# Four layers of crutches > Source: > Published: 2026-08-14 09:28:48+00:00 I just found it interesting to realize that when working with knowledge, we’re relying now on four layers of crutches: - Language - Writing - Computers - (Generative) AI Before language, there was knowledge that wasn’t encoded in a linear sequence of sounds that evoke images. But that’s too far to fathom. I wonder if some hominids didn’t carry misgivings about teaching (say) hunting through means that were not exclusively shared experience. It is much easier to think about a time where there was little written language. Socrates himself railed against the use of written language as a vessel for thought, because of its deleterious effects on memory and the imperfect way in which imprinted meaning (compared to oral teachings). We know about Socrates only because Plato wrote what he had to say. After centuries of explosion of writing, we finally developed computers. Computers not only gave us instant retrieval of essentially anything that was written, but it allowed us to automate the generation and processing of information. Writing then became not just something that guided humans through knowledge, but also something that directed the production of more knowledge (or at least, more information). We lost calligraphy and we kept on offloading our memory and processing even more. The explosion of AI into the mainstream adds a probabilistic component to computers. Before AI, almost everything that computers did was fully deterministic: a certain operation X would give you the same result Y. The triumph of AI is the triumph of probabilistic results, where you cannot predict what you’ll get. In that uncertainty, curiously, there is enough flexibility so that computers can now deal with a much broader set of inputs and make sense of them, to the point where we consider AI to be quite intelligent. AI can now almost effortlessly eliminate the blank page (or the blank search input). It now takes active effort to not offload most of our knowledge production, even our thinking, to AI. (Side note: it is really remarkable that what we stopped considering computers stupid when we found a way to make them flexible enough through probabilistic methods.) Each layer of crutches gives us great power, but there’s also great power in removing them, experimentally, in one’s own adventures. I don’t think you can mix and match (ie: remove written language but keep computers), so you can only choose at what level you work. But you can mix different levels in a project, by doing (say) X with AI and Y without. There could be interesting results at lower (higher?) layers, but it would be remarkable that we would go back and do most of our everyday production without all four sets of crutches. And it’d be also interesting to operate at lower levels as a sort of training, so that by doing without certain crutches you can then use them more strongly when you add them back.