gilch's Shortform 2019-08-26T01:50:09.933Z · score: 3 (1 votes)
Stupid Questions June 2017 2017-06-10T18:32:57.352Z · score: 3 (3 votes)
Stupid Questions May 2017 2017-04-25T20:28:53.797Z · score: 7 (8 votes)
Open thread, Apr. 24 - Apr. 30, 2017 2017-04-24T19:43:36.697Z · score: 3 (4 votes)
Open thread, Apr. 17 - Apr. 23, 2017 2017-04-18T02:47:46.389Z · score: 1 (2 votes)
Cheating Omega 2017-04-13T03:39:10.943Z · score: 7 (8 votes)


Comment by gilch on I would like to try double crux. · 2019-11-11T21:42:42.780Z · score: 1 (1 votes) · LW · GW

-Basically, since predictions are laws of nature, wrong predictions are miracles. We should expect to them to exist but to be rare.

Due to indexical uncertainty, we can always be surprised by low probability events. I don't see these as evidence of God though.

-Talking about aliens. Infinite hypothesis "God with such attributes exists" can be used only as approximation (that is, basically, our understanding of it). The finite hypothesis "aliens fake us to believe that God with such attributes exists" also can be used only as approximation, (that is our understanding of God + assumption that it is faked by aliens). Thus such approximation is longer and should be given smaller probability.

Around in circles again, but is there a difference this time? Do we agree "fake alien God hypothesis" dominates "infinite God hypothesis"? When using induction? You don't seem to be disputing it. But is "approximate God" simpler than "fake alien God"? That depends! How good is your approximation of "infinite"? How complex are your aliens?

But if you want to argue for a non-infinite God, that's OK with me, but even if you convince me, it won't be the infinite God you have convinced me of, but the finite approximation: Something more powerful than mankind, but not infinitely powerful. Something more knowledgeable than mankind, but not infinitely knowing... this sounds like you're describing advanced aliens. They're the same thing. I would then argue that the aliens are the reality and the "infinite God" is the approximation of them made by ignorant humans.

Even I would be willing to call such aliens "gods" given certain conditions, but we're using your definition of "God".

Can you convince me of approximately-God aliens? Maybe. My prior is not zero, but like the pet purple dragon from Mars, it would take a lot of evidence to convince me.

Comment by gilch on I would like to try double crux. · 2019-11-11T21:13:44.665Z · score: 1 (1 votes) · LW · GW

First, could you review the previous comment to see if you agree with the logic, and if not, what do you disagree in particular.

It feels like we are going around in circles at this point. I'm not sure where the disconnect is.

-if the total possible observation data is infinite, what is prior probability that it is exactly reproduced by finite hypothesis? I argue that it is infinitesimal

The set of all natural numbers is infinite, yet can be enumerated by a finite computer program (when run on an infinite computer, AKA, a Turing machine). There are many many other examples of infinite patterns enumerable by finite programs. And some of them, like "compute the digits of pi" seem pretty chaotic, yet their Kolmogorov complexity is small.

One wrinkle, which you might be alluding to, is that no program with infinite output ever halts. This is true, but there are halting programs that can compute any finite prefix of pi. And like I said before, at no point is your observation infinite. It's always finite so far. The infinity is never completed.

So the hypothesis "these are the digits of pi" is considered by Solomonoff induction, but maybe it looks like a weighted sum of a class of programs that say "compute pi up to the nth digit" for some n. These still compress quite well, (especially for compressible n's) so their Kolmogorov complexity is small. I don't think this is an obstacle for Solomonoff induction.

Has Solomonoff induction got it wrong? Close but not quite? I would argue no. I don't believe uncomputable sets can physically exist. There are no perfect circles. The abstraction called pi is the approximation, for whatever algorithm physics is actually running, which Solomonoff induction would eventually find.

Comment by gilch on I would like to try double crux. · 2019-11-11T19:24:25.610Z · score: 1 (1 votes) · LW · GW

If I would do this I would never go to this website discussing this with you.

A fair point. But I still think you are compartmentalizing.

for infinitesimal prior probability no evidence is enough.

It's never enough for induction, performed correctly. But an a priori deductive argument maybe could work. I've heard theists attempt these arguments, but have not found them convincing.

I can blame the same thing on you.

I am trying to find cruxes, not blame. I would rather leave our identities out of it and examine the question as objectively and impartially as possible. But your epistemology is extremely relevant in this case. It's the rights of Mor­timer Q. Sn­od­grass again. I don't think the God hypothesis has enough going for it to even justify raising it to our attention. If we had started with a good scientific epistemology, this would not even be a question. Instead we started with a biased indoctrination, and have to dig ourselves out of it.

I am not going to guess but there are so many stories of atheists who became atheists just because God didn't do what they asked. "I do not want to deal with such God that does not do what I want, therefore there is no God."

It's the availability heuristic again. Who have you heard these stories from? It's probably not the atheists themselves! You can't trust the clergy to be honest about this topic. They believe atheism is damnation, and so must present it as a sin. But for those raised atheist with a scientific worldview, believing in God seems as silly as believing in Santa Claus or the Tooth Fairy.

In my case, I was raised as a believer. My perspective changed due to an accumulation of a number of factors. The Problem of Evil was apparent to me in childhood. It introduced a doubt that I could not resolve. The biblical creation story also didn't align with what I read of science as a child.

When I expressed my misgivings, my church told me that God was a God of Truth, and the teachings of the Church could not possibly contradict the Truth, once it was properly understood. So I withheld judgement until I could learn more. I held both the religious and the scientific worldview in my mind at once, in the hope that they could eventually be unified. I was compartmentalizing, but I was conscious that I was doing so. I could speculate and philosophize in either religious or scientific modes, and I knew which was which. I saw the fruits of science. Computers and rocket ships and vaccines. I had church-related experiences I could only describe as spiritual. Surely they both had to be true?

I studied my faith in depth. I was warned of the sin of pride. I was uncertain how to interpret that, but after study, concluded that the problem with pride was an unwillingness to learn from error. I resolved to always be honest with myself. God was a God of Truth, after all, so honesty could not be wrong. I learned to think more critically. I found many satisfying answers, but my doubts on these points, and more, only deepened. There was evidence against the faith, that was for certain. Doubts remained, but abandoning my faith would mean damnation and I could never convince myself it was false beyond a reasonable doubt.

Then I learned that civil cases were judged according to the preponderance of the evidence, rather than beyond a reasonable doubt. In my commitment to honesty, I judged my faith again by this standard. Suddenly, many of the faith-promoting stories I had considered "evidence" no longer appeared that way. They were indistinguishable from no God at all. Once seen, I could not unsee it. Why was God pretending so hard not to exist? So we are less culpable for sins? Then why have a church at all? My faith was shaken (and not for the first time), but still I believed. I resolved to study more, to try and rebuild what I had lost.

In my church, we brethren sometimes minister to the other members, usually in pairs. I was usually too shy to participate, but I had studied enough to know answers from the scriptures. When ministering to one poor sister who was struggling, I went into religious mode and spouted off the relevant doctrine. This happened to be a point I had doubts about. And then the realization struck me: I didn't believe a word of it. I sounded that confident, and I didn't believe a word. I had lied to her. And worse I had lied to myself, the exact thing I had resolved not to do. I had so easily broken my commitment to honesty, just by studying doctrine. And if I could do it, so could any of the other members! They could sound so convicted, and yet not know! The testimony of the others I had been relying on may have been founded on nothing but air.

I still had my spiritual experiences, but they had always resisted critical examination. I finally understood that what I thought was the witness of the Holy Spirit, was only those around me interpreting my emotions for me in a certain way. They were spouting off doctrine memorized by repetition, the same as I had done to that poor sister. In another context, the same emotions could have been a witness for a completely different god. My faith was shattered to its very core.

My church regards all others as apostates. I had rejected them long ago. There was nowhere to turn. For a time, I considered myself agnostic. I told my story to a confidante, and she replied with something like, "so you're an atheist then". And in that moment, I realized it was true. I'm an atheist. I can't believe in God anymore, even if I try.

