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
Ah, it's not the ignorance per se that's cringeworthy (by some criterion): it's the badly mistaken idea. Some ideas are worse than others in terms of the damage they cause.
Some ideas are cringeworthy, but we may all have different thresholds.
People like Feynman, so they rush to his defense. Ad hominem. And unnecessary: he would have liked to know he was wrong so that he may improve.
Wheeler (both Feynman's thesis advisor and David's boss) knew Popperian epistemology well and may have introduced Feynman to Popper, but it isn't clear.
I know that you know who Wheeler was, but including it here in case others read it and aren't aware of the connections.
Yeah. Quote from David:
[...] I happened to mention Popper in the one conversation I had with Feynman, sometime in the 80s, and he did not say "who's that?" but replied meaningfully to the point.
Feynman seemed to show good understanding of Popperian concepts.
Yeah, he had an aversion to philosophy generally. But he read Popper, or at least he was familiar with his philosophy.
I enjoy Feynman, too, btw. It’s not a personal thing. He made a couple of simple but important mistakes. Inductivist mistakes, no less.
Though I should add that I don't know what he means by "definite" theory. One that isn't vague? Or perhaps one that doesn't require endless computation, as he explains later? Not sure.
Alas, even he made mistakes in his understanding of the creation of knowledge: "[W]e always try to guess the most likely explanation." cringe
Or, less obviously, a bit later: "You can always prove any definite theory wrong."
Sorry? I agree with the first sentence, but lost you after that.
I don’t think so, some wrong answers may still contain truth. Eg Christianity’s commandment not to kill.
From history we know all too well that some were enslaved even though they were much more obviously people than AGIs.
Not to mince words here, but it's the universality of people that implies that everybody is qualitatively equal.
All our laws should apply, but I fear they won't because most won't realize that AGIs are people.
Yes; horribly, that's what companies like @OpenAI and @DeepMindAI seem to be after.
No such thing as "advanced" btw, they are all people just the same.
That's just a map of US airline routes over South America, no?
@ReachChristofer @ToKTeacher @FallingIntoFilm @RatCritical @Malcolm_Ocean @DavidDeutschOxf @reasonisfun @Crit_Rat
Thank you, good point... not sure. Interestingly, whenever subconscious problem solving is successful, the solution does suddenly jump into consciousness (eg shower thoughts etc). So maybe it's something about the correction part of error correction...
@ToKTeacher @FallingIntoFilm @RatCritical @Malcolm_Ocean @DavidDeutschOxf @reasonisfun @Crit_Rat
... until I ride the bike completely subconsciously. Perhaps consciousness is either strongly correlated with error correction, or it may even be error correction.
@ToKTeacher @FallingIntoFilm @RatCritical @Malcolm_Ocean @DavidDeutschOxf @reasonisfun @Crit_Rat
Lately I have wondered if one is aware/conscious of wherever one is trying to detect errors. I learn to ride a bike: very conscious of it, I make mistakes all the time. Then gradually as I iron out the mistakes I grow less conscious of it...
@jamessseattle @ToKTeacher @Crit_Rat
(error correction being the primary ingredient of intelligence)
@jamessseattle @ToKTeacher @Crit_Rat
It can't be analog btw because error correction can only happen in digital systems.
And again, a single Turing machine can simulate multiple Turing machines, so parallelism is incidental at most.
@jamessseattle @ToKTeacher @RatCritical @Malcolm_Ocean @DavidDeutschOxf @reasonisfun @Crit_Rat
In order:
Could be but unimportant - single Turing machine can simulate multiple Turing machines.
Processor and memory.
Doesn't matter/is incidental (if even true).
Yes (if you mean spoken language).
@jamessseattle @ToKTeacher @RatCritical @Malcolm_Ocean @DavidDeutschOxf @reasonisfun @Crit_Rat
No such thing as virtual computer (if by "virtual" you mean "abstract"). Computers need to be physically built.
@ToKTeacher @jamessseattle @RatCritical @Malcolm_Ocean @DavidDeutschOxf @reasonisfun @Crit_Rat
Can you explain why its being analog or digital has any bearing on this?
@caerwy @MatjazLeonardis @DavidDeutschOxf @RatCritical @Malcolm_Ocean @ToKTeacher @reasonisfun @Crit_Rat
I don’t think the presence of understanding (ie knowledge) is indicative of consciousness.
It’s trivial to write a function that represents understanding of a prime number.
I guess that consciousness is related to error correction.
@zarzuelazen
This is the kind of vacuous nonsense that has earned philosophy its bad, navel-gazey reputation.
I agree that we start with conjecture and can then test against brain activity. Thank you; you have helped me realize something important about neuroscience.
Then how can we hope to reconstruct the software that caused these patterns? I think there are infinitely many pieces of software that would result in the same pattern.
@recursus @RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
Yes, we're in agreement here; though I had already agreed that architecture influences speed. But let me ask you this: can two different algorithms, when run, result in the exact same movement in hardware?
