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
Yes, good point. Perhaps what we are left with is that a universal explainer has the capacity to experience qualia, but doesn’t necessarily do so? Eg feeling hungry is a kind of knowledge?
If so, does that mean non-explainers cannot experience qualia?
Yup agreed. I’m after explanations at this point :)
Sounds like some neuroscientists are finally starting to see their own reductionism.
I personally stay away from qualia for the most part because they are utterly mysterious. If a universal explainer automatically has them, great; if not, I don’t really care all that much. May even make things easier if it helps avoid moral issues.
Depending on one’s definition of AGI, a universal explainer without qualia could be considered a little less than an AGI.
Lack of input/output devices doesn’t necessarily suggest lack of qualia. See aeon.co/essays/how-clo…
I agree however that even without qualia a universal explainer is genuinely universal in its capacity to explain.
Yeah, I still go back and forth on whether a universal explainer would automatically have qualia. I agree that it’s not obvious.
Episode 7 of the podcast on artificial creativity is out; as always, heavily inspired by @DavidDeutschOxf. This time, it's a Q&A episode.
@thethinkersmith @ToKTeacher @Crit_Rat @SamHarrisOrg
Neuroscience would need to violate computational universality in order to contribute.
Only half of all households in 1990 had a stove? How did people cook?
@tjaulow
The only thing universality suggests is a shared repertoire.
@thethinkersmith @ToKTeacher @Crit_Rat @SamHarrisOrg
That could all be genetic or memetic.
@thethinkersmith @ToKTeacher @Crit_Rat @SamHarrisOrg
I’m curious. How do primates demonstrate creativity?
@tjaulow
I think I understand what you mean, but I don’t think universality on its own suggests that.
It’s just a turn of phrase. It means that something that is considered an exception doesn’t break the rule precisely because it is an exception, as opposed to something that was not expected to occur.
For example, I could learn how to be attracted to men, if I chose to. This may be hard but can't be impossible, since women have that knowledge somehow.
Yeah, this helps. I think I was simply wrong about what heterosexuality implies. It means that there is different knowledge in men and women; but (given the right technology etc) nothing prevents either from learning what the other has.
While there are differences between kinds of knowledge any two people hold - e.g. you know something I don't - nothing forbids my creating that missing knowledge for myself. The universality here lies in the ability to create knowledge, not differences in existing knowledge.
But while there are no differences between people as explainers, there are undeniable differences between people as it relates to different genders.
x is a universal y if it can do all the z's all the other y's can do.
People are universal explainers. Any given person can in principle explain anything any other person could explain. That means in their ability to create knowledge, all people are literally the same.
Makes sense. I also just remembered David saying somewhere that people have an inborn fear of heights which they can exploit for fun (eg parachuting). But could they get rid of it?
How do we square gender specific preferences with universality? For example, most men are attracted to women.
Presumably, genes can create preferences, interests, etc in people, but can be overwritten by the mind?
- Knowledge is information that is adapted to a purpose.
- Knowledge is information with causal power.
- Knowledge is information that solves a problem.
Those are the three relationships that come to my mind.
RT @DavidDeutschOxf:
@notsurethomas @lynz_h55 @TheHalcyonSavan @PSTaylor13
Explanations never explain why they themselves are true. That wo…
First time I've seen it, but the description sounds promising.
@RealtimeAI @chrisalbon
Hahahaha omg that sounds so disgusting! 😂
Join me for a talk on artificial creativity this Sunday at 8 at Rainbow Mansion in Cupertino, CA:
The chicken's name is Ernie! I had no idea...
He's too invested now. He would probably, as most people, think it a public failure to change his mind about superintelligence.
Deeper reason: pessimism.
Haven’t read his books yet, but some of his papers I’ve read are among the worst “contributions” to the field. Complete nonsense. Will likely never come around.
@zombieinjeans
Don't know.
FWIW, definitions of knowledge I also like include "explanation" and "information that is adapted to a purpose", which latter definition includes being hard to vary.
And yes it has causal power and tends to remain physically instantiated.
Google the term "knowledge" and you'll get:
"facts, information, and skills acquired by a person through experience or education [...]"
😭😭😭
I was worried you’d say that. Will skip this one, thanks :)
@zombieinjeans @DavidDeutschOxf
Blind evolution often finds solutions in the biosphere. But knowledge in humans has the advantage that not every single explanation along the way has to work, so it has more flexibility.
Genetic pluralism may be part of the answer, too.
It doesn’t prove general relativity. Epistemological blunder.
RT @DavidDeutschOxf:
Agreed. (Disturb.) twitter.com/ZachG932/statu…
How was “A New Kind of Science”? It was recommended to me buy an AI person, but I haven’t gotten around to it yet.
@zombieinjeans @DavidDeutschOxf
I go back and forth on this. On the lowest lvl there seems to be blind evolution happening in our minds. In hindsight it looks goal oriented and purposeful, not random. So does biological evolution, though. But yes, things we have learned in the past can help with a new problem.
Got it - are you referring to the part where I invoke the hidden target function to gather return values and then reconstruct the function from those values as “explaining data”?
Got it. I think I understand what you mean by “objective”. And what do you mean by “criterion”?
What do you mean by “mismatch between criterion and current objective”? A mismatch between target function and its replica?
Yes, I think of it this way as well.
@ChristopherCode @DavidDeutschOxf
I know next to nothing about quantum operations, but the laws of epistemology should apply to function implementations on quantum computers as well.
@DavidDeutschOxf would know better.
How do they do this? Do they not implement a function of their own to imitate the other function?
It should apply to all explanations.
I’m not sure yet how to represent more complex functions such as human speech, let alone philosophical theories. But given lambda calculus’ universality, it must be possible.
FWIW functions without parameters are also explanations.
The sixth episode of the podcast on artificial creativity is out. It's another video episode, and as always, heavily inspired by @DavidDeutschOxf:
@madeofmistak3 @ghstgrllll
Not yet - thanks for the recommendation!
@madeofmistak3 @attractfunding
Can I throw “The Babadook” in the ring?
@madeofmistak3 @ghstgrllll
Mother! is unfortunately full of spaceship earth thinking, but good otherwise. I know it splits the crowd because half the movie is supposed to be annoying, but if you can ignore that, you’ll have a great time watching it.
RT @ToKTeacher:
Article summarising the prevailing intellectual view of humanity at a concentration approaching saturation. Unbounded pessi…
RT @davidarredondo:
Popper on Definitions: interesting. youtu.be/fgeMEr16yYk
Just watched it today myself. Love me some Popper.
@Azaeres @onnlucky @SerhiiHavrylov
Yes. I will get into some of that in the next episode. One thing to note is that there are always infinitely many functions that will fit the dataset perfectly. The question is how to find good, i.e. hard to vary, function implementations that match.
@RichardDawkins @center4inquiry
Nice. Add @Target for selling homeopathic eardrops I accidentally bought once thinking they were legitimate.
@Azaeres @onnlucky @SerhiiHavrylov
Ah, got it. Again, I’m hesitant to address problems of scale or performance without first knowing how the creative algorithm works. Only that knowledge would suggest performance problems. But you mentioned you had some background knowledge about this. Care to share?
@Azaeres @onnlucky @SerhiiHavrylov
That is precisely how an AGI does not learn. It’s induction again.
@Azaeres @onnlucky @SerhiiHavrylov
How do you know it will take immense resources without having a good explanation of how the program works first?
@onnlucky @Azaeres @SerhiiHavrylov
Indeed. We already know that modern day computers could run a universal creativity program. What's lacking here is neither hardware nor processing power: we lack a good explanation of how to write the program.
If resources were the problem, we would gladly wait to run it if we had a good explanation of how it works.
Bayesianism is inductivism in a cheap tuxedo. Has nothing to do with creativity. See bretthall.org/bayesian-epist… by @ToKTeacher
The problem with inductive processes is that they don’t exist.
Re intractable: I don’t know what you mean by “holistic nature”, but our minds are creative somehow, so it can’t be intractable. We just don’t yet know how.
I skimmed it. The causal approach is the right one. Also reading up on program synthesis now.
@SpaceTime_A @DavidDeutschOxf @ToKTeacher
Great question - I have no idea :)
It is creativity. And we are beginning to understand it. Baby steps :)
@ChristopherCode @dela3499 @DavidDeutschOxf
I responded in the document in purple.
@ChristopherCode @dela3499 @DavidDeutschOxf
Indeed. E.g. for objects in JS, Clojure etc one first needs an explanation of how to map keys reliably to values: a hash function. We know from lambda calculus that any such concepts can be built from pure, single parameter functions, because it is Turing complete.
Yeah, that’d be pretty neat. Being a universal explainer, the creative algorithm could do so.
@Doxosophoi
What do you mean by “epistemological facts”? Do you mean rules such as hard to vary? If so, my guess is knowledge of those rules is the result of the creative algorithm as well. If your question is how the creative algorithm creates new explanations, I haven’t covered that yet.
First time reading about this, too. First impression: Sounds like Occam’s razor applied to functions. I’m after good function implementations, which need not be the shortest. The best implementation of multiplication in this episode only happens to be the shortest, IIRC.
First time reading about Solomonoff’s inductive inference, so may be missing something. Wikipedia says it “is a theory of prediction based on logical observations”. Sounds inductive to me. Too focused on observation. No mention of knowledge creation as far as I can tell.
The podcast is now available on Apple Podcasts:
Thank you for reminding me. Episode 5 you will probably want to watch on YouTube since it heavily relies on video, but all others so far you can now listen to on Apple Podcasts as well: podcasts.apple.com/us/podcast/pod…
Today I learned that some recursive functions cannot be written in iterative form:
The fifth episode of the podcast on artificial creativity is out. This time it's a video episode. As always, heavily inspired by @DavidDeutschOxf.
@krazyander @ToKTeacher @RealtimeAI @DoqxaScott
One of the points of h2v is that you don’t even need to bother testing e2v theories.
@krazyander @ToKTeacher @RealtimeAI @DoqxaScott
You observe a conjuring trick and want to explain how it’s done. The explanation “the conjurer did something” has a lot of “evidential support” - you saw him doing something - but that’s a bad explanation b/c easy to vary. Doesn’t explain what he did and applies to any trick.
Glad to hear! Currently working on episode five, which will be about how the epistemological concepts we covered in episode four apply to computer programs and what that means. Hopefully ready to release soon!
@RealtimeAI @krazyander @ToKTeacher @DoqxaScott
You can tell an explanation is a hard to vary if you try to change parts of it and you end up diminishing its ability to fulfill its purpose. I.e. an explanation is hard to vary vis-a-vis a problem it purports to solve. Pick any theory and try changing parts; if difficult, h2v.
I guess that the flaws you mention are generally the result of irrational memes and/or bad explanations.
In order:
Not generally. A chemistry lab may blow up during an experiment or something, but that’s an exception. Extinction of ideas is fundamentally non violent.
Those may all be drivers but can be overwritten. GPPS still susceptible to irrational memes.