And after reading the Sequences, and understanding Bayes, I realized that the faith-promoting stories I had thought were evidence, and then eventually no evidence at all, were actually evidence against the church. The church had actually been preemptively preaching some of its worst stories, so we would learn to think of them in the best possible light, before we had a chance to hear a more critical presentation from anyone else.

Comment by gilch on I would like to try double crux. · 2019-11-11T04:05:06.125Z · score: 3 (2 votes) · LW · GW

well, yes, if we can not at all observe in all the infinite future infinite data, it does not make sense to talk about omniscient God

You are not a future hyper-mind made of basement universes and wormholes. You're a mortal human like me, with a lifespan measured in mere decades so far. Yet you claim to have knowledge of an infinite God. How did you come to this conclusion? By what method can you make such an assertion? Is this special pleading for a special case or do you use this method for anything else? Why should I consider that method sound and reliable?

My best guess: you were indoctrinated in childhood by your parents and community, long before you were old enough to develop critical thinking skills of your own. For obvious survival reasons, children are very inclined to learn from their parents and elders. The memeplex of any of the old religions must be self-sustaining, or they wouldn't still be here. They include psychological tricks to produce fake evidence, to stop questions, to make empty threats. They include answers to your questions or at least pretend to. It became part of your identity. You later learned of the methods of science, but they didn't become a part of you the same way. You compartmentalized the lessons and didn't use them to update your old thinking. You sought out evidence to support your belief instead of trying to disprove it to see if it would hold up, like a scientist.

Most people seem to use this method. You are not alone. And that's exactly the problem with it. People are using the same methods to believe in other religions that you already know to be false. How can that method be reliable if it so reliably produces the wrong answers? What makes you any different from them? Accident of birth. That's it. Your methods are the same.

Maybe that's a crux for me. If it could be shown that a God belief was founded on a sound epistemology that reliably produced good results, instead of these obvious fallacies, I would have a much harder time dismissing the proposition as a fraud.

Comment by gilch on I would like to try double crux. · 2019-11-11T02:14:53.106Z · score: 1 (1 votes) · LW · GW
  1. You can't deduce Born's rule - only postulate it.

Can't you? Carroll calls it "self-locating uncertainty", which is a synonym for the "indexical uncertainty" we've been talking about. I'll admit I don't know enough quantum physics to follow all the math in that paper.

  1. Most important, it does not give you a prediction what YOU will observe (unlike hidden parameters - they at least could do it). Yes, you know that some copies will see X, and some will see Y, but it is not an ideal predictor, because you can't say beforehand what you will see, in which copy you will end up.

Yeah, in this scenario, the "YOU" doesn't exist. Before the split, there's one "you", after, two. But even after the split happens, you don't know which branch you're in until after you see the measurement. Even an ideal reasoner that has computed the whole wavefunction can't know which branch he's on without some information indicating which.

So all your future observed data can not be predicted, only the probability distribution can be.

More or less. You can compute all the branches in advance, but don't necessarily know where you are after you get there. The past timeline is linear, and the future one branches.

Comment by gilch on Open & Welcome Thread - November 2019 · 2019-11-11T01:40:22.827Z · score: 1 (1 votes) · LW · GW

Try reading something boring. It works for me.

Comment by gilch on Open & Welcome Thread - November 2019 · 2019-11-11T01:38:32.683Z · score: 2 (2 votes) · LW · GW

Copied from my answer here.

  • Take melatonin at the appropriate time and dose. It's cheap and legal in the U.S., but most products have way too much. most insomnia drugs are not much more effective than this.
  • Avoid light at night, especially blue light. Light inhibits natural melatonin production, which interferes with your circadian rhythms.
    • If you can't darken your room completely, you can use a sleep mask instead. Get the kind with cups (like opaque swim goggles) instead of the kind that puts pressure on your eyes.
    • Use f.lux on your personal devices to reduce blue light after sunset or use one of the similar built-in features of your OS. Windows 10 has the new "Night Light" setting, macOS and iOS have "night shift" mode. Newer Samsung phones have a "blue light filter" setting. These options vary in quality and may have configurable intensity. More intense is more effective and it's surprising how much you get used to it.
  • Falling asleep is a common failure mode of certain types of meditation practice. You can use this to your advantage when suffering from insomnia in bed. Even beginners fail to meditate this way accidentally, so it's not particularly difficult to do on purpose. Focus your attention on the sensation of breathing or on the ringing in your ears. When you notice you are lost in thought, refocus your attention. But when you notice the dreaming arise without directed effort, dive in and let them take you. It works for me anyway. If not, at least you got your meditation in today. [...]
  • Track your sleep quality.
    • You can get smartphone apps that purport to do this using the phone's sensors. Some fitness trackers or smartwatches also have this function built in or available as an app. Accuracy varies.
    • You may have sleep apnea. Talk to your doctor about doing a sleep study to diagnose possible issues and treatments. Some people do much better on a CPAP, but there are many other treatment options.
  • Avoid eating late at night. This can cause indigestion, which can keep you awake.
    • if you suffer from heartburn, sleep on your left side to contain it better, because your esophagus attaches to your stomach on the right side (unless you're one of those rare people with backwards internal organs).
  • Exercise regularly. I'm not sure why this helps, but it seems to. Perhaps mental fatigue doesn't always line up with physical fatigue unless you actually make some effort physically during the day.
Comment by gilch on I would like to try double crux. · 2019-11-11T01:21:45.168Z · score: 1 (1 votes) · LW · GW

Anyway, there are a lot of speculations what can be and what can not.

I worry we may be getting lost in the weeds again. We need to try and find cruxes. Is this related to a crux of yours? What exactly are you getting at?

prior probabilities: that all our possible future observations are finite and that they are infinite are not negligible?

Even if time could be extended infinitely without the universe dying, there is no time at which the infinity has been completed. It's always finite so far.

An "immortal" being with finite memory in infinite time will eventually forget enough things to repeat itself in a loop, living the same life over and over again.

Can this be avoided? There are limits to any physical realization of memory. If you try to pack too many bits in a given volume of space, it will collapse into a black hole. And then adding anything more will make the event horizon bigger. Infinite memory requires infinite space and energy. Maybe with basement universes it could be done. They might have to communicate through wormholes or something. This is all very speculative, so I don't know.

Comment by gilch on I would like to try double crux. · 2019-11-10T23:51:02.254Z · score: 1 (1 votes) · LW · GW

Only if you're interested. I haven't actually read the whole book myself, but I have read LessWrong discussions based on it. I think the Sleeping Beauty problem illustrates the important parts we were talking about.

Comment by gilch on I would like to try double crux. · 2019-11-10T21:35:13.022Z · score: 1 (1 votes) · LW · GW

I'm pretty sure I said that:

But, when they encounter external environmental differences, their timelines will diverge.

I don't understand your point.

Comment by gilch on I would like to try double crux. · 2019-11-10T21:15:41.511Z · score: 1 (1 votes) · LW · GW

Could you please explain it in more details? I am confused. If I measure the spin of the electron that is in the superposition of spin up and spin down, I obtain with probability p spin up and with probability 1-p spin down. How to exactly predict using Tegmark multiverse when I see spin up and when I see spin down?

I'm not saying that a Tegmark I multiverse is equivalent to MWI, that's actually Tegmark III. I'm saying that Tegmark I is sufficient to have indexical uncertainty, which looks like branching timelines, even if MWI is not true. See Nick Bostrom's Anthropic Bias for more on this topic.

Comment by gilch on I would like to try double crux. · 2019-11-10T20:54:09.219Z · score: 1 (1 votes) · LW · GW

Side note: the Universe with finite number of states is quite depressive picture since it means that inevitably everything will just end up in the highest entropy state, so, the inevitable end of humanity.

Yes, according to our best current understanding of cosmology, the universe itself will eventually die (i.e. become unable to sustain life).