@recursus @RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
Do these performance characteristics not lie in the algorithm itself? Would X not also take 10 years to run on a desktop computer? (Assuming that computer would have the same memory and processing power as the brain you're comparing it to.)
@recursus @RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
Or to clarify (something I should have clarified before): hardware architecture can and does influence speed and processing power. But it doesn't qualitatively change anything about which algorithms the universal system can run.
@recursus @RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
Yup, agreed.
@recursus @RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
"not all algorithms that solve a given computational problem are equally efficient or robust"
Indeed, because this concerns the architecture of software.
@recursus @RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
"Not all architectures are equally good for running a given algorithm"
A universal computer, no matter its architecture, can run any computable algorithm (within its memory constraints).
@Malcolm_Ocean @RatCritical @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
When a computer breaks, shit can get weird, too: the fan keeps running, or the housing gets really hot, or it randomly flashes bright colors, some keys work while others do not, it keeps beeping for no apparent reason, it runs all programs fine except the calculator app... etc.
@Malcolm_Ocean @RatCritical @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
It does not. Evolution only optimizes the ability of the gene to spread through the population.
@Malcolm_Ocean @RatCritical @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
Evolution does not optimize for efficient ways to organize any alleged modules.
@RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
If we knew how to program consciousness and ran it on a computer made of chewing gum and vacuum tubes, those interested would start studying the properties of chewing gum in order to understand consciousness.
@RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
Somewhere in the brain there is memory, and somewhere there is a processor. Like in all computers. So what?
In order to understand brain functionality, one needs to understand the software that's running on the brain.
@RatCritical @Malcolm_Ocean @DavidDeutschOxf @ToKTeacher @reasonisfun @Crit_Rat
I haven't read it, but the problem with learning about the brain's functionality from its architecture (hemispheres, regions, parts, etc) is this: the brain is a universal computer. Since it's universal, its architecture does not matter.
I’ve only read one of his stories, and remember it being alright. But I love the video games by @frictionalgames which are Lovecraftian.
RT @andrewdoyle_com:
Politicians should never be invoking the skin colour of their opponents.
This is the cancer of wokeness: it convinces…
@ks445599 @bnielson01 @ToKTeacher
AlphaGo is machine learning. Machine learning is “learning from experience”. It’s empiricism. It’s impossible. Whatever is happening there, it’s not learning. No explanatory or any other alleged kind of knowledge is created.
@SimonDeDeo @wileyprof
Me neither. Finished it yesterday. Was a pain to get through.
I should add that it gets worse, though. I know some decision makers in Silicon Valley who know damn well it has nothing to do with intelligence but choose to call it that anyway because it sells. That’s fraud.
Learning is the creation of knowledge. Narrow AI is the following of predefined steps; the same as all programming so far. Using the term "learning" for something that clearly isn't learning is utterly misleading and dishonest (I'm mad at the industry, not you.)
@ChipkinLogan @scotthortonshow
Very well written. It really is denial of knowledge creation again: all the wealth is allegedly already present, just needs to be redistributed "fairly".
@onnlucky @bnielson01 @ks445599 @mizroba
Well, then maybe you should read more Popper (something I have recommended to you before).
Or at least google "Popper/Bühler four functions of language".
You will then learn that most of the things you listed require explanations, not merely descriptions.
@onnlucky @bnielson01 @ks445599 @mizroba
They're not. The point is it doesn't matter if it's vocalized or only descriptive. Popper explained that the descriptive function of language is built on top of the signaling function.
@onnlucky @bnielson01 @ks445599 @mizroba
None of the things you just listed require spoken language. The only way I can make sense of this is that you thought we were talking about programming language or something. We weren't.
@markcannon5 @bnielson01 @mizroba
Yes. Whoever builds an AGI will be its parent. His obligation will be to help it learn.
@bnielson01 @ks445599 @mizroba
Yes. Language is not required to make a universal explainer. How could it be? You need to be one in order to learn language in the first place. Again, no instruction from without possible.
Have the terms "eco-induction" or "green induction" been coined yet? Might be useful to refer to this sort of idea.
That's what this idea suggests: that there is knowledge somehow already present in the biosphere (not as adaptations!), and we just need to harvest/ingest it.
"How much do species and ecosystems contribute to the size and growth of economies?"
They don't. Only people do. And only explanations guessed by people tell us which resources, species, and ecosystems are at all relevant to economic growth. twitter.com/NaturePortfoli…
"But if [@DavidDeutschOxf] is right that evolution has led to humans as the ultimate knowers, where did this knowing arise?" from sfgate.com/opinion/chopra…
That is a problem the theory of evolution solves. David addresses it in "The Beginning of Infinity".
@ks445599 @bnielson01 @mizroba
Indeed. It could start making all kinds of conjectures about reality and criticize them. It just won't be able to test them. For example, it will know that it exists, and that therefore things can exist. It can then guess that other things might exist, too, what they may be, etc.
