The author makes a good point that it's important to define what "a good simulation" means.
On one extreme, we cannot even solve the underlying physics equations for single atoms beyond hydrogen, let alone molecules, let alone complex proteins, etc. etc. all the way up to cells and neuron clusters. So that level of "good" seems enormously far off.
On the other hand, there are lots of useful approximations to be made.
If it looks like a duck and quacks like a duck, is it a duck?
If it squidges like a nematode and squirms like a nematode, is it a [simulation of a] nematode?
(if it talks like a human and makes up answers like a human, is it a human? ;)
> If it looks like a duck and quacks like a duck, is it a duck?
ISTM that the answer is "in a way yes, in a way no".
Yes, in that we reasonably conclude something is a duck if it seems like a duck.
No, in that seeming like a duck is not a cause of its being a duck (rather, it's the other way round).
When we want to figure out what something is, we reason from effect to cause. We know this thing is a duck because it waddles, quacks, lays eggs, etc etc. We figure out everything in reality this way. We know what a thing is by means of its behavior.
But ontologically -- ie outside our minds -- the opposite is happening from how we reason. Something waddles, quacks & lays eggs because it is a duck. Our reason goes from effect (the duck's behavior) to cause (the duck), but reality goes in the other direction.
Our reasoning (unlike reality) can be mistaken. We might be mistaking the model of a duck or a robot-duck for a real duck. But it doesn't follow from this that a model duck or a robot-duck is a duck. It just means a different cause is producing [some of] the same effects. This is true no matter how realistic the robot-duck is.
So we may (may!) be able to theoretically simulate a nematode, though the difficulty level must be astronomical, but that doesn't mean we've thereby created a nematode. This seems to be the case for attempting to simulate anything.
At least this is my understanding, I could be mistaken somewhere.
I think this is also one possible answer to the famous 'zombie' question.
(aside: I'm not especially arguing with you; just thinking out loud in response to what you wrote)
> Something waddles, quacks & lays eggs because it is a duck.
Or: something does those things, period. We notice several such somethings doing similar things, and come up with an umbrella term for them, for our own convenience: "duck." I'm not sure how far different that is from "is a duck", but it feels like a nonzero amount.
I guess where I'm going is: our labels for things are different from the "is-ness" of those things. Really, duck A and duck B are distinct from each other in many ways, and to call them by one name is in itself a coarse approximation.
So if "duckness" is a label that is purely derived from our observations, and separate from the true nature of the thing that waddles and quacks, then does some other thing (the robot duck) which also produces the same observations, also win the label?
Luckily, I'm a solipsist, so I don't have to worry about other things actually existing. Phew.
I've never spoken to a self-declared solipsist before, though of course we all act like solipsists to a degree :-). Anyway, I will assume that solipsism is false for the rest of this post, that's another question.
It's amazing how many philosophical debates end up at the question of universals that you've just alluded to.
My own position, very briefly, is that when we predicate 'duck' (as in "this is a duck") of a given thing, we are describing reality, not just conveniently labeling some part of it in our own minds. If 'duck' is merely a label that we apply to something, then anything we predicate of 'duck' is merely something we predicate of our own mental categories. But this isn't so: the sentence 'ducks quack' refers to something real, not just our thoughts. But at the same time, the sentence is not referring to Duck A or Duck B, but to ducks in general. From this, it seems to follow that some general 'ducky-ness' must have a kind of existence (otherwise how could we predicate of it?), and that this 'ducky-ness' must be shared by everything that is a duck (otherwise, by what is it a duck?).
In the opposite scenario that you've described, all predication would be limited to our thoughts. Someone could say "ducks quack", and someone else could say "ducks never quack", and both would be right, because both would merely be describing their own thoughts. Obviously, all reason, science, possibility of communication, etc, is finished at this point :-)
Of course our labels can be wrong. Someone could mistake a swan for a duck. Also, there is infinite variation from duck to duck, so the 'ducky-ness' of each duck in no way tells us everything about that duck. Duck A and B are unique individuals. Also, the 'ducky-ness' only ever exists in a given duck; it's not like it has some independent ethereal existence.
Essentialism is the astrology of ontology.
> If it looks like a duck and quacks like a duck, is it a duck?
No, if it doesn't do everything else a duck does. You can have a robot dog, but you won't need to take it to the vet, feed it, sweep up it's hair, let it go outside to go potty, put up a warning sign for the mailman, or take it for a walk. You can have a simulated dog do all those things, but then how accurate will the biological functions be in trying to model it's physiology over time?
Will it give us insights into real dog psychology so we can better interact with our pets? Or does that need to happen with real dogs and real human researchers? Wildlife biologists aren't going to refer to simulated ducks to research their behavior in more depth. They'll go out and observe them, or bring them into the lab.