Of course, it contradicts nothing, but in this model any discussion of existential threats for the humanity (like superintelligence quite popular here) makes no sense since the end is unavoidable.

Again the laws of physics are what they are and don't care what I want.

But in the most likely scenarios, this will take a very long time. The Selliferous Era (when the stars shine) is predicted to last 100 trillion years, and we're not even 14 billion years into it. Civilization may continue to extract energy from black holes for a time many orders of magnitude longer than that.

It's not completely hopeless. Maybe in that time we'll figure out how to make basement universes and transfer civilization into a new one, as Nick Bostrom et al have argued may be possible.

But even if we ultimately can't, shouldn't we try? Shouldn't we do the best we can? Wouldn't you rather live for over 100 trillion years than die at 120 at best?

Comment by gilch on I would like to try double crux. · 2019-11-10T20:43:26.532Z · score: 1 (1 votes) · LW · GW

The observable universe probably has a finite number of possible states."

Not so sure about that.

Not sure either, which is why I said "probably".

For this you need at least

  1. The Universe to be finite (i.e. you can not have open Universe, only the surface of 4d sphere). It is possible, the measured curvature of the Universe is approximately on the boundary, but the open is also possible.

Note that I said "observable universe", not "multiverse" or "cosmos". There are regions of the universe that are not accessible because they are too far away, the universe is expanding, and the speed of light is finite. This limit is called the Cosmic event horizon

  1. The Universe to be discrete on microscale. Again, according to some theories it is the case, according to the others, it is not.

I think it is sufficient to say that the information content of the observable universe is finitely bounded. Space doesn't necessarily have to be made of pixels like some cellular automaton for this to hold. The Bekenstein bound is proven from Quantum Field Theory. How true QFT is, is another question, but experimental evidence proves that it is very true.

Comment by gilch on I would like to try double crux. · 2019-11-10T20:07:56.790Z · score: 1 (1 votes) · LW · GW

Yeah, given New Foundations, I'm no longer confident that "omniscience" is a logical contradiction, but neither am I confident that it isn't. And I still think it would take an infinite amount of evidence to prove inductively, so you would need some kind of a priori argument for it instead (or why believe it at all?). That's one obstacle down, but still a long way to go.

Comment by gilch on I would like to try double crux. · 2019-11-09T23:55:57.528Z · score: 1 (1 votes) · LW · GW
  • The observable universe probably has a finite number of possible states.
  • The laws of physics appear to be deterministic and Turing computable.

Therefore, an infinite theory would never be required. (And this makes me sympathetic to the ultrafinitists.) The laws of physics can be mapped to a Turing machine, and the initial conditions to a (large, but) finite integer. There is nothing else.

But I'm not sure that "all possible future observations" means what you think it means.

In the MWI, any observer is going to have multiple future Everett branches. That's the indexical uncertainty. Before the timeline splits, there is simply no fact of the matter as to which "one" future you are going to experience: all of them will happen, but the branches won't be aware of each other afterwards.

And MWI isn't even required for indexical uncertainty to apply. A Tegmark level I multiverse is sufficient: if the universe is sufficiently large, whatever pattern in matter constitutes "you" will have multiple identical instances. There is no fact of the matter as to which "one" you are. The patterns are identical, so you are all of them. When you make a choice, you choose for all of them, because they are identical, they have no ability to be different. Atoms are waves in quantum fields and don't have any kind of individual identity. You are your pattern, not your atoms. But, when they encounter external environmental differences, their timelines will diverge.

Comment by gilch on I would like to try double crux. · 2019-11-09T18:30:36.463Z · score: 1 (1 votes) · LW · GW

I still don't understand. Are you asking if the universe is deterministic?

Which sense of "exist" do you mean? Mathematically, where we can "have" imaginary things like infinite uncomputable sets, or physically, where we obviously can't construct an object corresponding to such a thing?

Solomonoff induction cannot be run on real physics. It's an abstracted ideal that can only be approximated. Maybe quantum field theory predicts the motion of particles to an accuracy of eleven digits, but that doesn't mean you can use it to predict the weather. You don't have enough computing power, and you don't know the initial conditions to that precision anyway.

Even AIXI, an ideal agent using Solomonoff induction (which can't be physically built), can only make probabilistic predictions based on observations made so far. There's always an infinite class of universes (hypotheses) that have produced the observations thus far, and they always disagree on the next bit.

There's no need to invoke quantum physics here. Given what we already know of relativistic physics, it's always possible that a particle could approach at the speed of light and mess up your plans. Because it's moving at light speed, there's no way in principle you could have observed it to take it into account in advance. Even AIXI can be "surprised" by low-probability events like this, even in a deterministic universe (because it has only observed a small part of the universe so far), and it has infinite computing power!

Comment by gilch on I would like to try double crux. · 2019-11-09T18:02:33.231Z · score: 1 (1 votes) · LW · GW

Strong enough evidence can overcome a very low prior, yes. And this doesn't have to take very many observations.

But more instances do not necessarily stack like that. That can only happen to the degree they are independent sources. For example, suppose you write a dubious claim in a book, then you make nine more copies of the book. Does that make the claim ten times more likely to be true? What if it's a hundred thousand copies? Did that help?

Of course it doesn't! You're re-counting the same evidence. The contribution of the nine books is completely screened off by the first; the new books have no new information.

I think the cases of miracle reports like weeping icons are similarly not independent enough. A thousand weeping icons is barely more evidence than one. It just means that the hoaxers copied each other's scam.

Furthermore, we already know that some similar instances of miracles were hoaxes. Shouldn't every new hoax report lower my prior that miracles are real?

Comment by gilch on I would like to try double crux. · 2019-11-09T04:09:13.709Z · score: 1 (1 votes) · LW · GW

the Universe is completely describable by a finite set of laws, no other reality behind

I don't consider this question well-posed. Physics seems to be working pretty well. But what do you mean by "Universe"? The part we can observe? Surely there's more to it than that. And "laws" can be dependent on context. The law that objects accelerate downward at 9.8m/s/s doesn't apply on Mars, but there's a similar law with a lower number and an underlying law of gravity connecting both cases. Laws that seem to be "fundamental" now are probably dependent on local conditions. The "symmetry breaking" observed in particle physics indicates this. And very simple rules like Conway's Life can produce very complex behavior, with emergent "laws", like "gliders travel diagonally". Is this law from a reality behind or in front of Life?

Comment by gilch on I would like to try double crux. · 2019-11-09T03:52:59.849Z · score: 1 (1 votes) · LW · GW

Errr not completely - you have prior and you have experience.

The posterior becomes the next prior when updating again, so we still call it a "prior" even though this is not the same prior as before. Sorry for the confusion. My current prior is my current level of belief/confidence.

Then if you observe miracles you update it to much higher probability

Higher, yes, but (say) ten times almost nothing is still almost nothing. And that's only if the likelihood ratio for the evidence favors the hypothesis by that much, which it doesn't.

but you can't do it if your prior is infinitesimal as now.

That's right. No finite amount of evidence can overcome an infinitesimal prior.

Your example "miracles" are evidence in favor of miracles existing (because we can hardly expect reports of miracles to be less common if miracles exist) but the likelihood ratio is very close to 1 because false positives (accidents, hallucinations, and hoaxes) are so common. On priors, these explanations are far more likely. That means your "miracle" reports are extremely weak evidence.

I cannot lower my epistemic standards on this, or I would invite in flat-Earthers, UFO-ologists and various other conspiracy theorists, not to mention all the other religions who have similarly dubious paranormal claims. Why should I favor your paranormal claims over theirs? It's special pleading.

Comment by gilch on I would like to try double crux. · 2019-11-09T03:29:11.127Z · score: 1 (1 votes) · LW · GW

However, there are different set theory axiomatics. Some of them allow universal sets

OK, that's a good point. I had not heard of the universal sets that contain themselves, which I thought would lead to contradictions.

should be able to kill the concept of Tegmark mathematical multiverse

I'm really not persuaded by the MUH, but at least it's based on reasoned a priori arguments. Do you have similar a priori arguments for God? There's no way for evidence to ever be enough establish omniscience by itself.

