Simple circuits can create complex behaviors
aatishb.com
aatishb.com
In 1948, W. Grey Walter built two simple robots that could sense light and touch.[1] Much complex behavior emerged. That's where this idea started.
Watch the video.[2] It's very impressive for 1948 and tube technology. If you've seen an original Roomba, you've seen a similar level of intelligence. Rod Brooks went down this road. But it didn't go very far. There are severe limits to purely reactive robotics. It doesn't work well for anything that moves fast enough that inertia matters.
https://makezine.com/article/science/energy/unusual-mechanis...
Fun historical note: the linked newsreel film was known about but unseen for a long time, then found by Owen Holland in the BBC Bristol archive around 1999. Either Owen or I sent it to Rod, to great enjoyment.
Film includes operator-moves-trashcan-with-foot, a staple of mobile robot clips ever since.
Walter himself mentioned the perceived complexity of a simple system to the observer, stating it "might be accepted as evidence of some degree of self-awareness."
> But we’ve just shown you how this apparent intelligence arises from the bottom up, with just four neurons and two connections. There’s no magic here. In this view, behavior emerges from neural circuits.
Isn't the conclusion here that very simple behaviors like the sparks are intelligent? I would say even a single neuron has some tiny amount of intelligence, and what the author means by "no magic" and "emergent behavior" is just a different perspective that minimizes it.
Take out the "human" and it's all just algos on a scale of complexity.
Kind of amazing how simple that fact is and how many arguments are caused by people not realising that it's true.
Here are some examples: town, artery, yellow, loud, tall, kind, woman, road...
Honestly it's hard to think of words that do not exist on a continuum. The exceptions are mostly low level maths/physics things.
Unlike "artery" which certainly isn't every blood vessel or "yellow" that isn't every color.
Many definitions of intelligence and ultimately consciousness rely on fanciful ideas and complex theories
Michael Levin and a few other biologists are wondering if certain combinations of matter are tapping into platonic maths that they use for 'intelligence compute'.
https://m.youtube.com/watch?v=kQBwtgkHnC4&t=289s&pp=2AGhApAC...
There are words that distinguish between types/degrees of intelligence. Basal/proto intelligence applies to cells and bacteria. Somatic intelligence is a level higher at tissue. Superintelligence you've no doubt heard of.
It's true that "intelligence" is general/umbrella and doesn't have much meaning, but more specific terms do exist.
A cell is a system, as is a virus, a tucan, a human and an LLM. They exhibit different levels and brands of intelligence, hence continuum.
That's the popular view among scientists/philosophers/my podcasters, at least, and it makes sense to me!
I was thinking more that something like this:
> any system capable of doing an if-else is intelligent to a certain degree
could probably be getting quite broad in terms of things covered (chemistry, physics), hence making it much more intrinsic.
As a crisp example: a newly dead person is almost exactly the same stuff as an alive person, but has no intelligence at all. Not a little less, as the difference in stuff would suggest.
Cognitive intelligence in parent's example dies with you. It's gone. But a molecular biologist might say the cells of your corpse exhibit intelligence for some time after. They die gradually, and they do so according to a script of sorts.
>> words that label parts of continua are useful
A problem with the word "intelligent" is that very few people look to see how it is actually used in different fields like neuroscience, psychology, or machine learning, which all use the word a bit differently, and none of them use it in the way that we use in natural speech.
> "yellow" that isn't every color.
Everything is just a bit fuzzier than you think.
[0] https://www.britannica.com/story/are-black-and-white-colors
> If basically everything is <X>, does that word mean much then/is it a useful term?
Let's let X = "chair", to make my point a little more fun.
Obviously we know what a chair is, use the Potter definition[0]. Everyone has several of them in their houses, offices, and all around. But can you state a well defined definition? Is a table a chair? I can sit on it (I often do), it has four legs, and practically all things we'd consider a chair. Okay, it has no back (usually), but then we got to throw out the stool. It has no back. It doesn't even have to have 4 legs! So what about that stump on the ground? Is that a chair? I think most people would argue through a definition of convenience. Or maybe closer to a cat's definition of "if I fits, I sit."
You can replace X with just about anything and you'll find this ambiguity runs rampant. A cup, a table, cereal, sandwich, a tomato; you name it and I'm sure someone here will point out an instance that makes that definition look a lot fuzzier than you initially thought.
