“A machine will never be as intelligent as a person.” To everyone who claims this, I want to answer: no, it’s a person who will never be as intelligent as a machine.
Because it all comes down to the definition of intelligence we give. And you don’t ask philosophers, literati or the random guy who showed up at a conference for definitions of intelligence. The most acceptable definitions that exist come from two places.
First channel: psychometrics
The first is psychometrics. Psychology, by measuring behavior, tries to estimate psychological variables such as intelligence, and there are batteries of tests built to measure intelligence in people.
Notice that now that we have these spectacular artificial intelligence models, we measure their abilities with psychometrics too. The famous “evaluation” everyone talks about in AI is nothing more than psychometrics applied to machines. You give them tests: multiple choice, summaries, answering questions, using tools.
Why tests? Because you can’t test artificial intelligence the way you tested traditional software, with unit tests and a test suite that hammers the main program and expects 100%. You can forget about 100%. So psychometrics is the first channel for measuring intelligence, applied to both people and machines.
Second channel: physics
The second channel - or better, the formal definition of intelligence - is found in physics. And in which physics do you find the most accurate definitions? You find them among the madmen, the scientists nobody listens to because they speak a complicated language, while people want easy answers - typically easy and wrong.
These definitions revolve around entities in relation to an environment, where they have to maximize some kind of score: the famous reinforcement, or value, as some call it. There are connections with formal economics here too. These agents face a set of environments they have to interact with and learn from.
The groundwork for this comes from the formalization of reinforcement learning: Sutton and all of the psychology that became more mathematical over time, partly thanks to artificial intelligence. And from information theory: in this continuous stream of information, including the actions I perform on the world and their effects, how do I figure out the patterns that govern the world, and the ones that govern my interaction with the world? Those are two problems of a different order that interact with each other, and you define them in terms of information.
Information, in turn, is based on probability. Multidimensional streams of data: the ones that enter the machine, but also the ones that enter your brain through the retina, the cochlea, touch and everything else.
The brain as a subsymbolic computer
The brain is a device that continuously monitors what’s in the environment, including monitoring yourself: proprioception, neurons that constantly track the state of your body, the vestibular system, balance. It’s a system that analyzes, compresses, evaluates and makes decisions based on incoming information, until that information is routed to the motor cortex, which sends signals down the brainstem, through axons, all the way to the muscle fibers - and you move.
The brain as a computer is an analogy from the ’80s that no longer holds if you mean a symbolic computer. But it still holds perfectly well as a subsymbolic computer. There are no little symbols; there are distributed representations. And these distributed representations are just like the ones in today’s artificial intelligence models.
Intelligence is not your grandfather’s tomato
Intelligence has nothing to do with “when I was little my grandfather took me to his vegetable garden and let me taste a tomato, it’s a beautiful memory”. That’s your subjective experience, your memory, and it has value in itself. There’s no need to measure it against intelligence.
Intelligence is something more abstract. Intelligence is like force, mass and acceleration, intelligence is like gravity: it’s something you measure. So if you’re sitting there arguing whether a machine is intelligent or not, you’ve already missed the point entirely. And if you then measure it against your personal memories and how much you care about your emotional experience - nobody is taking that away from you, but it has nothing to do with intelligence.
Compression and prediction
It’s about information. These intelligent devices and organisms have the property of being compressors. From an enormous, continuous stream of data, these machines and these brains simply ignore what is redundant and pay attention to what deserves it. And what deserves attention? What is less probable, less predictable.
From this definition of information and probability comes the analogy between brain and machine, not only as information compressors but also as predictors. The organism continuously tries to model the environment and its interaction with the environment. Why? Because it’s adaptive, because it was selected biologically as a mechanism. In the case of machines, we put it in there ourselves, with backpropagation.
The organism tries to understand how the world moves in order to interact, survive and reproduce in it. In machines the same thing happens, not for reproduction but to obey a law: an error that has to be minimized. So in my view it’s all much more down-to-earth, more mechanical. And it’s also more beautiful than many people describe it.
All this compressing and predicting - there’s also predictive coding, and there are spectacular neural models that haven’t reached the general public yet. At the intersection of biology and physics people are doing things that blow your mind.
Beauty is in physics
I wanted to say this and then I’ll wrap up. Treating intelligence as a physical property, as a physical phenomenon, does not diminish intelligence.
It’s like the guy who tells you “the world is awful, but I believe in the Lord because the Lord gives us grace”. You don’t need the Lord. You need to look at what there is in a way that lets you appreciate it. What is, is - and you’re the one who decides how to look at it.
Saying “no, it’s just a machine, the brain is just the brain, the mind is something else”? No, my friend. Mind and brain are the same thing, and the machine is intelligent. Machine and brain are the same thing.
If you don’t see beauty in that, if you feel diminished by admitting that we are physical organisms governed by physics and natural selection, that says more about the fact that you find life disgusting. If you can’t see the beauty and you have to make things up - bearded men controlling and judging us from the sky, a world created in seven days, virgins giving birth, and so on - then what you need is to understand that beauty lies in physics, in the world, in biology.
Look at your own body at the level of organs, tissues, cells, the organelles inside the cell. The plethora of life forms elbowing for room. Geology, and the timescales of all of this. If you don’t see beauty and don’t find spirituality in this, in my opinion you’re looking in the wrong place.