@bnielson01 @ks445599 @mizroba
I think you're flirting with induction here. A child does not learn to be intelligent, especially not via instruction from without. He is born intelligent. What knowledge he creates using that intelligence and whether IO may help with that are different matters.
RT @JeffBezos:
It’s time to go back to the Moon, this time to stay. @LockheedMartin @northropgrumman @DraperLab @blueorigin #gradatimferoci…
"How can we build beneficial AGI?" is yet another "Who should rule?" question. It should be replaced with "How can we detect our errors in building AGI?" and then, once built, "How can we help AGI learn?"
Stopping doing one thing and starting doing another is easily programmed and doesn’t involve the halting problem.
It wouldn’t need to be made of DNA. If it wants to study chemical reactions they can be exosomatic.
Ah, yes. Though to be clear, an AGI running on a computer is a completely non-biological phenomenon. No genes required.
That's fine, but then I still don't understand your original comment: twitter.com/EvanOLeary/sta…
Sure. But in order to learn about music?
Why would it have to be able to perform organic chemical reaction experiments in order to be able to learn anything?
Not sure I follow. Care to elaborate? I don’t think any biological substrate is needed.
I’m... not even mad. The article should be titled “In Flagrant Disregard of Computational Universality”. Some real gems in there. twitter.com/CloserToTruth/…
@reasonisfun @HeuristicWorld @dela3499 @ToKTeacher @Crit_Rat @RatCritical @BretWeinstein
Yet they can, and frequently do, lead to death. I suspect this is similar to how genes have to keep their hosts alive but often don’t.
@HeuristicWorld @ToKTeacher @reasonisfun @Crit_Rat @dela3499 @RatCritical @BretWeinstein
Thanks.
“But all we need to know is that for an expensive belief to travel through history with some population over a long period of time, it must be paying its way somehow.”
No. Static memes don’t survive by solving problems (“paying [their] way”).
@ChurchoftheSim @HeuristicWorld @ToKTeacher @reasonisfun @Crit_Rat @dela3499 @RatCritical @BretWeinstein
LOL, "rational religion". Classic example of a static meme trying to compromise, creating garbage in the process. "Sure, 100% religion is bad, but let's reduce it only to 50%" (or 20% or whatever).
@HeuristicWorld @ToKTeacher @reasonisfun @Crit_Rat @dela3499 @RatCritical @BretWeinstein
Broadly speaking: Yes, absolutely. It denies creativity by claiming that everything important that can be known is already known, it punishes dissent, creates bad explanations... in short: actively prevents progress.
I am not familiar with Weinstein's take on it.
The discovery of the equivalence of software engineering and reason, if I say so myself.
@KittJohnson_
Possible, but unimportant. A single UTM could simulate all of that.
RT @DavidDeutschOxf:
My robotic TED talk is online at last: ted.com/talks/david_de…
@RatCritical @HeuristicWorld @ReachChristofer @B_crawford_19
Thanks, but it appears my most recent comments were rejected. So was a post I made today.
Is there a way to permanently whitelist me? I am not that active on Reddit otherwise, so I have no idea how soon my comment karma will rise to sufficient levels.
Could one do a brisk walk near the poles of the earth as well to keep up?
I wrote detailed responses, but apparently they were all removed because my Reddit karma is not high enough to post comments in this Subreddit.
A critical discussion of @ReachChristofer's latest episode with me. twitter.com/FoundAtaraxia/…
"There is no distinction between the computer and the program. They are one and the same."
Utter nonsense.
@ReachChristofer @SamHarrisOrg @DavidDeutschOxf
A correction to a mistake I made: our computers are technically not Turing complete because they have finite memory. We may choose to consider them Turing complete for practical purposes.
Any run of the mill computer also has software interacting with hardware and vice versa. It does not chance the distinction between the two.
The "neuroscience of learning": topos.house/residency.html
The brain is a computer. People learn via a program on that computer. We need to study that program, not the computer.
@aroraharshita33 @james_ough
Here's your answer: you could, but only by increasing the brain's processing speed somehow.
The second link states you're trying to understand how humans learn. That's the right question to ask, but have you looked into whether we already know a fair amount about this? (We do.)
@MartvMegen @ReachChristofer @SamHarrisOrg @DavidDeutschOxf
Yes. Can you guess why?
I was on @ReachChristofer's podcast. We spoke about evolution, knowledge, machine learning, and the benefits of building AGI - among other things.
Thanks for having me!
Yes. Computation is when you use physical objects to represent abstract ones.
RT @ChipkinLogan:
Episode 3 of Fallible Animals, titled Progress as Error-Correction, is out now on iTunes, Youtube and Spotify!
https://…
Nice! Also, looking forward to learning about constructor theory.
RT @PessimistsArc:
'The Decline of Intellect' (The New York Times, 1900)
“The human intellect, like “the service” has long been “going to…
@ks445599 @markcannon5 @dela3499 @DavidDeutschOxf
I’m guessing he’s talking about weighted connections between neurons as inspired by neural networks, AKA coefficient matrix. Not about weighing ideas.