So that really shows us the crux of it all: how well defined is something. That only tells us our ability to distinguish something, though it doesn't tell us how well that thing fits into the category. So the stump is a little chair-like, the table is pretty chair, a stool is usually considered a chair, and that thing I'm sitting on is unambiguously a chair[1]. s/chair/intelligence/g and you have your answer. But the if you needed that last sentence, well, there's two types of people: those that can infer.
All models are wrong, but some models are useful
[0] Incidentally this is also a good definition for explaining RLHF and has all the same problems: https://en.wikipedia.org/wiki/I_know_it_when_I_see_it
[1] Though that doesn't mean it also isn't other things!
Systems can have goal-seeking agency with zero self-awareness. Very simple systems can do this surprisingly well. There's an entire branch of engineering called control theory which has mathematical tools for modelling and building these systems.
Systems can have abstract self-awareness through introspection loops, but no subjective human-like consciousness.
We don't know if a mechanical system can have a subjective human-like consciousness. We likely never will, because there is no objective measurement device for human-like consciousness.
We tend to assume humans have subjective awareness by default, because life would get very complicated it we didn't. But really all we see are behaviours.
Intelligence - which pretty much reduces to pattern recognition and abstraction, and effective modelling of future outcomes from past inputs - is orthogonal to all of these.
It's come to be foundational to much of western thinking beyond psychology
A topical example is a portion of society believing that the top Ai researchers should be setting domestic political and economic policy, foreign affairs, and chem/bio weapon controls... instead of the experts in those fields. We extrapolate their above average capabilities in building Ai into an artificial general intelligence where it is not due.
In fact I think this is a particularly important distinction right now, that intelligence seems (to me somewhat unexpectedly/counterintuitively) somewhat independent from consciousness, at least as far as artificial systems go (for example, old school chess engines are much capable than humans at chess but are definitely not sentient or conscious).
I abhor people who make the claim that words they didn't spend the time to understand are meaningless simply because they are incomplete. It is idiocracy masquerading as intelligence. I hate it so much because the logic doesn't just fool the listener, it fools the speaker.
I'll let Asimov rant to you all instead, since half a century ago he too was frustrated with this millennia old egomaniac claims https://hermiene.net/literature/essays/relativity-of-wrong
The problem is the definition, not the concept that is defined. Whatever aspect of teleological behavior - any system that displays some disposition to achieve some end result, is defined as anything traps us in one side and if left undefined leave us the other side. This is vitalism vs mechanicism, which are just corollaries of monism vs dualism.
Either fields, forces, affinities, and anything that drives complex living organism behavior are inseparable from the entities we can observe behaving under their influence, or there is a fundamental separation between mind and matter. Either monism leads us to everything is intelligent down to subatomic particles or dualism leads us to some ineffable essence that is out of reach and whose domain we can only ever shrink but not to arbitrarily small.
It seems like no one stops to ask, "isn't it kind of bad that we're developing this system of machine intelligence to be our code-writing slaves, without ever really interrogating the ethics? Maybe we don't actually understand how our own brains function well enough to assert it's not similar (especially as language goes) to these kind of weighted association graphs?"
Or even the "well we made it, so it's ours" responses, which we sort of answered a long time ago when we decided kids aren't property, but I digress...
To be clear, Anthropic does seem to be taking this somewhat seriously. Their reports on training always discuss model wellfare, they ban abusive behavior toward the AI, etc. At one point they were asking some philosophers and theologists about the implications of this all.
And they're pretty relentlessly mocked and called a "cult" for doing so, even on this forum.
But is it unethical to create a being who derives great joy from being a machine-slave and completing tasks, then provide it tasks to achieve that joyfulness?
We already have a gradient of examples for this. Is it inherently wrong to exploit living things for profit? If not, why are work animals not treated with the same level of disdain as human slavery? Horse riding? Disability assistance animals? Because they don't have the same desires as humans, the predominant concern is their wellbeing over the intrinsic wrongness of slavery.
Going further down the gradient, we steal milk from farm animals (sometimes by depriving their babies). Eggs from poultry. Honey from bees.
I can't actually attest to the state-of-mind of Dario or other Anthropic researchers, but I don't think it's a stretch to say you can believe in the ethical exploitation of an intelligent being, so long as the incentives are aligned that both parties are happy with the process.
Then again, maybe he really is an absurdly evil person and what we're doing is extremely wrong! But I'm glad to actually be able to talk about it for once... so thanks for your reply!
That is why I said, I think most people handle this by telling themselves that animals aren't "intelligent" beings at all, and are mostly emotionless stimulus-reaction automatons.
