From Laplace’s Demon to AI Ethics: Biases, Surveillance, and the Limits of Control

During their latest episode of the VALUE: After Hours Podcast, Carlisle, Taylor, and Eric Cinnamond discussed From Laplace’s Demon to AI Ethics: Biases, Surveillance, and the Limits of Control. Here’s an excerpt from the episode:

Jake: Shall we get some veggies-

Tobias: You can say veggies.

Jake: -under the wire before it’s too late?

Tobias: Yeah, let’s do veggies.

Jake: Yeah. I try when I’m at my best to match up the veggies along with the guest. And as I mentioned before that Eric is to me one of the great bottom-up economists out there, coming through all these earnings transcripts and then offering these real time insights that happen before, they show up in official statistics. But it got me thinking about this other fellow who made predictions about what it would be like if we could measure everything.

And his name was Pierre Simon Laplace. and he was a French scholar and polymath, and he lived from 1749 to 1827. And he made important contributions to engineering, mathematics, statistics, physics, astronomy. He has his name on a few equations, which is always about as big as it gets, if you can get your name on something. One of them is in the field of partial differential equations. Don’t ask me to tell you what that is. But apparently, they use it for describing tidal flows. He was also the first to predict the existence of black holes. In 1814, he set out a mathematical system of inductive reasoning that’s based on probability, which today we would probably recognize as Bayesian statistics.

So, anyway, today, we’re going to be discussing this thought experiment that he had called Laplace’s demon. And it’s an exercise in causal determination. It goes as follows. Imagine that there’s a demon which knows the precise location and momentum of every atom in the universe, and their past and future values for any given time. It could then be calculated and predicted with perfect accuracy. By the way, Laplace never used the term demon in any of his writing. It was just added later for embellishment by others. But what if you could measure every single little absolute detail, and then could you accurately predict a future at that point? And this is, I think, an interesting thing to think about AI then. And as we’re trying to measure everything using AI, can AI do predictions that are better than–? Can it measure every little atom in the universe effectively?

Now, there are four arguments against Laplace’s demon in the physics realm. Number one is entropy. Laplace’s demon was basically vanquished with the discovery of irreversibility and entropy, the second law of thermodynamics. It’s based on the premise that there is reversibility. But many thermodynamic processes are irreversible. So, imagine like, you stir a cup of coffee with cream in it, you can’t get the cream to be unstirred ever at that point. There’s also the problem of quantum indeterminacy. This is basically like Heisenberg’s uncertainty principle. There’s properties like position and momentum, which you can’t know both at the same time. And then there’s chaos theory and the sensitivity to initial starting conditions. So, this is like the classic butterfly effect, where small changes produce these huge inaccuracies and predictions. This one’s a little bit less convincing, because the premise of Laplace’s demon is that you know every initial condition with perfect accuracy.

And then the last one is computational complexity. There was recently some work that had a proposed limit on the computational power of the universe. The limit is based on the maximum entropy of the universe using the speed of light, the minimum amount of time taken for information to move across the Planck length. And it’s been shown to be about 10 to 120 power bits. So, basically, it would require so much computational power to ever figure this out that the amount of time that’s eclipsed since the universe started is not enough to ever really do it.

But let’s think a little bit about Laplace’s demon and AI, because it’s an interesting topic right now. AI has the same problems of trying to predict outcomes based on data, like, what is known today, what is going to happen tomorrow? And in this context, it’s a useful metaphor. The AI is trying to predict future events. And if it had complete information, maybe it could, but it’s never going to have– There’s always fundamental limits to how much information you can actually have. And AI faces that same type of constraint. There’s also the quality and the quantity of data that the AI is trained on. So, in that way, that mirrors the impossibility of Laplace’s demon having complete knowledge.

And then there’s some ethical considerations, actually, just like Laplace’s demon raises questions about really, like, free will around judicial sentencing, loan approvals, recruitment. If you’re using AI for some of these things, there are these biases that can be baked into the AI based on the training data. And if an AI is that good at predicting individual human behavior, what does that imply then it is personal responsibility and free will?

And then the last one is kind of omniscience and control. The idea that if an entity could predict every future event, what kind of surveillance then would be required? Like, you almost start to run into minority report, the sci-fi movie with Tom Cruise predicting murders and things like that. So, what does that mean then for privacy, and autonomy and control of governments? So, hopefully, there’s a little bit of maybe thinking through some of Laplace’s demons and the implications of that from a physics context that we can then apply to AI to help us give us some kind of metaphorical framework to explore the capabilities and limitations of AI.

Tobias: What about natural intelligence? Does that suffer from those same limitations?

Jake: Yes, of course.

Tobias: More inconsistent.

Jake: True.

Eric: On the ethical side, this is timely. My daughter’s English class had an essay assignment, and they discovered half of the students used AI to write it.

Tobias: How did they discover?

Eric: And the teachers now are using AI to find AI.

Tobias: Ah.

Eric: And so, this is going to be an issue for the academic world, because half the class decided it was a lot easier [chuckles] to have AI write their essay and the other half did it. So, that’s just one example. But there’s going to be a lot of ethical issues.

Tobias: How do we pivot back to investing, JT? You’ll learn to plan.

You can find out more about the VALUE: After Hours Podcast here – VALUE: After Hours Podcast. You can also listen to the podcast on your favorite podcast platforms here:

Apple Podcasts