Dennis Hackethal’s Blog
My blog about philosophy, coding, and anything else that interests me.
Tweets
An archive of my tweets and retweets through . They may be formatted slightly differently than on Twitter. API access has since gotten prohibitively expensive – I don't know whether or when I'll be able to update this archive.
But in case I will, you can subscribe via RSS – without a Twitter account. Rationale
@joe_shipman @DavidDeutschOxf @ESYudkowsky
I recorded an impromptu episode of my podcast to go into more detail about this:
I recorded an impromptu episode about the question of "value alignment" in artificial intelligence research:
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Not really, no. Here's a hard-to-vary explanation of explanation: youtube.com/watch?v=DHR6ro…
The mistake Popper identified as the who-should-rule-question permeates inquiries into AGI and is behind the pseudo problem of “value alignment.” An example of how critical philosophical knowledge is in this field and how urgently we need more of it. twitter.com/joe_shipman/st…
It’s undesirable because when taken seriously it leads to enslavement and murder of AGI.
2/2
One can’t know, nor does one need to. The very question seeks an authoritative answer (see the only superficially unrelated economist.com/democracy-in-a…; disable JavaScript to circumvent the paywall). And mistakes are what enables progress in the first place.
1/
@jchalupa_ @ReachChristofer @adilzeshan @RealtimeAI @dela3499 @ToKTeacher @Soph8B
Video game "AIs" work according to the same mechanisms as animals even when given only incomplete information about their environment. (In)complete access to model of environment is not the defining factor. The ability to create knowledge is.
@jchalupa_ @ReachChristofer @adilzeshan @RealtimeAI @dela3499 @ToKTeacher @Soph8B
It's possible, but not part of our best explanation of creativity. FWIW creativity does not only apply to humans but all people (humans, AGIs, intelligent ETs, etc).
1/
@jchalupa_ @ReachChristofer @adilzeshan @RealtimeAI @dela3499 @ToKTeacher @Soph8B
Moment to moment response to environmental factors != creativity.
Eg video game "AIs" respond to the player's movements and decisions moment to moment. Since we build them, we know they do not use creativity for that. It all follows predefined algorithms that have reach.
@NathanPMYoung @IamtheWay13 @DavidDeutschOxf @ToKTeacher
I'm happy to immediately drop the term "AGI" and replace it with whatever you find more appropriate as long as we know we are talking about the same thing.
@NathanPMYoung @IamtheWay13 @DavidDeutschOxf @ToKTeacher
I explain my reasoning for the sharp distinction between narrow AI and AGI and why progress in one is not progress in the other here: soundcloud.com/doexplain/2-wh…
@NathanPMYoung @IamtheWay13 @DavidDeutschOxf @ToKTeacher
Just because the vast majority of researchers use a word as it is commonly used does not mean they are successful.
I'm on board saying that narrow AI researchers achieve great successes in the domain of narrow AI. Sadly, that says nothing about their contribution to AGI (nil).
@NathanPMYoung @IamtheWay13 @DavidDeutschOxf @ToKTeacher
That’s the AI researchers’ confusion, not Deutsch’s. :)
They don’t understand universality. That’s why AI researchers will not make progress in AGI.
Entirely different field with a misleadingly similar name.
@NathanPMYoung @IamtheWay13 @DavidDeutschOxf @ToKTeacher
Not entirely sure what you mean by “AI is a superset of AGI” but in terms of problem solving it’s the opposite because an AGI could do everything all narrow AI programs could do (and then some).
AI is like all other programs; execution of predefined tasks, no creation of new knowledge to solve novel problems.
AGI = universal problem solver; creates knowledge. Opposite of not creating knowledge.
BTW, perfect value alignment is neither feasible nor desirable.
@IamtheWay13 @NathanPMYoung @SmashAGrape @DavidDeutschOxf @ToKTeacher @SamHarrisOrg
Yeah humans = AGIs (both are people) because of universality.
@RealtimeAI @pmathies @ReachChristofer @dela3499 @ToKTeacher @Soph8B
I don't understand the question. Please elaborate.
@RealtimeAI @pmathies @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Both frogs and dogs may well have universal computers for brains. So it's not the brains. It's the software installed on those brains that determines whether you can train that brain.
@ReachChristofer @EAMagnusson @RealtimeAI @dela3499 @ToKTeacher @Soph8B
Exactly. Fallibilism must not lead to paralysis during decision making. Otherwise we run the risk of turning fallibilism into a strange version of the precautionary principle.
We should act on our best explanations without hesitation. Anything else is like Pascal's wager.
