Daniel Susskind's argument for AI tutors has drawn two letters in reply, and neither of them disputes what the software can do. The Peel and Galia Collective in London goes after one sentence — that human mentors are too expensive to provide for everyone — and answers it with a question of its own: expensive compared with what? Jane Roland Martin, who starts by agreeing with Susskind, gets stuck on a different sentence: his assumption that the basic skills of the past will still be the basic skills of the future. Between them the two letters mark the places where the case for machine tutoring is load-bearing.
The collective's position is redistributive and it says so plainly. Direct resources toward mentors for everyone, rather than letting the owners of technology companies accumulate enormous fortunes, and you reduce the anxiety about lost jobs while giving every child the quality of education that well-off, screen-avoiding families currently want only for their own. The students now protesting against AI, the collective writes, are not only worried about their chances after graduation; they can see that deploying this technology enriches a few and worsens working conditions for most. Children do not need new data centres, or automation that consumes vast resources and does serious environmental damage. A good future for children, the letter says, will not be built by education-technology grifters.
The second half of the letter takes on imitation. Susskind is impressed by these programs' ability to imitate cultural products that already exist — stories, podcasts. The collective's answer is that if what you want is new ideas rather than reworked versions of old ones, you have to give up the appeal of endless personalisation that chatbots offer. Susskind separates educational AI from social media, which he calls "dehumanising"; the rejoinder is sharper than it first looks. Social media at least once helped people stay in contact with one another, while AI, by the collective's reading, does not create that connection at all.
The letter ends on a small scene Susskind borrows from Sam Altman: on a car journey it is convenient to play a podcast about your own children's interests. Those interests, the collective replies, are developed first of all by talking about them. A society is not a set of tasks waiting to be solved but a run of substantive interactions with other people — people who get tired and distracted, which is exactly why their time has a particular value.
Martin's letter arrives from the other direction. She endorses Susskind's central warning: parents fail their children when they prepare them for the world the parents lived in instead of the world the children will live in. What she cannot follow is why the foundations of that older world should remain foundational. Her evidence is her own recent week. She signed an agreement disposing of an inheritance over the internet by following a link, and she dictated this letter into the microphone of a desktop computer. Those two examples, she argues, are already enough to cast doubt on the received idea of what the necessary basics are. She agrees with Susskind that AI should be used critically rather than blindly — and says that demands the same critical treatment of a school curriculum designed to prepare children for adult life in the 19th and 20th centuries, instead of accepting its contents as given.
The two letters do not agree with each other, and that is the more interesting fact about them. One says the content of education is roughly right and the funding is a political choice; the other says the funding question is secondary to a curriculum built for a century that ended. Put together they cancel: if the basics are due for wholesale revision, it is not obvious what the universally funded human mentor is supposed to be teaching.
Neither letter costs its own proposal. The collective's argument turns on cost being a distributive choice rather than a hard constraint, which is a serious claim, and it passes without a single number — not what a mentor for every child comes to, not what the fortunes being redirected amount to, not how the two compare. Martin's examples prove less than she needs them to. Clicking a link to execute a legal document and dictating a letter are evidence that interfaces got easier, not that literacy stopped being foundational; you still have to understand what you signed.
What survives both letters is Susskind's sentence. Nobody in this exchange contests that human mentors at scale cost more than software at scale; the collective says pay for it anyway, Martin says teach something else. That asymmetry is the whole engine of the AI tutoring argument, in education and everywhere else it will be made next: one side of the ledger requires a budget line that has to be won every year, the other requires a subscription that is already being paid.