So you really want to compare neurons to thermostats and these “governors”?

Some might disagree, but I think it is a much better comparison than all-or-nothing circuits that just represent a 1 or a 0 😊
(How Computationally Complex Is a Single Neuron?)

That’s a good point: neurons have a graded response and integrate a lot of signals, so that is already more sophisticated than a circuit

I guess that your perspective matters: for a hardware engineer, it really is about the nuts and bolts of how stuff works at the lowest levels 😅

but I suppose that from the point of view of software and mathematical models you can just treat neurons as numbers or a function

And that is why you would say that we need to look at analog computing and real time computing?

If you want to understand why it is “neuromorphic” computing, yes!
Otherwise it is just fancy statistics

Ooh, that might be a controversial take indeed! 😏

Ha! 😄 I thought you might like to make the article a bit thought-provoking

But then at a higher level of abstraction, when you put those artificial neurons together into a neural network …?

Well, sure, you get all the bells and whistles of machine learning and artificial intelligence, which is what your readers expect I guess

You seem dismissive about that? 😅

I don’t mean to, no, but after hearing about this 19th century engineer, Smee, from Julia I started to think differently about it

How so? What made you change your mind?

Well, just making a network out of stuff, doesn’t necessarily make it more “brainlike”, you know

Smee just tried to link all the concepts we explicitly know we have, but we more often than not don’t know what and how our brains are doing

So copying the structure of the brain without copying the bits and pieces it is made of doesn’t make it “neuromorphic”?

In a way, yes. We don’t have to copy everything, of course, all the biological stuff, and perhaps not even all the chemical stuff

Just the bits that are relevant to computation, even if we don’t really see it as computation in the classical sense?

Exactly! 😊

And perhaps we end up with artificial neural networks that aren’t terribly efficient or useful

So it depends a bit on whether you use neuromorphic computing to get quick results or to better understand the brain or computing?

Yes, and we might learn something unexpected from our brains, which are still the most efficient computers on the planet

That was great stuff! And totally unexpected for me

You might want to check up with Cho about some of the details regarding how neurons work though …

Excellent idea, I’ll do that right away 😊

I remember Julia reporting about that, but I’m not entirely clear on how that is connected to this and neuromorphic computing

OK, so think of old-fashioned scales: the bigger the weight, the more the needle moves, right? 😊

Sure, but is that a computation?

If you think about it: yes it is ! 😊
It is certainly information processing: it converts one magnitude into another. Or take a thermostat

Ah, yes, I know that it is the favorite example of some cognitive scientists for information processing 😄

There you go: temperature acts on a sensor which then does something, like turning on the heating.

How do you think the human body does that?

You do have a point there … I suppose that at a very basic level they work the same

That is indeed my point: you need to understand the basic level before you can appreciate what large nets of neurons can do and why that is and isn’t something special

Fine! It’s OK, 😊

I think we can get the readers on board with this, if indeed it is going somewhere more or less familiar or expected.

Sure, we all know where this is going to go, but if you don’t see why one neuron already computes, then a neural network will just seem like voodoo

So we have very simple systems in our body, which are similar to simple analog computers, and therefore we can speak of neuromorphic computing?

Yes, because those simple analog computers and circuits are the basic building blocks for the much more complex artificial neural networks

Alright then! It is good to make that connection very explicit …

Certainly. The main point is that most computations aren’t like doing math on a blackboard, but so to say implicit

So while you are moving your arm to hit the tennis ball and the sunflower is tracking the sun, they’re not “crunching numbers” as such, you mean?

Exactly, but it is still computation and it is part of a causal chain that makes stuff happen. There’s definitely rules being followed

I think an example would be very helpful here …

Ok, think about steam engines: they had these big metal balls mounted on a valve. The faster they turned, the more they closed the valve (Centrifugal governor)

in effect they continuously monitored and regulated the pressure. These were called “governors” and the example works exactly like the case of the thermostat

Oh, wow, that’s so cool! 🤭

I never realized what those rotating ball thingies were meant to do!

You can totally compare them to neurons too: as soon as a some impulse exceeds a threshold, it elicits a response (Artificial neuron)

Analog computing? I hadn’t really thought about that one …

Well, in the beginning, it was quite superior to classical computing, because it was so much faster

Faster? 🤨

Yep, it could make quite complex calculations: a century ago analog computers for fire control on battleships solved differential equations in real time 😊                                                            

The who with the what now?! 😄

I don’t mean to go as far back as ancient Greece, but at the beginning of the 20th century shooting the big guns on a ship accurately required a lot of calculations, in real-time (Gears of war: When mechanical analog computers ruled the waves)

I had no idea about this … 🤔

Just like you don’t fall over when you play tennis, because your neural system integrates information about your limbs and the balance organ in your inner ear

ships used gyroscopes and analogue computers to keep their guns on target despite rolling with the waves

Wow, that certainly would involve quite a lot of computation I guess

It does! Analog computing does this in a very different way from digital computing, but that doesn’t mean it’s not computing …

Ok, ok, fine! So these analog computers work more or less like the inner ear, therefore we might as well call it neuromorphic computing?

I’m sure some people might disagree, but basically, yes. You’d have to ask Cho about the specifics of how the brain does stuff like that though

I must say, this isn’t really what we were looking for originally …

I guess this kind of measuring, integrating information, and reacting in real-time is a lot less sexy than finding faces in photos and animating them …

Maybe it’s that, but I also suppose that most people wouldn’t consider this as computation at all

I’d say that someone skateboarding is solving a lot of differential equations in real-time! Or they’d crack their skulls … 😅

Just playing devil’s advocate here 😄

Just because we can describe the system using equations, doesn’t mean that the system is computing anything …?

Ah, yeah, I can see where that is coming from .. 😊                  sunflowers don’t do astronomy just by following the sun

That’s it, yes! So what are these navy devices and the inner ear doing exactly that makes them “computers”?

I had a whole conversation with Julia about analog vs digital, I guess this is the same kind of problem right here

Hi Sam, I wanted to thank you for helping us out with our article!😊

My pleasure, I enjoyed working with Julia and you 😄                        

Can I ask you another favor though?

Sure, happy to oblige 😊                                                                          

For our next issue we’re thinking of looking at Neuromorphic Engineering

Very interesting, yes! 😊                                                                           

There’s a lot of technical problems with the classical approach, so we thought about presenting some alternatives

Of course, but neuromorphic computing can mean a lot of different things

Yes, that’s why we need your help! I know it is not simply “doing things like the brain”

Indeed, in a very literal sense we still can’t do that                              

But Julia’s historical approach did convince me that we are getting closer and closer, right?

I would agree, yes, but we’re still a long way off                                   

OK, fair enough, but would you willing to collaborate on an article discussing the various approaches?

Certainly! Do you plan on going as far back as the 19th century like Julia did?

Not this time, thought I found it quite interesting what she uncovered

So mostly 20th century stuff?                                                                   

Yeah, we wanted to look at the parallel developments from the 1950s onwards

Side by side with the “official” history of AI? Interesting! I like it already

Yes, that was the plan, good to heave you on board!

Can I suggest Analog Computing as the first topic?