A whole cottage industry has erupted to litigate the ethics of AI. It's sprawling all around us. Politicians eager to burn taxpayer money to protect us all from something so new, so unprecedented, so magical, so potentially catastrophic that we so absolutely must get in the way. Endless commissions. Two-hundred-thousand-dollar-a-year gigs off a completely fraudulent intellectual debate.
Can I offer some context?
A hundred years ago, the anthropologist Bronisław Malinowski wrote a brilliant essay, Science, Magic and Religion. Malinowski's insight was that magic is not a primitive precursor to religion. Magic is a form of science. You are not communing with the divine --- it's about pressing a button to get a result.
You don't know how that machine works. You just know that when you press its button, something happens. The coffee brews! That desirable scent fills the air! And you don't need to understand the $5,000 worth of engineering inside that steel box. You just enjoy your coffee.
Malinowski observed that magic serves a psychological function: it provides an illusion of control over the practically uncontrollable. When the Trobriand islanders fished the calm inner lagoon, no magic was called for. Their practical knowledge was sufficient. But then they also went deep-sea fishing. It was there that storms, currents, and sharks could kill without warning. So it was there, in the deep seas, where the Trobrianders performed their magic rituals.
Trobrianders were not irrational. But uncertainty is terrifying, you are human, you need a lever to pull.
A hundred years have passed. What we call AI today is not a mystical thinking machine. It is an implemented form of statistics. At its most fundamental, conventional AI is an animated Markov model of chain probability.
Andrey Markov was the mathematician who enabled us effectively to model cognition as a sequence of linked probabilistic events.
For every link in a chain of data, a statistical probability dictates the next link. Start with "can of beer" and the model calculates: 90% probability the next link is Uncle Erind, 50% Bora, 0.01% Maia. You ascertain the known events, you calculate the probability of the next one.
So that's what conventional artificial intelligence is. It's not magic. It's a model for statistically predicting the next link in a chain.
We understand this completely --- we built it. It only seems mystical when the mechanics are hidden from the observer. Pull back the curtain and instead of magic you see the tool. The magic only happened because you commanded it to happen.
The real magic is marketing. Some of those $200k-a-year roles are threatened by machines that can now do in seconds what took teams of people weeks. Nursing confusion seems to be the magic here --- the ritual to perform over this sea of threat is to invoke danger. You feel you can continue making a pretty penny by convincing the rest of the tribe that a statistical model is an existential threat. That is the illusion of control in action: if you can make the public afraid of the tool, maybe the tool won't replace you.
Is it always cynical and conspiratorial? Not entirely, I don't think.
Some people have genuinely not recovered emotionally from the shock of a machine mimicking human language so fluently. We believed that language was uniquely ours. Now it seems to be replicable by a sophisticated chain of probabilities.
Here is my answer to your question about the danger of AI: the most immediate danger is the one being manufactured --- by those who seek profit from your fear of it.
Ongoing thread. More from EP Pajo to follow.