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
IIRC DNA is a general purpose storage medium, so yes, I think it could encode a nuclear spaceship. (That’s not to say such a thing would evolve biologically.)
@DoqxaScott @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
No. They’re on a molecular basis?
@DoqxaScott @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
I have been thinking about how cool it would be to build self-replicating machines. I don’t think it’s been done. Might tell us a thing or two about evolution.
@DoqxaScott @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Nuclear von Neumann probes? :)
@DoqxaScott @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Sure but none of the ideas built by creativity did. Those are created at runtime.
@RealtimeAI @dela3499 @ReachChristofer @ToKTeacher @Soph8B
Examples of that: learning how to communicate with people, or building complex structures that are completely different from anything the species has built before, or every individual creative animal being entirely unique in character.
@RealtimeAI @dela3499 @ReachChristofer @ToKTeacher @Soph8B
... and therefore would not have evolved biologically, and therefore could not be encoded in the animal's genes, and therefore must have been created by the animal itself, at runtime (i.e. during its lifetime).
3/
@RealtimeAI @dela3499 @ReachChristofer @ToKTeacher @Soph8B
Here's what I'd consider evidence for creativity in an animal: knowledge for which there would have been no genetic precursors in its ancestors, or which would not have given the ancestors' genes a better ability to spread...
2/
@RealtimeAI @dela3499 @ReachChristofer @ToKTeacher @Soph8B
Combination of inborn ideas about what things to avoid and shape recognition algorithm?
1/
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Also note how the dog is not looking at the stick he's pulling out, or even at the tower, but at its owner. It has no idea what it's doing or why. It wants to please his owner.
2/2
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
I don't think it guessed. It just had to update parameters and make associations. Dogs know from birth what sticks are, how to put things in their mouths, and how to seek reward and avoid punishment.
Put these things together and you get enough reach to play Jenga.
1/
@univ_explainer @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
I made critapp.com for that reason - I think it has better tools for discussion than Twitter. If you'd like to try it out, shoot an email to contact@critapp.com and I'll make an account for you. :)
@univ_explainer @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
... and it makes sense for dogs to have genetically evolved such criteria for success because people have been selectively breeding them.
2/2
@univ_explainer @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Ok, a few things (Twitter’s character limit is terrible):
Narrow AI can already do what the dog does there.
AGI can only be achieved in a jump, so we must skip to it somehow.
What the dog counts as success is genetically given (praise by owner)...
1/
@univ_explainer @RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
To be clear, you mean an AGI?
The dog trick is cool, but can be accommodated by a reinforcement algorithm with enough reach, which dogs seem to have. (Note how she praises the dog. Also note how the dog watches her closely, presumably for facial cues indicating success.)
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
While it's impressive that animals can do this, it does not require any creativity on their part - that's why I don't consider it learning.
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
Yes, though I'd be careful with the word "learn" there - the animal may have been an inborn reinforcement "learning" algorithm, which, coupled with inborn shape recognition algorithms, updates parameters to categorize something as "not dangerous" after several interactions.
@univ_explainer @ReachChristofer @RealtimeAI @dela3499 @ToKTeacher @Soph8B
Why did they have almost no chance to explain anything?
In any case, note that "universal explainer" also signifies an ability, not a guarantee or even chance of success.
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
Yes, I don't disagree that the replication strategies of memes differ from ideas that never become memes. But... so what? :)
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
I claim that they do replicate within minds, just not necessarily across people :)
@univ_explainer @ReachChristofer @RealtimeAI @dela3499 @ToKTeacher @Soph8B
All babies make that jump to universality long before they learn to speak. This universality lies within people - it is not induced or awarded by outside factors such as technology (let alone the fact that one needs creativity to make technology in the first place).
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
Eg "selecting" a preference between fight or flight can be done according to inborn algorithms that do not involve creativity (variation and selection in the evolutionary sense).
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
Ah, got it. I'll claim that's a stretch of the phrase "variation and selection" as it strays a bit from evolution because it doesn't refer to variation and selection of replicators.
@dela3499 @RealtimeAI @ReachChristofer @ToKTeacher @Soph8B
Can you give an example of variation and selection in animal brains?
@RealtimeAI @dela3499 @ReachChristofer @ToKTeacher @Soph8B
Why bring recursion into this?
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Well, people are universal explainers, so even if other organisms have some limited creativity, that marks a pretty sharp distinction. They would all have an infinitesimal repertoire compared to people.
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Could all of those things not be encoded genetically? At the risk of sounding like a broken record, the presence of knowledge, including that which changes behavior, is not evidence of creativity. That knowledge may have emerged from biological evolution.
