I was happily sitting on a bench in Black Swan Park, waiting for something to happen, when it hit me: artificial intelligence (AI) has a three-body problem. There are simply too many moving parts to predict what’s going to happen.
Whether AI saves or kills us all is, after all, dependent on how the computer code gets written. After that, our fate may be sealed, but it still depends on what some nerd had for breakfast.
If the wrong constraint is omitted, such as “thou shalt not kill”, you’ll get a real problem down the line. Allowing government regulators to define the constraints of AI will prove to be as useless as the Driver and Vehicle Licensing Agency (DVLA) on a good day.
We don’t know what’s going to happen yet – it’s just too dependent on too many different factors that are, themselves, up for grabs.
As Nigel Farage told me last week about net zero, the evolution of these vast trends tend to be defined by big events. A severe blackout might wake people up to the risks of intermittent power, for example.
The risk was always there, and the trend was always playing out. But an event defined the trend, nevertheless.
The first person killed by a machine using AI could tip the balance. Or AI might spot an asteroid headed towards central London that the Hubble Space Telescope missed, thanks to human error. Then we’d all forgive AI for any subsequent sins. Especially, if it came up with the solution to the asteroid problem (if you consider it to be a problem in the first place).
What’s the reference to the three-body problem?
I’m glad that you asked.
It’s not just a reference to the book, Three Body Problem, nor the upcoming Netflix series that is based on the book, but a very old physics problem. So, I better ask someone else to define it.
Professor of Mathematics, Richard Montgomery stated in Scientific American that:
One of the oldest quandaries in mathematics and physics is called the three-body problem—the question of how three bodies, mutually attracted by gravity, will move in the future if their current positions and velocities are known.
Isaac Newton first posed this problem, along with the simpler “two-body problem.” Later, in the case of three bodies, the question was found to be practically “unsolvable”—it is essentially impossible to find a formula to exactly predict their orbits.
My own explanation goes like this…
Imagine an empty universe. Now, put two moving objects in it – something like the earth and the sun, for example.
We know how these two objects will move as their mutual gravitational pull acts on each other. It’s predictable and maths can model it accurately.
If Elon Musk wants to travel to one of the two objects, he knows where it’ll be, when, and what speed it’ll be moving at.
But what happens if you add a third moving object (planet)? Now, you have three gravitational pulls that are interacting with each other.
Scientists haven’t figured out how to mathematically model the movement of three such objects in a way that predicts their paths accurately and indefinitely. The complexity of adding the third object’s pull is just too high.
It’s a bit like having three children, I think. (I’ll be able to let you know how and why in November.)
AI, in much of the same way as the three-body problem, is defined by complexity and too many unknowns that all interact with each other. There are just too many ways that things could go for us to truly predict AI’s future.
Take, for example, one of the predictions that I have made: in an age of AI, data analysis will become so powerful, convenient and effective that the availability, quality and access to data itself will become the key constraint, instead.
In other words, every school child will be able to ask AI to write their essay, but the quality of that essay will depend on the database in which the AI had to draw upon. Give the AI a database of Shakespeare and Byron to work with and the teacher will suspect something is up…
A database of past A+ essays would be disproportionately valuable to students, as many young, entrepreneurial businesspeople know. Such purveyors will be selling databases, not their essay writing skills, in the future of educational institutions.
So, data becomes the bottleneck in an AI world. Which implies that those companies that own vast amounts of data, and control access to it, will be the ones that profit from the AI boom.
Indeed, in some ways, companies already are doing so, with such companies dominating stock market returns in the US, this year.
But what if data is freed up by law? What if companies with too much control over data are broken up under anti-trust legislation?
What if the ability to anonymise data to protect privacy makes data freely available to everyone?
Or if data is simply stolen, copied and replicated too easily to be constrained with firewalls?
What if government declares itself to be the sole repository and source of all data?
What if people begin to protect their own data, or demand to be compensated for its use?
So many unknowns and moving parts…
One of the fun implications of the three-body problem, which the book is about, is that attempting to take advantage of opportunities in a three-body world is difficult.
If you can’t predict what’ll happen, and the potential for severe, unexpected impacts is quite high, how do you try and profit?
What if Elon Musk didn’t know where Mars would be, when nor what speed it was travelling at? Would people still buy a ticket?
In finance, the distinction between uncertainty and risk is used to tease this idea out. You can point to, define, quantify and analyse risk. Uncertainty is what emerges from behind the door and punches you in the stomach while you’re doing the maths to work out the risk.
A three-body world is defined by uncertainty, not risk. Because the complexity is so high, bizarre and seemingly random things keep happening. There are just too many possible outcomes for Isaac Newton, let alone us, to spot the patterns and make predictions.
The impact of the smartphone, the internet, cryptocurrencies and sub-prime mortgages were defined and calculable. We could conceive of and measure their impact on our lives. Nothing dramatically unexpected happened.
AI, however, has a three-body problem.
And yet, my friend and tech investing expert, Sam Volkering is doing his best to make order from the chaos. As many scientists have discovered, there are in fact three-body constellations that do result in predictable formulations of movement.
For example, a central object with two orbiting ones that are precisely opposite each other. That combination of three bodies results in a predictable and stable set of future movements. The two objects orbit the central one. The stars align, as it were.
And this, Sam claims, has occurred in a certain set of AI investments.
Until next time,
Nick Hubble
Editor, Fortune & Freedom
