Do Students Dream of AI?
by Marcus Birkenkrahe
When a human wants to dream (in his sleep, I mean), it is not beneficial to be interrupted all the time. That explains, to me at least, why it is so hard to dream about what to do with AI.
Dreaming requires immersion: sinking deeply into a topic, mulling it over for a long time, sometimes for years, until the mind comes up with the next thought, or with another question, or with a solution. This is more like the recursive creation of a puzzle than the solving of one. AI, as we use it now, is a puzzle-solving machine: it answers whatever we put in front of it, quickly and (mostly) well. It doesn’t sink, and it doesn’t mull. Ask it something you’ve been turning over for weeks, and it hands back an answer before your own mind has had the chance to come up with the next question. Dreaming also needs other humans, dead or alive, inaccessible or accessible, met through their written or spoken works or through conversation.
When a field changes too fast, this process is disrupted. Not that the field necessarily changes in its foundations (in the case of AI we don’t even know what those foundations are), but it changes daily as far as its input for our human dreams is concerned. Every new model, tool or announcement is another interruption, and the dreamer never gets past the first stage of sleep.
That’s why it is difficult for my students to answer honestly, or deeply, or at all, when I ask them “What do you want to do with AI?”, or when they try to answer the question of their own career aspirations. Sure, they have aspirations. But the word comes from the Latin spirare, to breathe, and breathing is made difficult at present, because the pace of development is so rapid.
I don’t have a remedy for this, other than the obvious one, which is also the hardest: to slow down on purpose, and to look away from the daily news long enough for something to settle. That goes for my students, for my research, and for me. Whether the field will let us is another question.