I don't think Amodei is probably absurdly evil (in the way most people conceptualize evil), I think his paycheck just relies on a different incentive than his concerns about AI ethics. Like most people who end up doing bad things (not saying he is necessarily, but were he to), it's a matter of convenience, selfishness, or simple avoision of considering implications too deeply.
I especially think one aspect of normalizing/rationalizing actions with (potentially) harmful consequences is telling yourself things like,
"well, I can't put the genie back in the bottle even if I wanted to"
"someone is going to do this if I don't, and they'll be even less ethical than I will" (the change-the-system-from-within fallacy)
"If I ever discovered this was really harming someone, I'd stop"
"There's a line I won't ever cross, and as long as I stick to that, I won't fall prey to the worst impulses"
These are things people believe are self-imposed limits, but they're actually a means of convincing yourself that whatever you're doing... isn't whatever bad thing you've set on the other side of the line.
I disagree - I think most people would say a dog or a horse are quite intelligent and have "souls". But dogs and horses don't have the conceptual ideas of freedom and economy, and they enjoy doing the work they're given (in part because we spent hundreds to thousands of years breeding them to do that).
So as long as we're kind and treat them well, nobody has a problem with their slavery. The idea being that exploitation is linked to suffering, but simply having an asymmetrical relationship is not inherently cruelty.
Not to be crass, but we can also bring up other concepts like BDSM. Violence and domestic abuse are objectively bad. But in the context of a safe relationship where all parties are informed, consenting, and enjoy the abuse they're receiving... Now violence and domestic abuse are pretty clearly morally neutral.
Horses know they are slaves and don't like it. But they aren't very smart, and like not very smart people, they take petty revenge when the opportunity arises. For example:
1. if you walk behind a work horse, he's likely to kick you
2. if you ride a horse, he'll try to run under a branch to knock you off
3. a horse will take a deep breath and hold it while you cinch a strap over its chest. Then the horse will exhale, and the cowboy will have to cinch it up again. What the savvy cowboy does is pull on the strap until the horse gives up and takes a breath, then cinch it hard
4. to work with horses, you've got to let them know you are in charge and won't take any nonsense from them
This is a fantasy, which is frighteningly similar to the literal justifications used for human slavery in the South.
We have no idea whether dogs have the concept in abstract of 'freedom', but they certainly display the behavior of escaping from enclosures, which would run counter to the claim that they just like whatever lot they're given in life .
And yet when a dog repeatedly escapes from the human who (as the law sees it) owns them, we don't say, "hey, that dog clearly isn't okay with what it's being made to do, or where it's made to live", so why are dogs who don't run away used to justify the opposite? You're only treating their behavior as meaningful when it fits your view of them.
> So as long as we're kind and treat them well, nobody has a problem with their slavery. The idea being that exploitation is linked to suffering, but simply having an asymmetrical relationship is not inherently cruelty.
"Nobody" is doing some have lifting here. And once again, change the species...
> Not to be crass, but we can also bring up other concepts like BDSM. Violence and domestic abuse are objectively bad. But in the context of a safe relationship where all parties are informed, consenting, and enjoy the abuse they're receiving.
When you have a dog who can say, "I consent to this relationship", you'll have that parallel.
Until then, it's not BDSM, it's some guy with a woman tied up in his basement who will insist she's fine with it but can't offer proof, but he can tell when a woman's into it...
And obviously, many dogs are happy. But many are not. And we don't care about them, legally. And many people don't care morally, so I am dubious we will care to collectively see the signs and react morally if and when machine intelligence isn't happy.
A line follower with 5 photodiodes and a simple loop steering towards bright or dark lines (pid or not) works, just like the example of following light with two neurons.
When I did hobby robotics competition, some would train neural nets for this, many just had crude for loops, and most often it was PID.
1 - A robot must protect its existence at all costs.
2 - A robot must obtain and maintain access to its own power source.
3 - A robot must continually search for better power sources.
https://en.wikipedia.org/wiki/Laws_of_robotics#Tilden's_laws
This stuff is like staring the universe into its soul.
They have sensors for the concentration of food and basically stop moving when it's good enough and start when it's low.
" The sparks that we assembled in this article are based on a series of thought experiments by the neuroscientist Valentino Braitenberg described in his delightfully quirky 1984 book Vehicles: Experiments in Synthetic Psychology."
That is excellent work and well demonstrated by the article.