disable-at-install-time option. I don't know how it will shake out.

I expect a lot of corporate policies saying "turn that crap off on your work machine." I certainly will on my Windows box (which runs nothing but Steam anyhow). That's not really the question. The question is what security exploits arise to target the average user -- the one who never changes the defaults and clicks any "Install" button that pops up.

The longer-term question is, what happens if the AI bubble really does sputter out and MS leadership stops believing it will save them?

I don't mean "when AI goes away". I think we're past the point where it will go away. I mean the likely outcome: people stop shouting "AI will solve all human problems!" and start treating it like, I don't know, copy-and-paste. A neat trick which is ubiquitous and makes some things easier, but nobody thinks about it much.

(I remember the first article I read about copy-and-paste, back when it was a Xerox PARC tech demo. This Alan Kay article in 1979? Mm, not sure. Anyway, I didn't get it. "How does that help?" Hard to see until it's available everywhere, but I sure copied and pasted a lot in writing this post.)

Anyway, neat tricks are commodity, not differentiator. If it's ubiquitous, it'll be on Windows and Mac and anything Google-connected. Where will Microsoft be then? Doomed? Or just another tech company with some revenue streams?

I know that seems like a fate worse than death in today's venture-unicorn market, but somebody's gotta do it.

Comments from Andrew Plotkin

Comments from Mastodon