Can a machine think?

Section 1Self-driving too much Cars

One thing that’s probably pretty obvious to most people is that cars don’t think. If a car, left alone by itself, suddenly started moving forward, we wouldn’t ask what the car is thinking. We’d chalk it up to gravity and a driver who doesn’t know how to park.

But how far does this go? What happens if your car not only moves forward, but keeps going, exits the park, gets on the highway, drives to a port, finds its way to a container, gets itself shipped across an ocean to another country, etc., all without active monitoring or teleoperation?

At that point we’d be more interested in the internals of the car. What was going on in the processing unit, perhaps, what it was thinking?

Section 2Computer person

A common complaint that comes up when someone says their AI agent “thought” about something is to recoil and complain about anthropomorphizing.

“You cannot,” many say, “Attribute words that describe what humans are doing to Computers.”

I’d say that’s coming a little too late. For, perhaps, the past few decades, that’s all we have done. Constantly.

Consider the word “Computer” itself. That used to be a human. You would call a human a computer because they computed stuff. Now Computers are virtually exclusively silicon-based devices.

Or consider the various things we say that computers have and do. We say that computers have memory, that they can read and write. That they can die, sleep, be awake, that they have parent processes and child processes, etc.

So a more sophisticated argument wouldn’t be that anthropomorphism is bad, but that it can go too far. A computer can “die” because its processor no longer works, not because every cell in its body has self-destructed, but surely a computer cannot think. This goes too far. Thinking, many say, should be exclusively reserved for the human race.

But why?

Section 3The Tradition of Human Thought

People have been thinking about the concept of thought for a very long time.

The classical place to start is Aristotle, and his ideas of the rational soul.

For him, humans, and only humans, had logos, with which came the ability to speak and reason. To understand the abstract.

Descartes came many centuries later and concluded that you could separate humans from machines and animals because not only could humans say words, but they could arrange those words as a response to open-ended questions. They could use reason as a universal tool for all situations.

Many traditions have built on this; we even named our species after our ability to think. Homo sapiens, the wise man.

But these beliefs were merely asserted, and AI is bringing them into question today.

Many human traits have already been called into question before. Our most popular initial self-identity was that we were unique aspects of nature, specially created. It’s possible that much earlier, in hunter-gatherer societies, we saw ourselves as more deeply embedded in nature, but as civilization began to ramp up, we truly began to see ourselves as above everything else.

The theory of Evolution by natural selection put a big hole in this self-identity; it showed how we evolved from the very nature we tried to set ourselves apart from.

We’ve been scrambling ever since to find a stronger ground to stand on. If humans are just another kind of animal, then maybe thinking isn’t special? Sure, only humans have logos, only humans can think and reason, but maybe that’s just a unique trait within the animal kingdom. Maybe another kind of entity can do the same?

Section 4The Elevation of the Animals

The idea that only humans possess the ability to reason was first challenged by early evolutionary thinkers like Darwin, who argued that the difference between humans and animals was one of degree, not of kind.

In other words, animals can reason too! Just not as robustly as humans can.

This led to people hunting for evidence that animals could, in fact, think and reason. After a few false starts, they did start to find them.

For example, in 1917, experiments by the scientist Wolfgang Köhler with chimpanzees showed them figuring out how to get food placed out of reach by stacking boxes, not by blind trial and error, but by what appeared to be insight.

In the 1960s, Jane Goodall, who spent years studying these apes, found extensive tool use and complex social dynamics amongst them. Louis Leakey, a celebrated paleoanthropologist, commenting on the event, said we had to “redefine tool, redefine man, or accept chimpanzees as humans.”

This led to the creation of cognitive ethology in 1976, which studied animal minds extensively.

Fast forward to today, and we actually take animal cognition seriously. We actually reasonably accept that animals think. Not to the extent that humans do, but surely.

So now, thinking is actually no longer solely a human trait. Most reasonable concepts of thought accept that animals can think. We will all agree that animals don’t think as humans do, but they do.

So what does this mean? Human thought is a subset of thought.

Section 5Unbundling Human Thought.

If I ask you, What’s 12 + 13, you can give me an answer, but so can a calculator.

But we agree that you are thinking, and the calculator is not. So what’s the difference?

Is it because you’re aware of the process of generating the answer? If this is the case, then “Thinking” is merely conscious processing. It is algorithm execution bundled with experiential phenomena.

Under this view, we genuinely don’t know if AI systems can think, because we don’t know if they’re conscious. So the processing exists, but the consciousness might not.

But if thinking is just reasoning, if there is no consciousness requirement, then actually this would rule out the calculator, but not modern AI systems.

Now what people usually argue is that the very presence of consciousness itself grants you special reasoning powers.

The reason why we insist that humans are special because we consciously process thought is that we think this leads to unique results that another kind of system cannot achieve.

But what AI is increasingly showing us is that functionally, the results of thinking require only processing, not consciousness.

Thus we are forced to make a choice.

Is thinking conscious processing? If so, then thinking is not required for open-ended problem solving and creativity.

Is thinking merely processing? If so, then AIs can think, but only humans are conscious thinkers.

We are the modern Louis Leakey, and we must reinvent yet another aspect of our self-identity.

Section 6Can an AI model think?

I think that thinking should simply be open-ended, goal-directed, multi-step Information processing

Then, humans are simply conscious thinkers. This way, we aren’t surprised when AIs learn everything that we can do, and even surpass us.

Of course, this raises the question: what if AIs are actually conscious?

My response is that this doesn’t matter. The AIs are not intelligent because they are conscious. Their ability to generate art, music, solve mathematical puzzles, write, code, and operate robots is downstream from their processing ability. Their consciousness, if any, is likely orthogonal to this.

Such that, while we might quibble about whether they are conscious, we don’t need to debate whether they think, and thus, the functional implication of that: competence.

We must accept that the AIs will be able to simply do things! Things that only humans could do before, and they may be able to do this entirely unconsciously.

So that while we exist to be able to experience the universe, we can no longer claim a monopoly on advanced reason.