I had assumed AI adoption was still on its way up and would keep going for a while yet. Ofcom reckons 54% of UK adults now use AI tools like ChatGPT, Copilot or Gemini, up from 31% in 2024, which means a bit under half don’t, and I had those people down as not having got round to it yet. Pew’s American figures from February say otherwise. Among people who don’t use chatbots, 60% give lack of interest as a major reason and only 29% say not knowing how, and around seven in ten say they’re unlikely to start within the year. That isn’t a queue waiting to be shown, it’s a decision most of them have already made, and I had it the wrong way round.
I noticed I had it wrong because of a piece Brian Madden wrote at the end of June about how he reads AI news, where the short version is that he ignores most of it. He’s a futurist at Citrix, and his argument is that the job isn’t predicting what happens but narrowing the range of things that plausibly could, which he does by living about six months ahead of everyone else so his starting point is further along than theirs. He doesn’t need to see further, he just needs to stand closer. Brian also borrows the Bezos move of asking what stays the same rather than what changes, which for him comes out as workers still needing authenticating and someone still needing to answer what ran and who told it to run. That’s a good list and it’s the list you get from where he’s sitting, which is a vendor watching how work gets secured.
The trouble is that standing closer is his job and it isn’t mine, and it tells you what a technology can do rather than whether anyone will bother. Those turn out to be different questions, and the Pew numbers are what happens when you answer the second one by extrapolating from the first. So the three things below are what I use instead, and they’re all things I’ve watched happen rather than things I’ve predicted.
The exit
The first is that leaving gets harder than arriving, and Microsoft spent about two years from 2004 building a version of that for its own customers. The scheme was PlaysForSure, and it existed to promise that a track bought from one of these certified stores would play on one of these certified players. Then in late 2006 they shipped the Zune, and the Zune wouldn’t play PlaysForSure files. The EFF pulled the detail out of Microsoft’s own press release: the footnote listed what Zune software could import, unprotected WMA, MP3 and AAC, and protected WMA simply wasn’t on it. Anyone who had bought music on Microsoft’s own recommendation either re-bought their library or stayed put. MSN Music had already shut by then, and Microsoft kept issuing DRM keys for a while so people could move purchases to a new machine, before deciding the support wasn’t worth it and stopping, which meant those tracks would fail the next time someone changed hardware.
The Zune was a reasonable device and it didn’t matter, because the people it hurt most were the ones who had followed the vendor’s advice, and they were the ones who couldn’t move.
No opt-in required
The second is that you can decline something entirely and still get whatever it does to everything around you. I’ve never had TikTok, never installed it, and don’t have much intention of doing so. It has still rearranged shops I walk into, with boards inside and outside bookshops now saying BookTok and tables organised around what’s been popular on a platform I’ve never opened. Those books are there because of decisions made by people responding to something I can’t see, and declining the app didn’t keep any of it away from me, it just meant I couldn’t tell what was causing it.
Declining the app does not prevent its effects from rearranging what appears around you.
The bookshop tables are a choice someone made, but a lot of what arrives without asking is arithmetic. Vizio, who make budget televisions, ran two business lines and reported them separately after going public in 2021: Device, which is the TVs and soundbars, and Platform+, which is advertising and viewer data from their content recognition system. In the third quarter of that year Device brought in around five hundred million dollars of revenue and made $25.6 million of gross profit, while Platform+ brought in $86 million and made $57.3 million. By late 2024 the device side was losing money outright while Platform+ made $115.8 million, and Walmart bought the company that December. Their CEO said on an earnings call that they’re willing to sit at a low margin on the TV itself because the unit becomes valuable once it’s in someone’s house, which is why the version without the data collection isn’t really sold anymore.
Declining is not the same as not knowing
The third is the one I got wrong at the top, and it’s worth being specific about why. Not adopting something looks the same from outside whether the person has never heard of it or has heard plenty and doesn’t fancy it, and those two produce completely different forecasts. The first one closes on its own as awareness spreads. The second doesn’t close at all.
Which puts me in that group on TikTok rather than outside it. I’m not behind on it, I’ve declined it, and what that costs is that a fair amount of what’s going on reaches me late or in translation. Reply All, the Gimlet podcast that ran until 2022, had a recurring segment called Yes Yes No where the hosts took an internet reference their boss didn’t understand and worked through it line by line until he did. I listened to that for years in London and didn’t know most of the references either, which was the enjoyable part.
Three predictions
Those three are what I’ve got instead of standing six months ahead, so here’s what they give me, with what would show me I was wrong.
Cars will not be sold new without a permanent cellular connection by 2030, in any mainstream segment. I’m looking at buying one at the moment and there isn’t a good option, which is a strange thing to find in a market this size. Used doesn’t solve it either, partly because used isn’t the cheaper route it used to be. Wrong if a volume manufacturer ships something genuinely offline and sells it.
Paying a subscription for capability in hardware you already own becomes ordinary rather than a story by 2030. Cars are where it’s most visible with heated seats and performance unlocks, though it’s already in cameras, printers and appliances. Wrong if a regulator or enough consumer pushback kills the model.
Passkeys will succeed and passwords will survive underneath as the fallback, and the fallback is where the attacks go. Every non-technical person I’ve raised it with says the same thing, which isn’t that it’s difficult but that they’re worried about what happens if they lose the phone, and no amount of explaining that it’s more secure has moved that. Wrong if the password fallback actually gets removed at scale rather than just deprecated in the documentation.
The 2008 panel in the Pew survey I wrote about last time got the device mostly right and said very little about who would end up holding one. I’ve just done the same thing in the other direction, and I’ll find out around 2030 which half I got wrong.