@ReachChristofer @RealtimeAI @dela3499 @ToKTeacher @Soph8B
Yes, people do eventually die if they don't solve problems. But I don't think the absence of creative thought = death. Eg if you run on autopilot for a few minutes, that won't kill you.
Of course, the underlying message rings true: problem avoidance eventually kills people.
@RealtimeAI @ReachChristofer @dela3499 @ToKTeacher @Soph8B
Here's a criticism of one of the ideas in BoI :)
The artificial intelligence research community is in bad shape...
Actually, I take it back - progress is the result of that. So still need a word for it. :)
@PopperPlay @ReachChristofer @DavidDeutschOxf
I’m about to publish something on this, stay tuned. :)
@PopperPlay @ReachChristofer @DavidDeutschOxf
Yeah those with side effects transform minds, and, if they get a mind to act, the world.
The motivation for treating ideas as functions is to solve the problem of how to encode ideas in a computer program.
Ideas need not return the same output for same input, nor do functions.
@PopperPlay @ReachChristofer @DavidDeutschOxf
PS: The above is more of an answer to your question “Is there a way we can show that all possible conjecturing and problem solving descends from a single algorithm?” from popperplay.com/problem/Qb6ij0…
@PopperPlay @ReachChristofer @DavidDeutschOxf
So the explanatory universality of people is powered by the computational universality of functions. Those two universalities are deeply intertwined.
4/4
@PopperPlay @ReachChristofer @DavidDeutschOxf
Since Lambda Calculus is computationally universal, all ideas in the mind can be expressed as functions, and so the above is the same as saying that it’s a functional replicator in a mind that explores the space of all possible functions.
3/
@PopperPlay @ReachChristofer @DavidDeutschOxf
A more elaborate one: ideas replicate imperfectly within a creative mind and thereby inadvertently explore the space of all possible ideas. This is how sometimes ideas evolve in a mind that happen to solve a problem/explain something.
2/
@PopperPlay @ReachChristofer @DavidDeutschOxf
Some quick arguments for the explanatory universality of creative minds:
1) What couldn’t one guess? (nothing , it seems)
2) Humans are so far off the mark (we have built space shuttles age cured diseases etc) that it just makes sense to think they are universal.
1/
@ReachChristofer @ks445599 @PopperPlay @DoqxaScott @DavidDeutschOxf
Yeah, IIRC, there is no computation a quantum computer can perform that a classical UTM couldn’t. It’s just that some of those computations run intractably slowly on UTMs compared to quantum computers.
@DoqxaScott @RealtimeAI @ReachChristofer
I once saw a video of a monkey swiping pictures on an iPhone. Cool, but not evidence of creativity.
@DoqxaScott @RealtimeAI @ReachChristofer
What do crows/monkeys/other animals do that couldn't be explained exclusively in terms of biologically evolved adaptations? Do you have a video showcasing such behavior, or maybe an article explaining it?
@DoqxaScott @ks445599 @RealtimeAI @ReachChristofer
I think conjectures are the result of imperfect replication of ideas in the mind.
@ks445599 @DoqxaScott @RealtimeAI @ReachChristofer
I'd leave out any considerations involving pattern matching because they are too close to empiricism. It's a mistake I have made in the past myself. Empiricism is tempting so it does sneak back into mind here and there if one isn't careful.
@DoqxaScott @RealtimeAI @ReachChristofer
I mean, maybe we can consider the result of any algorithm running in the mind a conjecture, but thinking of creativity as pattern matching is a dangerous path into empiricism, which is really creativity-denial.
@DoqxaScott @RealtimeAI @ReachChristofer
That's why I wrote "intelligence/consciousness" a number of times, because if you have one you automatically have the other.
@DoqxaScott @RealtimeAI @ReachChristofer
I didn't mean to suggest that intelligence and consciousness are the same thing.
I think intelligence = creativity. Same thing just different words. And I think consciousness, among other things, is epi-creative, meaning it arises from creativity.
@DoqxaScott @RealtimeAI @ReachChristofer
Ah - you’re saying the result of, say, a pattern matching algorithm is a conjecture?
@DoqxaScott @RealtimeAI @ReachChristofer
There may be value in it, idk, I’m just pointing out that one is an error and the other a result of one. They are different things. So I don’t think the comparison applies.
@DoqxaScott @RealtimeAI @ReachChristofer
Yes, with the proviso that no (or only little in the case of inborn ideas) knowledge of how to solve particular problems is given and needs to be evolved at runtime instead.
@DoqxaScott @RealtimeAI @ReachChristofer
Sounds like empiricism. Not sure what you’re trying to say. Please elaborate?
