(An edited translation of this Norwegian blogpost, for the benefit of my non-Norwegian students.)

I’m currently reading and discussing a lot about artificial intelligence (like everyone else, for that matter, but I’m on a sabbatical and new – in this case, not so new – technologies are what I’m working on and teaching about.)

One of the reflections I have about it is “wow, so little has changed.”

I first got into the discussion about artificial intelligence in the 80s, when I worked in the IT department at BI and was lucky enough to have a boss (Erling Iversen) who read a lot and was plugged into the AI ​​discussion as it was then. Through him, Fred Wenstøp, and eventually Charles Stabell and Øystein Fjeldstad, I was introduced to books like Heins Pagel’s Dreams of Reason, Bolter’s Turing’s Man, Hofstadter’s Gödel, Escher, Bach (as Erling used to say, “there are two kinds of IT people, those who have read Hofstadter and those who have not.”), Rumelhart and McLelland’s Parallel Distributed Processing and somewhat later Resnick’s Turtles, Termites and Traffic Jams. (Blogged about here.)

All of these books, and the ideas they contain, can help us understand what artificial intelligence is and how we should handle it. That understanding is important, because today AI is starting to become commonplace (much like PCs in the late 80s, the internet in the second half of the 90s, mobile and social media in the 2000s) and I see in any case that the same misunderstandings and expectations that were prevalent then also apply to AI.

Then it might be worth looking at some previous discussions about AI, because almost all the arguments you see have been made before and have not lost their validity, even though language models can now be run rather than just described.

The most important criticism of AI – and especially whether machines could ever achieve some form of intelligence – came from Hubert Dreyfus, a philosopher who questioned many of the assumptions that underpinned the idea that if a machine becomes powerful enough, it will become smart. Dreyfus articulated early on what is becoming clearer and clearer as neural networks become commonplace: That the brain is not a computer, even though parts of it can be simulated by a neural network without context; and that a neural network is not a brain, even though it can produce something that at first glance looks like intelligence. Thinking is not computation, and computation is not thinking, quite simply.

I found a film on YouTube the other day that describes and illustrates many of Hubert’s points, especially that knowledge is created in a context and cannot be reproduced freely from it. The film takes a number of artists and others with deep knowledge, and shows how that knowledge and the results of it occur in interaction with both one’s own body and the environment. As Dreyfus says in the film: “the source of meaning in our lives isn’t in us – that’s the Cartesian tradition – and it isn’t in some supreme being, but it is in our way of being in the world.” (https://youtu.be/fcCRmf_tHW8?t=4629).

I find one of the statements particularly interesting: “One of the dangers of technology is that it relieves us of the necessity of developing skills.” Ideas and knowledge often come as a result of limitations, of obstacles that one must overcome. With the shorter and shorter distance between desire and fulfilment, will we all end up as spoiled brats?

Probably not, because we want context and interaction.

Plus ça change, plus c’est la meme chose…