Comment by gilch on I would like to try double crux. · 2019-11-08T06:39:20.325Z · score: 1 (1 votes) · LW · GW

That's your crux? Lesser interpretations than E won't do?

I am not convinced that E is logically coherent. It's as meaningless as "married bachelor".

  • Suppose that God's memory is the set of "all facts" O.
  • The set of all subsets (or powerset) of O, we'll call p(O).

Then, for any given fact f, there is a further fact f ' stating that it's either in or not in each subset of O in p(O).
Thus, there must be at least as many facts as there are elements of p(O), which, being the powerset of O, by Cantor's Theorem must have a strictly greater cardinality than O.
But we assumed that O contains all facts. Contradiction!

And Cantor's Theorem holds even for infinite sets! Q.E.D.

Did I just disprove God?

Comment by gilch on I would like to try double crux. · 2019-11-08T04:45:47.761Z · score: 1 (1 votes) · LW · GW

Wouldn't the prior probability of God to exist be a crux for you? I.e., if you change your prior probability from infinitesimal to somewhat not negligibly small, would it change your position? At least the infinitesimal probability is a crux for me.

Getting past an infinitesimal prior to a tiny finite one is a long way from "more likely than not".

But more simply, my prior is my position. If you get my prior belief for the proposition "God exists" over 50%, then you've won: at that point I've become a theist by definition (though maybe not a very confident one). This isn't a crux--It's the original proposition!

Comment by gilch on I would like to try double crux. · 2019-11-08T04:26:17.293Z · score: 1 (1 votes) · LW · GW

Yes. Omniscience is a crux for me.

OK. What does "omniscience" mean? The root words translate to something like "all knowing". But what is "all", and what is "knowing"? What's the minimum qualification? Each successive option seems harder to prove:

Option A: (sufficiently advanced aliens) God's knowledge isn't infinite or anything, just far beyond our current level. "Omniscience" is more metaphorical than literal.

Option B: (semi-omniscient simulator) God can look up any past event in the world simulation, but isn't simultaneously conscious of all of them and cannot predict the future short of actually simulating it. He does not know the all the logical implications of His knowledge and can be surprised by events. (Janet from The Good Place might be at this level.) Although perhaps he can rewind the simulation and try a different timeline, if He makes any changes, He can't always predict what would happen without actually trying it. He may also be ignorant of events in His native plane, outside of the world simulation.

Option C: (halting oracle of the first degree) God is a halting oracle machine able to solve the halting problem for any Turing machine, but is unable to solve the halting problem for halting oracle machines like Himself.

Option D: (higher-order halting oracle) God is a halting oracle machine able to solve the halting problem for any Turing machine, and halting oracle machines of some finite degree less than His own, but is unable to solve the halting problem for higher-order halting oracle machines like Himself, or those of any higher degree. There may possibly be beings of greater degree that know things God doesn't.

Options A, and maybe B seem at least possible, but very very far from proven. Option C seems unprovable using any finite amount of evidence, but probably has a logically coherent definition. Option D seems unprovable even with infinite evidence, but again seems coherent.

Or did you have some other option in mind? I don't know how to get past Option D without self-referential paradoxes invalidating the whole definition, but perhaps you have some new math for me?

Comment by gilch on I would like to try double crux. · 2019-11-08T03:44:41.885Z · score: 1 (1 votes) · LW · GW

I'm not sure this is part of the authoritative definition of doublecrux,

I'm not sure if anyone has an authoritative definition of doublecrux yet. But as this is my first real attempt at it, I appreciate guidance. We did open with the Litany of Tarski, but I might have lost sight of that for a moment. I maintain that I at least need to understand what my interlocutor is saying before I can conclude that he is right.

you try to persuade yourself that they're right and you're wrong ... (I don't think this is quite right, obviously the goal is for both of you to move towards the truth together ...

Again, the Litany of Tarski: If a God exists, I desire to believe that is the case. An update for either side is a victory. But the goal is not to fool myself or give up, or give in to confusion. The update must be an honest one, or the whole exercise is empty.

Comment by gilch on I would like to try double crux. · 2019-11-07T06:09:07.529Z · score: 1 (1 votes) · LW · GW

Looks like we started to go on circle

Yes, I just started to notice that after re-reading this thread. It seems like we're talking past each other without understanding. For Double Crux to work, we're not supposed to aim for direct persuasion until after we've identified the double crux, or we'll get "lost in the weeds" discussing the parts that aren't important to us. Have we found it yet? I think we have not, and that's what went wrong here.

I have yet to identify a single crux, but part of that might be because I don't understand your concept of God. I don't know what crux could possibly convince me your God exists, because I still don't know what "God" means (to you).

I'm honestly not that familiar with the Eastern Orthodox tradition. Protestant sects are more common in my country. The God concept worshiped by the average churchgoer here seems laughably naiive, and logically incoherent, but it does have some differences with what you've described so far. And the apologists, even in my country, seem to have a different definition that the average churchgoer (in my country), probably because the naiive definition is so indefensible. It's motte-and-bailey rhetoric--a combination of bait-and-switch with equivocation.

So I'll ask again: Is omniscience a crux for you? That is, if a source you would consider authoritative (the bishops, the Patriarch, archeology, visions from God, whatever it takes) explained to you that omniscience was not an attribute of God as He revealed Himself, but a later misrepresentation made by sinful philosophers, would you then say your God does not exist?

If you answer, "Then my God still exists and is not quite omniscient as I had once believed," then omniscience is not a necessary attribute for your God definition, and there is no need to discuss it further, because it is not a crux.

But, if you answer, "A 'God' that is not omniscient is no God of mine," then omniscience is a crux for you and we need to nail down what that means, because it might be closely related to a crux of mine.

Comment by gilch on Explaining Visual Thinking · 2019-11-05T02:46:39.887Z · score: 3 (2 votes) · LW · GW

Do you visualize the icosahedron as one object or do you split it up and consider each separately, but reminding oneself that it is actually one object?

I have looked at a d20 long enough and from enough angles (it's very symmetrical) to have memorized the whole icosahedron, and can visualize it that way, at least as an opaque object from the outside.

But the mnemonic technique of chunking is a valid strategy for visualization. Short-term memories must be "refreshed" or they fade away, but if you juggle too many at once, you'll drop one before you can get back to it. Making each face a chunk would be 20, which is too many. 3-5 chunks is a more reasonable number. My favored decomposition of the icosahedron is into a pentagonal antiprism with pentagonal pyramid caps. That's 3 chunks, and two of them are the same thing. Other decompositions may be useful depending on what you are trying to do.

More complex objects can be visualized as hierarchical decompositions, though not always in their entirety. Recognition is not the same as recall. The resolution of a weak visual memory may be just enough to recognize a new example (but too low to count the faces, say). A really low resolution image is more of a handle than a structure, but it can point you to the memory of the real thing.

Comment by gilch on I would like to try double crux. · 2019-11-04T04:34:21.334Z · score: 3 (2 votes) · LW · GW

Equivocation. The algorithmic (Kolmogorov) complexity cost of the conjunction of "simulated X" and Y is finite, but the "real X" is infinite, therefore, the former must be preferred by Occam's razor. "Simulated X" is a deception by aliens and is not a full halting oracle, but a finite approximation of one. It can't do everything the "real X" could.

I do not believe that aliens are performing miracles, just that that explanation is infinitely more probable on priors than an omniscient God. The miracles you have pointed to so far are best explained as natural accidents or hoaxes, not nearly enough evidence to even suggest aliens.

Comment by gilch on Explaining Visual Thinking · 2019-11-04T03:17:39.770Z · score: 1 (1 votes) · LW · GW

We know from IQ tests that working memory abilities vary. Those with aphantasia can't visualize at all, while others report that not only can they visualize a tiger, they can count its stripes. My visualizations are not that stable. The number of stripes would probably change as I attempt to count them. But visual thinking can be improved with practice, at least in my own experience. Things that took a lot of effort to visualize the first time become simple recall after that. The bigger your bag of tricks, the more likely you can find one that applies to a novel situation.