@DoqxaScott @RealtimeAI @ReachChristofer
Yes. Sometimes adaptations have enough reach to incorporate use of new tools etc.
Knowledge of any kind, no matter how sophisticated, is not evidence of intelligence.
@DoqxaScott @RealtimeAI @ReachChristofer
Well, a conjecture is the result of an erroneous replication in a mind, so I wouldn’t compare it to transcription errors per se.
But yes there are many differences between biological evolution and what I call functional evolution in a mind.
@DoqxaScott @RealtimeAI @ReachChristofer
Don’t see why those couldn’t have been genetically programmed?
@DoqxaScott @RealtimeAI @ReachChristofer
Not pedantic, good point. Errors in transcription do indeed happen somewhere in plant. But no evolution within plant. Hence not intelligent.
@DoqxaScott @RealtimeAI @ReachChristofer
I think conjectures and refutations are components of intelligence regardless of whether they are made consciously.
@DoqxaScott @RealtimeAI @ReachChristofer
There is no variation and selection happening within plants. They happen across plants.
And yes I think only people are intelligent.
@DoqxaScott @RealtimeAI @ReachChristofer
That's not creating knowledge. It's just updating some parameters and it all happens to genetically given instructions.
@DoqxaScott @RealtimeAI @ReachChristofer
Genes are within plants, sure, but they are not intelligent/conscious because new knowledge is not created from within them.
@ks445599 @RealtimeAI
It’s a way to avoid explaining that by saying that consciousness is somehow already present everywhere.
Similar to how Lamarckism, empiricism etc state knowledge is already present somehow.
2/2
@ks445599 @RealtimeAI
Agreee. Also note that panpsychism is not an explanation. It’s just a statement: everything is conscious to some degree. That’s too easy. Doesn’t explain what consciousness or at least what gives rise to it.
1/
@RealtimeAI @ReachChristofer
Both would only be intelligent/conscious if knowledge originated from within them.
3/3
@RealtimeAI @ReachChristofer
In the case of plants, the knowledge originated in biological evolution and the plant just inherited it through genes.
In the case of a Roomba, the knowledge originated in a the minds of programmers and the Roomba “inherited” it through programmatic instructions.
2/
@RealtimeAI @ReachChristofer
Cool :)
I don’t think a Roomba is intelligent/conscious. Both Roombas and plants contain knowledge, no doubt. But to determine whether they are intelligent, one needs to determine the origin of that knowledge.
1/
@RealtimeAI
Okay, so does a Roomba. Is a Roomba intelligent/conscious?
I haven't thought enough about whether children should be given more lenience than older people. I have a hunch that yes, they should, but I wasn't really commenting on that.
Whichever way one argues, the argument should invoke knowledge, or a lack thereof - not brains.
@RealtimeAI
Because they move? What about a speck of dust flying through the air?
Flawed, reductionist reasoning. All people, including children, are universal explainers. That their brains develop until 25 doesn’t change that.
Why are people so set on ignoring software?
RT @ReachChristofer:
To celebrate the tenth episode of Do Explain, @DavidDeutschOxf stopped by to talk about our distant past, why genes an…
Gettier problems are pseudo-problems. Relativity is tentatively deemed true because it is a good explanation.
@BryanMageeNews @HermesofReason
He's rocking those sunglasses.
@skeptic_thomas @ks445599 @DavidDeutschOxf
Don’t know. Persuasion, I suppose.
@ks445599 @skeptic_thomas @DavidDeutschOxf
Coercion isn’t just about physical force. It’s a psychological state in which one idea arbitrarily wins over a conflicting one without solving the conflict.
Yes - they're not creative because their pattern recognition algorithms were designed by biological evolution.
Creativity, OTOH, is evolution happening inside a mind, during the person's lifetime. People are not given pattern recognition algorithms genetically but create them.
RT @mattstark256:
The polaroid game now has a working title: Viewfinder #madewithunity #gamedev #polaroid https://t.co/B6ArM6Ezn1
Creativity is when you solve problems through conjecture and criticism.
The question should be the other way round: is pattern processing possible without creativity?
The answer: no, because one's creativity creates one's pattern processing algorithms.
@ChristopherCode @DavidDeutschOxf
Well, say you can memoize the original function - the person. Then you're not running it. You're just looking up results in a behavioristic table.
In "The Beginning of Infinity", @DavidDeutschOxf asks: "What is the difference between a computer simulation of a person (which must be a person, because of universality) and a recording of [it] (which cannot be a person)?"
An attempt at an answer: