When we talk about computers, we tend to distinguish the machine and its programming: hardware and software

OK, yes, I think everyone is familiar with those terms                         

Great, but I’d like to point out how wrong that is, in some respects …

Where do we draw the line between those? Are they really that different? Doesn’t the design of the one determine the design of the other?

Whoa, those are tricky questions! 😅
Remarkably similar to the kind we ask about humans; is that what you’re after?

In a sense. We are so used to the hardware/software distinction, that we forget that somewhere they overlap and interact

Can you give me some examples? Otherwise, I fear our readers won’t all get it …😅

Sure: flip-flops are at the same time a hardware device and a software abstraction, a circuit and a one or zero

Right, so there is a level where the distinction is arbitrary: it is both a thing and information

Exactly! 😉 But then of course at other levels things are much more clear cut

But what did you mean with “wetware” then?                                       

Ah, yes, well as far as I know, the brain is like that all over: it is tremendously difficult to keep the hardware and software apart

Oooh, now I get what you mean: a classical computer architecture is completely different

Right, but if the brain is an information processor, you can try to just copy the informationally relevant bits

So like Smee you can capture the same informationally relevant relations in a different hardware?

Up to a point, of course. We can try to copy some of the structure of the brain, but only in software really

What do you mean? 🤔                                                                             

Well, most “artificial neural networks” (ANN) actually are software simulations that run on ordinary beige box computers

Aha, so as you said: a lot of simplification and abstraction 😊                

Correct! We still don’t know exactly how the brain does all its information processing or how to copy that efficiently

and then there’s the whole digital/analog problem …

I’m not sure it would qualify as an artificial neural network nowadays, but there was this doctor and metallurgist, working on electro-biology: Alfred Smee

He thought that all thought was based on electrical activity in the brain which we could use as a model for a machine

That’s amazing! 😁 I didn’t know that … 

It seems he conjectured you’d need an artificial neural network the size of London to model a whole human brain

Ha! 😅 He was probably right, early computers where absolutely massive, and supercomputers still are

With the technology of his time this could probably never have been built, he even guessed that the machine would self-destruct by using it …

It really is surprising how old some ideas are, and has to wait for technology to catch up so we can realize them!

But a lot of there machines were never built: Babbage’s analytical engine, for instance

Or Turing’s Machine, indeed, the idea was enough to prove that something was or wasn’t possible

That’s very interesting!  🤔                                                                       

Well, I guess that you could compare and contrast Smee and Babbage in that respect

Babbage did not consider his machine as a model for the mind, but developed his ideas from the technology that was available

Smee tried to design a machine based on what they then understood of the brain, but the technology wasn’t there yet

But now it is?                                                                                               

Not exactly, as I mentioned to Manuel, artificial neural netword can’t really compete with our organic brains

So the idea of copying the brain still can’t be realized?                       

With some extreme simplifications, maybe a little bit …

Tell me more! 😊                                                                                        

I think I need to make a few distinctions first, like between hardware, software, and … wetware

Wetware!? 🤔                                                                                              

Hi Sam, I wondered if I could ask you some questions about the history of AI?

Hi Julia, sure! Is this for an assignment?

No, for an article I am co-writing with John                                           

Oh, interesting! Ask away!

We were wondering about some of the things you mentioned in your interview with Manuel

Yeah, that did remain somewhat superficial

We’re doing a more in-depth article now, so you have room to explain

That’s great! So what exactly did you want to know more about?

We were talking about Dreyfus’ critique of AI and you mentioned embedding robots in the world to make them learn

Right! I think I mentioned that at the end, when talking about AGI: artificial general intelligence

That would be human-level intelligence, right?                                     

Indeed. Pre-programming all that didn’t work out, not even when trying algorithms inspired by human problem solving

Yes, I read about Newell and Simon, this was Dreyfus’ main target, no?

Correct, but there are other approaches as well, such as making a computer learn, instead of programming it

OK, so machine learning, using artificial neural networks?                

Those networks are only very roughly similar to the structure of a human brain and not really meant as a model of how humans do it

But they would be a serious alternative for the older approach?      

In a way, but we are still very far from AGI, so far we don’t even know how far …

When did this new approach really get off the ground? I’ve found texts referencing artificial neural networks back in the 19th century …

What!? That can’t possibly be right!

Hi Sam, can I bother you for a minute?

Hi Julia, sure thing! What’s up? 😊                                                           

I’m collaborating with John on a new article on the history of AI

Oh, wow, there’s lots of interesting stuff in there! How can I help?😊                               

My question would be: how much does your body matter to your mind?

Wow! You’d need a philosopher for that, not an engineer! 😊          

Let me rephrase that: how much does the hardware determine the software? 😉

Oh, right! Yeah, if you put it that way … Quite a bit actually!😊         

I don’t know about the mind in general, and the “laws of thought”,   

but a computer is definitely conditioned by its basic architecture    

say, whether it can process things only serially, one after another, or in parallel, etc.

I guess the constraints can be found mostly at the lowest levels?

Yes, where hardware and software meet, in a sense, where the 1 and 0 of math become electromagnetic charges and streams

OK, but at higher levels you can do the same algorithm on different machines?

Absolutely! Most computers implement the same kind of high level, abstract programming languages

And whatever we call “AI” mostly happens there, so isn’t constrained much by the hardware?

Well, most of the differences you’ll see between human intelligence and AI actually does boil down to implementation

how the various bits and pieces communicate, the size and speed of memory, etc.

last question: why do we call it “artificial” if it just works like human intelligence? 😅

Perhaps we shouldn’t! But I guess it mostly means “made with tools” instead of “fake”. Some prefer to speak of “machine intelligence”

OK, this has been most helpful! 😊

Happy to hear it! 😊Let me know how it turns out!