Visualizations need not be static images. They can have motion as well. I can rotate simple 3-D shapes in my mind, for example. Rotating a cube is pretty easy. I can even do an icosahedron, though that one took some practice. But counting the leaves on a tree would be too difficult, never mind rotating the tree without changing (or glossing over) their number. There are limits to the resolution. You can also do transformations other than rotations, like scales, shears, extrusions, etc. These visualizations are useful in computer graphics and in topology.

In the case of mathematics, I find visualization most useful for generating examples, especially counterexamples. Using the visual query process I described, one can try to query for a shape that meets certain constraints. Sometimes one example (or counterexample) is all it takes to prove a theorem. Sometimes the query produces the example, but sometimes it fails to meet all the constraints and I have to query that part again. Pointing out the part that failed a constraint can bring more examples to mind. You have to give these mathematical objects a visual form to gain the benefits of visual thinking, but there are many morphisms one might try. Besides single examples, you might also be able to enumerate a set of them, or notice a pattern that can be repeated to infinity.

I can generate candidate visualizations much faster in my head than I can draw them on a whiteboard, but then communicating that insight to another person may require a diagram.

A visual thinking riddle: go in one hole and come out three. What am I? I solved this one visually pretty quickly. Try to generate candidate visualizations and see if you recognize the shape.
(Answer: grrfuveg.)

Comment by gilch on Multiple Moralities · 2019-11-03T20:52:07.331Z · score: 1 (1 votes) · LW · GW

You can't persuade a rock that cheesecake tastes better than dirt, no matter how clever your arguments. Not every possible mind might agree with us, even in principle; there are No Universally Compelling Arguments.

We are born already in motion with godshatter instincts suited to our stone-age ancestral environment of evolutionary adaptation. This part is not simply arbitrary: it had the requisite survival value, or we would not be here talking about it. We are then acculturated to our surrounding society. This part is learned, but not entirely arbitrary either, because culture itself is evolving and subject to selection pressures.

To whatever extent we judge our society to be suboptimal, we must use our evolved/acculturated minds to do it. What else could we use? But there are reasons we are the way we are, principles that are not simply random.

Comment by gilch on I would like to try double crux. · 2019-11-02T18:55:24.120Z · score: 1 (1 votes) · LW · GW

because of the amount and the longitude. Everything that could been inside should have gone away.

I don't think that word means what you think it means. Did you mean "latitude"? Hawaii has a more tropical climate because it is near the equator, not because it's near the prime meridian. Or did you mean "longevity"? Either way, you can't expect all oils to evaporate quickly like water, even in a warm or arid climate. It's true that there are some oils used in paint that dry quickly, but there are other oils that do not dry out, even after a long time.

The previous person was killed and tortured, the icon disappeared. The murderers were not found. It would be quite crazy idea, knowing this story, to make this mystification. To put your life under risk for what? For stupid hoax? You must be crazy to do it.

Or greedy. How about for money? Fame? There are known examples of hoaxes with motives like these.

And this person (current keeper) serves in police

A policeman might be even less afraid of criminals trying to kill him, because he already has to deal with them.

If I would need to do such a hoax, I would put some source of myrrh inside, and refill it periodically. It can be done of course. I could even believe that it can be done such that observer, taking the icon, would not notice any difference from the usual icon. But is it possible to avoid X-rays somehow? They travel by plane, I bet they do not put icon into luggage (it is too precious). So they must go with it as hand luggage. There the custom, using X-rays, observe small vessels inside the icons and asks what is it. And it is done, the hoax is over.

If I needed to perform such a hoax, I wouldn't have to modify the icon at all. The Roman Catholics tend to use statues instead of icons, and they have examples of those weeping too. I heard of one case where someone was caught applying the "tears" with a squirt gun. You don't need to carve channels or secret compartments for a hoax. It's enough to have a spray bottle and vegetable oil, which are available for purchase pretty much anywhere. And if you use a non-drying oil, you don't even have to re-apply it! It will stay "wet" long after water would have dried out.

-different myrrh-streaming icons, as long as it passed the check by church officials not only on the local level

I don't trust church officials any further than I can throw them. They have a fundamental conflict of interest. Their loyalty is to the Church, not the truth. The fact that someone is a priest makes me even less likely to trust them. The Roman Catholics have been rocked by child sexual abuse scandals which have been all over the news in this country, and worse, they attempted to cover them up to maintain the Church's reputation. A cursory web search reveals the Eastern Orthodox seem to have similar problems for similar reasons. The Churches cannot be trusted to be honest with us.

These officials are clearly not rationalists (if they were, I would not expect them to be religious!) I don't expect most of them to even be scientists. But even if they were, scientists can still be deceived.

Comment by gilch on I would like to try double crux. · 2019-11-02T18:14:21.604Z · score: 1 (1 votes) · LW · GW

Yes, MWI still has indexical uncertainty. This is a property of the observer, not the universe, which remains deterministic. But you can still simulate the wavefunction on a Turing machine and use it to make predictions, which was my point. It's in the space of hypotheses of Solomonoff induction.

I don't really prefer non-local theory, but the laws of nature are what they are and don't care what I want.

Comment by gilch on I would like to try double crux. · 2019-11-02T18:07:53.307Z · score: 1 (1 votes) · LW · GW

No. We use the approximation, and approximation has the same size for both of them (we consider the case of comparing hypothesis "There is a God with such and such attributes" and " There are aliens forge us to believe that there is a God with such and such attributes"). The algorithm of construction of this approximation, though, is simpler for pure God's hypothesis (using the mere fact of its existence, not formulating the hypothesis itself, like we establish dualities between different types of string theories using that M-theory exists but without formulating it) since it does not require transitional link of "hidden aliens".

I'm not understanding this part. If we already assume that aliens and God exist (which is not allowed because it's begging the question) then of course it's simpler to assume God explains the evidence than to introduce the additional hypothesis that the aliens are also trying to fool us.

But without committing the fallacy of begging the question, we are left with the conjunctive hypothesis of "aliens exist" and "they are trying to fool us" that dominates "there is an omniscient being" (which must have an infinitesimal prior), never mind all the other attributes of your particular God.

Comment by gilch on I would like to try double crux. · 2019-11-02T17:59:57.402Z · score: 1 (1 votes) · LW · GW

I tell you soon after the discovery of muon that there is another particle, like the electron

It took a great deal of evidence to nail down both the existence of new particles and their properties to that degree of precision. It's already strong enough to overcome a low prior, but due to mathematical symmetries in nature, some particles were even predicted in advance of experimental discovery. In other words, they had a high prior given what was known, which is why scientists were willing to go to the great expense of looking for them.

We do not have any strong evidence for God, and assuming omniscience alone gives Him an infinitesimal prior, which means no amount of evidence could ever be enough.

Comment by gilch on Explaining Visual Thinking · 2019-11-02T01:40:26.197Z · score: 12 (8 votes) · LW · GW

Are you (or is your student) claiming to have aphantasia? If so, you'd have to use something external to hold the image.

Visual thinking is like other kinds of thinking, but using the visual sensory channel. It's a kind of short-term memory with limited capacity.

If you remember the days when we had to look up phone numbers in books, you might have had the experience of loading a seven-digit number into your short-term memory as an auditory loop, speaking it over and over again in your mind's voice (and hearing it in your mind's ear), long enough to dial on the phone. You could actually speak the numbers aloud for a similar effect, but most people don't have to do this to "hear" it. But a moment of distraction, especially an auditory distraction, can cause you to lose that memory.