I suppose that's where a nematode is interesting — it's maybe juuuust simple enough that a real nematode on a plate of agar (as described in the article) might be able to be simulated well enough that we could actually make useful, and even long-term predictions about it based on a mere model.
Not to say I'm fully convinced, but I can see the appeal.
This is an interesting point really. At what level of duck-ness do we decide that it's acceptably close to a duck? I agree that taken ad-absurdum, just because something looks like a duck and quacks like a duck, it doesn't mean at all that it is a duck. I can enclose a raspberry pi in a fake duck and it will fulfil the above criteria, and perhaps from a distance it can be mistaken for a duck, but it has practically nothing to do with ducks. At the same time, it might be enough if our objective is to make some low cost garden decorations :)
What I'm trying to say is: as long as the simulation fulfils the objectives set out, it's useful, even if it is very far from the real thing.
Then the next question is: what are the objectives here?
> I can enclose a raspberry pi in a fake duck and it will fulfil the above criteria, and perhaps from a distance it can be mistaken for a duck, but it has practically nothing to do with ducks. At the same time, it might be enough if our objective is to make some low cost garden decorations :)
Agreed, it depends on what data you want out of the simulation. If you want to see how your dog will react to a duck, maybe it's good enough. If on the other hand you want to see how a duck will react to getting poked, well... your raspberry pi is worse than useless.
Assuming a dog only cares about how a duck sounds and not how it smells. We know that wouldn't work for other dogs. Which brings up something about simulating other animals. They're not human, and likely have sensory experiences that differ from our own. Perhaps a nematode worm is simple enough that we don't have to worry, but a dog or a duck are complex enough that we might leave that part out of the simulation. Or just not know how to fully simulate dog olfactory processing.
On this note, I'd like to recommend Phillip K. Dick's "Do Androids Dream of Electric Sheep?", which goes beyond the what was adapted in "Blade Runner" on many fronts, but particularly in regards to what it feels like to own a robotic animal (or a flesh one for that matter).
>Wildlife biologists aren't going to refer to simulated ducks to research their behavior in more depth.
I'm pretty sure behavior is simulated all the time in everything from migration to predator prey dynamics, to population dynamics, and so on. If we don't use simulations to understand all the little nuances and idiosyncrasies of behavior right now that's probably just because at present that's extremely difficult to model. But I suspect they absolutely would be used if such things were available. Of course, they would be treated as complementary to other forms of data, but wouldn't be disregarded outright.
That's an incredibly narrow slice of properties of ducks, nematodes (and humans).
Is there truly so little that makes up the soul of a duck? No mention of laying eggs? Caring for it's young? Viciously chasing children across the lawn of the local park? (I know that's usually the prevue of Geese, however I have seen ducks launch the occasional offensive against too curious little ones)
dude idk if you're trolling, but if not, the gp meant - if something exhibits the properties of a duck is it a duck.
GP's comment still stands: how many properties of the duck does a simulation need to exhibit before it can be considered an accurate simulation?
If it looks like a duck and quacks like a duck but isn't made of duck meat, you probably don't want to eat it.
If it looks like a duck and quacks like a duck but doesn't have feathers, you probably don't stuff its skin covering in your pillows.
Not really trolling, unfortunately.
GP, in their parenthesis, made the insinuation that if it, which I would take as a LLM, talks like a human and makes up answers, like a LLM is wont to do, like a human is it human? >(if it talks like a human and makes up answers like a human, is it a human? ;)
While I don't subscribe to the idea that humans have a soul, or some other dualist take, I do think that there is far more to a human than just our cognitive properties. So to convince me that something is human takes more than just listing to it talk to me or make things up during the discussion.
So, too, with a duck. Sure, if all I have to go on is hearing a quack then I would say yeah that's most probably a duck.
Just like if you told me a barn was red when we say just one side, I'd say it's probably red.
I know, I know, I am fun at parties.
It's like the idea that a civilized society would rather have a criminal go free than an innocent man be convicted. We are more than a brain, of course, but how confident are you? What's an acceptable level of risk for you potentially killing someone?
If you can simulate a behavior then that means you have a working understanding of how to get that behavior.
Not necessarily. You can simulate a high-level process independent from the low-level processes that make it up.
For example you can simulate traffic without simulating the inner workings of every car's engine, or even understanding how the engine works.
I might nitpick the "understanding" part. Lots of ML-type statistical models produce good results, but we can't very well explain how they work.
Or maybe by "working understanding" you mean "we have a black box that does the thing we wanted."
You can simulate behaviour with a lookup table, but that's not the same as understanding.