Hi John, I just finished discussing things with Sam                               

Great! So what did he have to say about this whole “laws of thought” thing? 😊

Well, there’s a bunch of different approaches  😊                               

Either you try programming them in from the start, or you make the AI learn them on its own                                  

Would either of those then yield the same “rules” humans use?

Probably not, there’s still far too many simplifications and shortcuts in the models

Generally people care more about the results than the procedure  

So as long as it works, it isn’t relevant whether or not an AI uses the same process a human would use

Exactly, moreover we don’t really know what rules humans use      

basically, that computers cannot understand the world like humans do

We don’t know how we get stuff done?

Perhaps I should also go talk to a psychologist or a logician, but: no, not really

We can’t properly generalize how we solve problems to how any problem can be solved in general

and we don’t know precisely enough how the rules are “programmed” in the brain

OK then, so AI just is a tool we use to solve problems, whether or not in a human way

and the whole “laws of thought” thing then doesn’t really matter, right?

I think we might have to split out how AI is used in different context.

The AI in an automated assembly line certainly doesn’t need to think and behave like a human

but research in computer science and in psychology does have to ambition to discover these laws

So mostly we’d need to split up theory and application

Perhaps that would be the best. And AI would certainly be on the theoretical side

Very good, the only thing we would still need is a good example or metaphor to make it clear to the reader

In the beginning people thought General AI was just around the corner

Like in the ‘50s and ’60s? 😊

Yes, Newell and Simon were convinced that they already had a thinking machine (Computer Simulation of Human Thinking)

That does sound a bit ambitious, for the kind of equipment they had … 😅

Of course hype like that brought out many critics, famously Hubert Dreyfus

Not sure I’ve heard of him …

He was a philosopher, not an engineer, and wrote a book on “What Computers can’t do”

basically, that computers cannot understand the world like humans do

Oh, I remember Sam saying something about the world not being a chessboard , is it like that?

I think it is exactly that: computers lack the background knowledge about how things work

if it isn’t explicitly there in the algorithm, in the program, it doesn’t really exist for the computer

But that’s not how humans think, right?

Yep, Dreyfus was a phenomenologist and looked at how humans experience and act in the world

and AI’s lack the specific human way of “being in the world” as Heidegger called it

So that’s why AI’s would have a lot of trouble with stuff a little kid can do effortlessly? 😊

Indeed, but this also suggests a solution: put the AI in the world like a human 😊

Ah, of course, instead of a disembodied algorithm, you give it a robot body, and senses, and make it learn, and …

Bingo! Humans aren’t disembodied minds, so if you want human intelligence … 😊

Good stuff! But we should definitely check in with Sam about this too.

Sure, will do, and perhaps you can talk to Manuel, about the interview he did him?

Yes, and see if he has more material.

Great! Then we’ll catch up later 😉                                                         

See you soon! 😊

Remember that you mentioned Newell, Shaw, and Simon to me?

Yes, did you find out how they would fit in all this?😊

Absolutely! 😊 Their research is a great fit for this topic                  

They tried to figure out the laws of thought to turn them into an algorithm

OK, so from psychology to programming? 😉

Exactly: humans use algorithms too, we just need to figure out which ones.

and then translate them into somethings a computer can use.

Yes. Basically they thought that thinking is just “problem solving”

and that there are some general rules that we use all the time, whether playing chess or doing math

I think Manuel mentioned this to me at some point: it’s all information processing

That is indeed the kind of approach: humans and computers use algorithms to solve problems, to process information

So the laws of thought that we discover through psychological research can be turned into a program

There’s just one problem …. it doesn’t really work all that well.

How do you mean?

The idea was: humans are intelligent, we make computers do what humans do, and then we have intelligent computers, AI.

However, there’s quite a difference between a computer algorithm and how humans solve problems.

The analogy works really well for chess and math, and is terrible for, like, more mundane problems: grocery shopping, navigating traffic, etc.

So the whole “laws of thought” stuff didn’t work out in the end?

Not really, and the link to logic was lost too: programmers simply went with what works, the most pragmatic option

So rules of thumb and heuristics and not deduction and statistics

Indeed, and then of course the way a computer solves a problem becomes very different from how a human does.

OK, so I guess we should include some of the setbacks and critics of AI as well

I think that would be a good idea, so we can make it clear how things are different now. 😉

In philosophy you had two camps: one group considered logic as the physics of thought

logic is the set of laws that describe how we think, just like the law of gravitation describes how things fall

Ok, that’s clear. The other group? 😊

The others thought logic is like an ethics of thought: what we should do if we want to think right

I do understand the difference between those, but I’m still not sure why this would matter so much … for AI 🤔

Well, if I got things right, then the point is that in one case we can’t disobey the laws of thought, it’s just how we are wired?

In the other case we choose to try to follow those laws, but we can fail. An AI could do the first, but not the second.

Ah, now it is becoming clearer what the issue is! An AI can only do what it’s told to do, not choose what rules to follow.

Yes, also the difference would be between logic as an ideal theory or as a human practice

Right, so is logic pre-given or do we develop it as seems fit to us?🤨

Exactly! That why everyone was accuring everyone else of being some kind of “psychologist”.

What so bad about being a psychologist? I don’t get it …

In German there’s a difference between someone who practices psychology and philosophers who apply psychology where it doesn’t fit

Somewhat like science and scientism?

Yes, that’s a good analogy: overextending the tools of one discipline to everything else

But why was everyone name-calling everyone then? Isn’t is clear to which camp they would belong?

Well, if logic is hard wired in the brain, then you would be a “psychologist” in one sense, 😊

but if logic is something humans build to fit their own needs then you would be a “psychologist” in another sense.

This is all very confusing, I hope we can simplify that for the article.

Yes, perhaps we should leave the whole “psychologist” thing out, and focus on the physics vs ethics

That would seem like a good idea, and make the link to AI more explicit!

I’ve actually got some good material on that from another group!

OK, so I did some digging and I’d like your feedback 😊                     

Great! Let’s have it!😊

So, I looked up the idea of “laws of thought” and whether we can put them in a computer

First issue: if we find laws of thought, what kind of laws are these?

What do you mean? 🤔

Are they laws like traffic regulations, laws you should follow, but sometimes don’t?

Or are they laws like the laws of nature, which you follow whether or not you want to?

Oh, I see. I remember something about prescriptive and descriptive laws …

Exactly! That’s it. So do the laws of thought tell us something about how the brain is wired,

how we should think if we want to think the right way, or are these ideal laws

Ideal laws? 🤨

Not sure what the right expression is here: these would be descriptive laws, but not like natural laws 🤔

Oh, you mean like math and logic, and so on

Yes, we don’t discover them by generalizing from facts, but they are formal, ideal

OK, so now we have different types of laws. We can explain that to the reader, but then what?

Well, there’s a bunch of psychologist, philosophers, and engineers that argued back and forth on this

and it seems to matter a lot to whether and how we can put them into a machine.

Right, so can we just build a copy of the brain and then it will automatically follow the same laws, or not?

That is indeed the type of question! But there is more! 😊