Visual thinking is the same, but you use your mind's eye instead of ear. It's like having an imaginary whiteboard, but it requires concentration to hold an image, and a moment of distraction can erase it. You're not exactly drawing in lines either (unless you choose to). This whiteboard has limited capacity in the same way your auditory loop has a limited duration. You can choose which parts of the image to focus on, and that can have more detail (just like your real eyes can see things they're looking at directly better than things in the periphery), but if you stop focusing on an area for too long its detail fades and you lose the memory. If you try to exceed your capacity, then the part of the image you refresh might not quite have what was there before, the same way you can accidentally remember the wrong number if you try to keep too long of a string of digits in your auditory memory.

In the same way you can query your long-term memory for the sound of something, by holding the question in your mind until the sound arises (E.g. What rhymes with "muffin"? Or what does a dolphin sound like?), you can query your long-term memory for an image (E.g. What does a dolphin look like?).

And this can be done without using your mind's voice. Perhaps you can use your mind's eye to "read" the text of the question. But you don't even have to use words. You can use the concepts behind the words more directly. And sometimes these concepts are visual, or can have a visual representation. Even kinesthetic concepts have enough of a spacial component that they can be diagrammed visually in a very natural way. Many nouns correspond to visible things, and verbs correspond to visible actions. Holding such an image in memory along with the intention to query memory can cause an answer to arise to consciousness, the same way a query in words can.

The intention to query for a sound is like stopping for a moment to listen, while the intention to query for an image is like stopping to look. You have to make a space for it in your mind in the appropriate sensory channel.

You can also query your subconscious for things you've never seen before. Maybe you don't know what an ichthyosaur looks like, but someone tells you it looks like a dolphin, but with two pairs of fins instead of one. Running a hypothetical query may be enough to produce the image in your mind's eye. If that's not working, you can try to force it by "painting", which takes more effort: If you had a whiteboard (and artistic skill) you could draw a dolphin, and then draw another pair of fins in the pelvic region. You can do the same thing in the mind's eye. Start with a dolphin image, and then while concentrating to keep it refreshed in memory, make a change to it: refresh that part differently on purpose.

Comment by gilch on I would like to try double crux. · 2019-10-31T04:28:34.239Z · score: 1 (1 votes) · LW · GW

Thus, the fact that we need to throw to the trash can all our current assumptions and build the theory based on new assumptions does not mean that we should put small probability to this new theory and need hidden parameters or hidden aliens given the same observations.

The alien hypothesis dominates the God hypothesis, because God is infinitely improbable, but aliens are only finitely improbable.

However, this approximate version of the whole hypothesis can be short enough to compete with no God hypothesis (remember, I was talking about the width of the function? ).

You seem to be arguing that we can bias our prior to accept an approximate God at the very edge of the "width". I say the rights of Mor­timer Q. Sn­od­grass are being violated.

Why your God,

Rather than Allah, the Fly­ing Spaghetti Mon­ster, or a trillion other gods no less com­pli­cated—never mind the space of nat­u­ral­is­tic ex­pla­na­tions![?]

Comment by gilch on I would like to try double crux. · 2019-10-31T04:00:38.495Z · score: 1 (1 votes) · LW · GW

QM is fundamentally probabilistic.

No. That is not fundamental at all. Bell's Theorem only rules out local hidden variables. The Many-Worlds Interpretation and De Broglie–Bohm interpretation are deterministic.

Comment by gilch on I would like to try double crux. · 2019-10-31T03:51:50.213Z · score: 1 (1 votes) · LW · GW

I think you completely missed my point about the toast. I was trying to be humorous by referencing an actual case, but one that I found especially silly.

It's just pareidolia. It's the same as seeing animals in clouds. But which animal you see depends on which animals you're familiar with. Toast patterns are noisy, so are clouds. The human perceptual system is constantly trying to recognize what it knows in what it sees, and seems particularly good at finding faces. And we have a pretty good idea how this works. See DeepDream.

Yes, I can see that the pattern resembles a human face, and a feminine one. But I personally think that the toast looks more like Abby Sciuto from NCIS than most Virgin Mary paintings.

Comment by gilch on I would like to try double crux. · 2019-10-29T23:38:09.677Z · score: 1 (1 votes) · LW · GW

Sorry, what's a MEU?

Comment by gilch on I would like to try double crux. · 2019-10-29T03:50:28.992Z · score: 1 (1 votes) · LW · GW

Hmmm. Indeed, you are totally right here. I actually never thought that incomprehensibility is directly connected with the omniscience. Thank you very much for this, it make me to reconsider a lot of things.

An update of beliefs! We are making progress.

However, this approximate version of the whole hypothesis can be short enough to compete with no God hypothesis

So are you weakening the original claim? You are no longer trying to persuade me of an omniscient being, but only a sufficiently knowledgeable one?

What is the difference between such approximation and the same approximation for alien teens etc? Why would we prefer the God hypothesis to the alien teens?

Yeah, at this point, I think we may be talking about aliens, not God, but we're going to use your definitions of the terms. I personally wouldn't expect omniscience of a small-g "god".

Well, because to say: "there is a God with such and such attributes" is simpler than to say "there are alien teens who form around us a reality such that it looks like that there is a God with such and such attributes".

I don't really agree with that, and here is an illustration of why:

Suppose I tell you that I have an aunt that owns a dog. I think most people would just believe me. Aunts are not at all rare, and neither are dogs. Maybe I could be lying to prove a point, but dogs are so common, that I probably could have picked another relative with no need to lie about it.

Now, suppose I tell you that I have an uncle who owns a tiger. I think most people would not just believe that easily. There certainly are people who own tigers though. So maybe you'd be persuaded with a little more proof. Maybe I could show you a picture. That might help until you realize that you've only ever met me online, and have no idea what I look like. Maybe I'm not the man in the photo (I could be a woman for all you know), and maybe the owner is not my uncle. Maybe I could do a video chat with you and you could see I have the same face. That would help, but maybe I used Photoshop on the tiger picture to insert my face. At some point though, the evidence would be good enough, or you'd call my bluff.

Now, suppose I tell you that I have a nephew who has a pet purple martian dragon. Your first impression might be, is that a Pokemon? A toy? (Even understanding what another person is saying requires some shared priors.) "No, I mean it's literally an alien creature from Mars," I say. Did he tell you that? Kids have wild imaginations. "No, no, I saw it." OK, we know life exists on Earth, there's no physical reason why it couldn't exist on other planets. It's not outside the realm of possibility, but you're going to need a lot more evidence than for the tiger.

Now, suppose I tell you that I have a niece with a pet genie. He can turn invisible and follows her to school. She gives him lamp rubs and sometimes he grants her minor wishes using magical powers when he's in a good mood. Does this seem more or less likely than the purple dragon? The dragon is at least compatible with what we know of science. Magical powers, not so much.

The above stories are an illustration of how you, or people in general are already using priors. The lower the prior, the more evidence is required to overcome that prior.

Now suppose I tell you that I have an internet acquaintance who has an invisible friend named Steve, whom he communicates with via mental telepathy, although Steve seems oddly reluctant to answer sometimes. Steve has phenomenal cosmic magical powers and can rearrange stars and stuff. "Have you ever seen Steve do this?", we ask. "No, but I've seen him remotely draw pictures of his mother on toast." I don't know about you, but I'm gonna need a little more proof than that. Right? Does this sound more or less likely than my niece's genie? Even the genie could explain the toast. Not only is Steve invisible, he has stronger magic? Wouldn't we need at least as much evidence as for the genie? Or for the dragon? The tiger?

Oops, hold on. My acquaintance tells me I got the name wrong it--was a glitch in Google Translate. His real name is not Steve--it's Jesus. (. . .) I guess that settles it.

But why do we need to say that there is the omniscient God at all if all we are going to do is to use approximations? Well, let me give you an analogy from mathematical physics.

Past some point these cases are indistinguishable with the finite amount of available evidence anyway, so I would argue that the difference is meaningless, at least from the perspective of induction on evidence: it makes no difference to the resulting predictions.

However, the difference may still matter to arguments of logical necessity, and if your faith has some creed that cares about the distinction, the weakened definition of God may still be a problem for you.

Comment by gilch on I would like to try double crux. · 2019-10-29T02:27:29.102Z · score: 1 (1 votes) · LW · GW

Not exactly. First, I can predict that if I throw the stone it will fall down and stuff like that. A miracle may happen, but the probability for it to happen from nowhere is very small (also not zero).

OK, so when things behave as normally expected, that's just laws of nature, but whenever you're surprised we can blame it on the witch?

So, first, such theory can predict something (like myrrh-streaming icon mentioned above, or healing, or answer to the prayers). Second, the predictions do not always coincide with no God theory predictions

This point is very important: The theory must make predictions to be knowledge--if your theory is equally good at explaining anything (like the witch), then you have zero knowledge, because it fails to constrain anticipation.

The sword of prediction cuts both ways:

the [advance] ex­pec­ta­tion of the pos­te­rior prob­a­bil­ity, af­ter view­ing the ev­i­dence, must equal the prior prob­a­bil­ity. ... If you ex­pect a strong prob­a­bil­ity of see­ing weak ev­i­dence in one di­rec­tion, it must be bal­anced by a weak ex­pec­ta­tion of see­ing strong ev­i­dence in the other di­rec­tion. If you’re very con­fi­dent in your the­ory, and there­fore an­ti­ci­pate see­ing an out­come that matches your hy­poth­e­sis, this can only provide a very small in­cre­ment to your be­lief (it is already close to 1); but the un­ex­pected failure of your pre­dic­tion would (and must) deal your con­fi­dence a huge blow.

In other words, how strong a piece of evidence should appear to you, depends on your priors; strength is not a property of the evidence alone. If you are claiming that your God hypothesis is not equally good at explaining anything (like the witch), and if you are (rationally) very confident that God exists, then you must have a weak expectation of seeing strong evidence the other way. That's a crux, right? What would be a big surprise to your theory?

A corollary to conservation of evidence: Absence of evidence is evidence of absence, when the observation would be expected.

Comment by gilch on Artificial general intelligence is here, and it's useless · 2019-10-27T19:33:37.540Z · score: 1 (5 votes) · LW · GW

You're fighting a strawman, George. You clearly do not understand our real arguments. Attempts to point this out have only been met with your hostility. I do not have the patience to tutor one so unwilling to study.

If you have any desire to cross the inferential gap, I will refer you to the LessWrong FAQ

If your post involves topics that were already covered in the sequences you should build on them, not repeat what has already been said. If your post makes mistakes that were warned against in the sequences, you'll likely be downvoted and directed to the sequence in question.

That is exactly what is happening here. The symptoms of this dialogue are diagnostic of an inferential gap. Your case is not the first. Read the Sequences, George. Especially the parts we've linked you to.

On the other hand, we're well aware that it can take a long time to read through several years worth of blog posts, so we've labeled the most important as "core sequences". Looking through the core sequences should be enough preparation for most of the discussions that take place here. We do recommend that you eventually read them all, but you can take your time getting through them as you participate. Before discussing a specific topic, consider looking to see if if there is any obvious sequence on that topic.

Comment by gilch on I would like to try double crux. · 2019-10-27T19:14:33.393Z · score: 1 (1 votes) · LW · GW

I find the MUH philosophically dubious. I also disagree with Wikipedia's characterization of the CUH as adding an additional hypothesis on top of MUH (I'm not sure if that's how Tegmark sees it, or if that was just an interpolation by the editor). Instead, the CUH is throwing out the dubious axiom that allows things like uncomputable sets to exist, which means by Occam's razor, I think the CUH is the simpler hypothesis. I don't exactly buy the CUH either, but I don't have a better idea.

How is that relevant? It is perfectly possible for a mathematical universe to be a form of Platonic realism.

I disagree with your interpretation of "perfectly possible", but even if I hypothetically grant you that a halting oracle exists, how can an agent ever be rationally justified in believing that it does? It's something that takes an infinite amount of evidence to prove. The method clearly can't be induction.

Comment by gilch on I would like to try double crux. · 2019-10-26T21:20:08.273Z · score: 3 (2 votes) · LW · GW

I have yet to find a crux for myself, much less a double-crux, but not for lack of trying. I have maybe identified co-cruxes, that is, obstacles that would be necessary to overcome for me to believe in God as defined, but which individually proving would not be sufficient to prove God to me. Perhaps that is progress.

Trying to summarize,

I don't think we've exactly nailed down what "God" means, but I think establishing at least the possibility of each of the necessary attributes of God would be a co-crux (because God as defined must have all of them). Necessary attributes we've identified so far include

  • "omnipotence" (in the limited sense of being able to alter a simulation, and actually doing so),
  • being a "mind" rather than a simple law of nature, and
  • "omnibenevolence", which runs into the problem of evil, and the problem of defining evil.

I haven't heard back enough on "omniscience" to know what definition we're using, but I am concerned that "infinite knowledge" must exclude God a priori from the hypothesis space of any kind of induction, which rules out using evidence to prove God exists, because no finite amount of evidence can ever be enough to prove an infinite amount of knowledge. That may still allow some kind of deductive logical necessity argument though. And perhaps it can be weakened somehow as we have done for "omnipotence".

Am I mistaken in my approach here? This is my first real attempt at Double Crux. Are these co-cruxes pointing to a crux I have that I'm not seeing?

I'm not sure if we can find a double crux before we first agree on questions of basic epistemology. My interlocutor is studying the topic as I bring points up, but it does take time.

Comment by gilch on Artificial general intelligence is here, and it's useless · 2019-10-26T20:17:44.976Z · score: 3 (2 votes) · LW · GW

My argument here is that it doesn't, it's an empty formalism with no practical application. ... There are literal physical limitations to computational resources.

Where do I even start with this? That argument proves too much. You could apply the same argument to engineering in general. "Well, it would take infinite computing power to sum up an integral, so I guess we can't ever use numerical approximations." Please read through An Intuitive Explanation of Solomonoff Induction. In particular, I will highlight:

But we can find short­cuts. Sup­pose you know that the ex­act recipe for bak­ing a cake asks you to count out one molecule of H2O at a time un­til you have ex­actly 0.5 cups of wa­ter. If you did that, you might not finish the cake be­fore the heat death of the uni­verse. But you could ap­prox­i­mate that part of the recipe by mea­sur­ing out some­thing very close to 0.5 cups of wa­ter, and you’d prob­a­bly still end up with a pretty good cake.

Similarly, once we know the ex­act recipe for find­ing truth, we can try to ap­prox­i­mate it in a way that al­lows us to finish all the steps some­time be­fore the sun burns out.

Comment by gilch on What video games are more famous in our community than in the general public? · 2019-10-26T20:05:40.872Z · score: 5 (4 votes) · LW · GW

I'm not exactly sure what the downvotes are for, but this question seems difficult to answer without a survey. We'd have to know both what games are popular in the general public and in this community.

I can point out a few games of interest to rationalists (in my opinion) though.

Morrowind has crafting mechanics that let you implement a kind of intelligence explosion: drinking intellect enhancing potions lets you craft better potions. Lather, rise, repeat... You can boost your abilities far enough that, for example, you can craft a dagger strong enough to slay a dragon in one hit. (This might also work in other Elder Scrolls games. You can do something similar in Skyrim too, I think.)

Arimaa is more of a board game (thought it has computer implementations). It was designed as a toy problem for AI, intermediate in difficulty between chess and go due to its relatively high per-move branching factor. It's played with a chess set, so you probably have the equipment you need already. The high branching factor gives the game a very fluid feel that's quite different from chess. This seems less important now that we have Alpha Go, but it's still a fun game in its own right.

Poker is more of a card game, but again has computer implementations. Playing poker well requires you to make probabilistic judgements under uncertainty. It's a kind of applied rationality, and can be profitable too, but only if you're more rational than your opponents.

The Credence Calibration Game by CFAR. This is intended as more of a training system than a game per se, but it's a computerized game nonetheless.

Dual n-back. Supposedly trains your brain. This game is really freaking hard and maybe not enough fun to be worth the effort given its paltry benefits.

Magic: The Gathering. Again, it's a card game, but there are computer implementations. It's one of the games I recall being discussed here on LessWrong, but it's also quite successful with the public.

Dungeons and Dragons. Again, it doesn't require a computer, but there are computer games based on it (basically all RPGs, yes, but also some that use the actual D&D mechanics). Munchkin exploits are interesting because we rationalists are also looking for that kind of thing in the real world. It's also a source of a lot of nerdy lore, some of which gets talked about here.

Speaking for myself, I'm a fan of indie games. There's a lot of crap, but the ones that stand out do so because they're good, not because they're marketed. They also tend to be cheap: you can usually get several for a dollar in the Humble Indie Bundles, so you can enjoy a variety. One example that I liked was Reassembly, which puts an emphasis on creativity.

I've also started playing VR games. Wands is great.

Comment by gilch on I would like to try double crux. · 2019-10-26T18:11:57.600Z · score: 1 (1 votes) · LW · GW

You can't prove a halting oracle exists inductively. How could you? Solomonoff induction is doing induction perfectly, and a halting oracle is not even in the hypothesis space, because the space contains only computable functions, and the halting problem is not decidable. And even if it were, what use would that hypothesis be to you? You can't get any predictions from running a program on a halting oracle machine when you don't even have one.

From the Wikipedia article:

Tegmark claims that the hypothesis has no free parameters and is not observationally ruled out. Thus, he reasons, it is preferred over other theories-of-everything by Occam's Razor. Tegmark also considers augmenting the MUH with a second assumption, the computable universe hypothesis (CUH), which says that the mathematical structure that is our external physical reality is defined by computable functions.

So this is a point that Tegmark himself considers fair. The CUH would not have a halting oracle.

Comment by gilch on Artificial general intelligence is here, and it's useless · 2019-10-26T18:01:02.766Z · score: 3 (2 votes) · LW · GW

Optimising for one thing is pretty automatically optimising against other things,


but AI theorists need a concept of general intelligence -- optimising accross the board.

No. There is no optimizing across the board, as you just stated. Optimizing by one standard is optimizing against the inverse of that standard.

But AIXI can optimize by any standard we please, by setting its reward function to reflect that standard. That's what we mean by an AI being "general" instead of "domain specific" (or "applied", "narrow", or "weak" AI). It can learn to optimize any goal. I would be willing to call an AI "general" if it's at least as general as humans are.

Comment by gilch on Artificial general intelligence is here, and it's useless · 2019-10-26T17:35:01.964Z · score: 3 (2 votes) · LW · GW

This is the point where I start to think that, although we both seem to speak English, we certainly understand fully different things from some words and turns of phrase.

Why bring up AIXI if you yourself admit you are not going to defend the approach as a good path to AGI ?

Implementing true AIXI is not physically possible. It's uncomputable. I did not say that "AIXI" is not a good path to AGI. I said that a Monte Carlo approximation of AIXI is not a good path.

And it's not that it can't work, clearly it does (and indeed any intelligent agent, humans included, is going to be some kind of an approximation of AIXI) but that the Monte Carlo AIXI has certain problems that make the approach not good:

  1. It's not efficient; other approximations besides Monte Carlo AIXI are probably easier to get good performance out of, the human brain being one such example.
  2. Direct attempts at approximation will run into the wireheading problem. Any useful agent with physical influence over its own brain can run into this issue unless it is specifically dealt with. (AIXI's brain is not even in the same universe it acts on.)
  3. The Sorcerer's Apprentice problem: any useful reward function we can come up with seems to result in an agent that is very dangerous to humans. The AIXI paper takes this function as a given without explaining how to do it safely. This is Bostrom's orthogonality thesis: a general optimizer can be set to optimize pretty much anything. It's not going to have anything like a conscience unless we program that in, and we don't know how to do that yet. If we figure out AGI before we figure out Friendly AGI, we're dead. Or worse.

From the Aribital article:

AIXI is the perfect rolling sphere of advanced agent theory - it's not realistic, but you can't understand more complicated scenarios if you can't envision the rolling sphere.

Comment by gilch on Artificial general intelligence is here, and it's useless · 2019-10-26T16:32:52.245Z · score: 3 (2 votes) · LW · GW

Think of “possible” here as meaning “possible to formalize”

Why not say “possible to formalize”, if that is what is meant?

I DID say that, and you just quoted it! I didn't say it in advance, and could not have been reasonably expected to, because it was obvious: AIXI is based on Solomonoff induction, which is not computable. No-one is claiming that it is, so stop putting words in my mouth. I cannot hope to predict all possible misinterpretations of what I say in advance (and even if I could, the resulting advance corrections would make the text too long for anyone to bother reading), but can only correct them as they are revealed.

If AIXI, is possible, the "Artificial general optimisation" would make sense. Since it is not possible, the use of optimisation to clarify the meaning of "intelligence" leaves AGI=AGO as contradictory concept, like a square circle.

No, that does not follow. Optimization does not have to be perfect to count as optimization. We're not claiming that AGI will be as smart as AIXI, just smart enough to be dangerous to humans.

Which brings in the the other problem with AIXI. Humans can model themselves, which is doing something AIXI cannot.

AIXI can model any computable agent. It can't model itself except approximately (because it's not computable), but again, humans are no better: We cannot model other humans, or even ourselves, perfectly. A related issue: AIXI is dualistic in the sense that its brain does not inhabit the same universe it can act upon. An agent implemented on real physics would have to be able to deal with this problem once it has access to its own brain, or it may simply wirehead itself. But this is an implementation detail, and not an obstacle to using AIXI as the definition of "the most intelligent unbiased agent possible".

Comment by gilch on I would like to try double crux. · 2019-10-26T04:49:17.548Z · score: 1 (1 votes) · LW · GW

I'm not sure if this is all correct. What you're describing doesn't exactly sound like Solomonoff induction, but you do seem to have a grasp of the principles involved.

Solomonoff induction does not discard any program that is consistent with the observation so far. But for any observation string there are an infinite number of programs that produce that string. There is a shortest one, and infinite class of that program prefixed by some whole number of no-operations (computations that undo themselves). And compilers implementing that same program in encodings of other programming languages. And interpreters implementing that same program in encodings of other programming languages. And entire universes containing people who happen to be simulating one of these (which may be considered an unreliable type of interpreter). And arbitrary nestings of any of the above any number of times. None of this is discarded. But again, the set is infinite, so no matter what distribution you choose, the probabilities after some point must decrease for it to converge.

The point about the witch isn't that "witch" is a complex cost to encode in a program (although it is), but that "she did it" fails to compress the data at all, because you still have to encode what the pronoun "it" is referring to. Because a "witch" can be blamed for literally anything, adding a "witch" to the uncompressed hypothesis "it" adds no predictive power whatsoever. (If you can compress "it" some other way, then you can make predictions without the witch and she is useless to your model.)

God, who can likewise be credited for anything (even what looks like evil--"all part of God's plan", or "God works in mysterious ways", right?) is the same as the witch: no predictive power over "it" whatsoever. And worse, God's complexity cost is not just relatively big like any intelligent mind (such as the witch) would be, but literally infinite if we say that God is omniscient: If God is a "halting oracle", then God is not even contained in the set of all computer programs, because He is not computable: He can't even be a hypothesis, only approximated. And to get a better approximation, you must use a longer computer program that encodes more of Chaitin's constant, which is provably not compressible by any halting program. Better approximations of God get bigger without limit. The approximate God hypothesis has literally infinitesimal probability--you can't escape it: The better the approximation gets, the less likely it is.

And the true God hypothesis is not even in the running. It literally cannot be proved by induction at all. Nor can you take God as an axiom. (I will dismiss it as the fallacy of special pleading: applying this privilege lets us prove anything, even false gods.) The only hope then is proving deductively from some logical necessity, or giving up on omniscience as defined, which of course, opens the possibility of there being beings greater than whichever God you choose.