Posts
Comments
Which (possibly all) of the VNM axioms do you think are not appropriate as part of a formulation of rational behavior?
I think the Peano natural numbers is a reasonable model for the number of steins I own (with the possible exception that if my steins fill up the universe a successor number of steins might not exist). But I don't think the Peano axioms are a good model for how much beer I drink. It is not the case that all quantities of beer can be expressed as successors to 0 beer, so beer does not follow the axiom of induction.
I think ZFC axioms are a poor model of impressionist paintings. For example, it is not the case that for every impressionist paintings x and y, there exists an impressionist painting that contains both x and y. Therefore impressionist paintings violate the axiom of pairing.
Can you by chance pin down your disagreement to a particular axiom? You're modus tollensing where I expected you would modus ponens.
I didn't follow everything, but does this attempt to address self-fulfilling prophecies? Assume the oracle has good track record and releases its information publicly. If I ask it "What are the chances Russia and the US will engage in nuclear war in the next 6 months?", answers of "0.001" and "0.8" are probably both accurate.
What what sorts of output strings are you missing?
Calculating Kolmogorov complexities is hard because because it is hard differentiate between programs that run a long time and halt and programs that run a long time and never halt.
If God gave you a 1.01 MB text file and told you "This program computes BB(1000000)", then you could easily write a program to find the Kolmogorov complexity of any string less then 1 MB.
kolmogorov_map = defaultdict(lambda x : infinity)
for all strings *p* less than 1000000:
run *p* for at most BB(1000000) steps
save output to *o*
if (*p* halted and kolmogorov_map[*o*] > len(p):
kolmogorov_map[*o*] = len(p) # found smaller program
else:
# *p* does not ever ever halt and has no output
Replace BB(1000000) with a smaller number, say A(Graham's number, Graham's number), and this calculator works for all programs which halt in less than A(Graham's number, Graham's number) steps. That includes pretty much every program I care about! For instance, it includes every program which could run under known physics in the age of the universe.
Eh, don't take it personally. I'm guessing commenters are implicitly taking the title question as a challenge and are pouncing to poke holes in your argument. I thought your essay was well written and thought provoking. Keep posting!
Don't know, not the original author. What do you think the chances are than an email on the third page of your inbox will ever get a reply? Inbox purgatory seems to me like a way to give up on something without having to admit it yourself.
If my inbox has more than 40 or 50 items in it I feel demoralized and find it harder to work through newer items, so the easiest way for me to stay at steady-state is to keep my inbox at zero or close to it.
Counterpoint: I've kept to an empty inbox for many years, but know people with ever-growing inboxes whom I consider more organized and responsive. I've never declared email bankruptcy during my professional life and don't know the consequences.
And nothing in here says anything about how to deal with that situation.
I read the advice as:
If you still have unresolved emails from 2015 in your inbox then keeping emails in your inbox isn't causing them to get resolved. Accept that, get a clean slate, and move on.
Make a folder called "old inbox" and put all your old emails there. Now you have an empty inbox! The costs of putting your old emails out of sight are less than the benefits of keeping an empty inbox going forward.
HLS students of any skin color have high IQs as measured by standardized tests. The school's 25th percentile LSAT score is 170, which is 97.5th percentile for the subset of college graduates who take the LSAT. 44% of HLS students are people of color.
The book to read is Reasons and Persons by Derek Parfit.
If love your simulation as you love yourself, they will love you as they love themselves (and if you don't, they won't). You can choose to have enemies or allies with your own actions.
You and a thousand simulations of you play a game where pressing a button gives the presser $500 but takes $1 from each of the other players. Do you press the button?
What do you mean by "commit suicide" here? Memorize the results of 5 more coins?
Spit balling hacks around this:
Weigh hypotheses based on how many steps it takes for them to be computed on a dovetailing of all Turing machines. This would probably put too much weight on programs that are large but fast to compute.
Weigh hypotheses on how much space it takes to compute them. So dovetail all turing machines of size up to n limited to n bits of space for at most 2^n steps. This has the nice property that the prior is, like the hypotheses, space limited (using about twice as much space as the hypothesis).
Find some quantum algorithm that uses n^k qubits and polynomial time to magically evaluate all programs of length of n in some semi-reasonable way. If such a beast exists (which I doubt), it has the nice property that it "considers" all reasonably sized hypotheses yet runs in polynomial space and time.
Given n bits of evidence about the universe, consider all programs of length up to k*log(n) run for at most n^k2 steps. This has the nice property that it runs in polynomial time and space.
I also found the answer to a question I've been researching for ~3 years.
Boy, did you ever! Congratulations!
I'm not sure if coin flips are quantumly random, or just hard enough to predict. Feels like coins would still work as well in a Newtonian universe. I tried to go with something that something that is clearly caused by quantum effects, like measuring if electron is either polarized up or down or down. Luckily, there's an app for that.
I set up an experiment to test quantum anthropics.
Flip four quantum coins. If they all came up heads, stop. If any of them came up tails, flip 5 more coins and (using mnemonics) think really hard about the exact coin flip sequence. If I find myself in a universe where first four coins came up all heads, then with p < 0.0625, quantum weirdness kept me from finding myself in one of the universes the state of my consciousness split me 512-ways.
I got access to a quantum random number generator, resolved to do the experiment, called a friend and told them I was about to do the experiment, and... chickened out and didn't do the experiment.
I do not know how to interpret these results :-/
Minor naming feedback. You switched from calling something "supervised learning" to "reinforcement learning". The first images that come to my mind when I hear "reinforcement learning" are TD-Gammon and reward signals. So, when I read "reinforcement learning", I first think of a computer getting smarter through iterative navel-gazing, then think of a computer trying to wirehead itself, then stumble to the meaning I think you intend. I am a lay reader.
Other answers I've considered:
o) Simpler universes are more likely, but complicated universes vastly outnumber simple ones. It's rare to be at the mode, even though the mode is the most common place to be.
p) Beings in simple universes don't ask this question because their universe is simple. We are asking this question, therefore we are not in a simple universe.
2') You don't spend time pondering questions you can quickly answer. If you discover yourself thinking about a philosophy problem, you should expect to be on the stupider end of entities capable of thinking about that problem.
Oh! So you're saying the spectrum of the acoustic noise at a given temperature will be the spectrum of black body radiation! Yes, I could definitely believe that. That is high-frequency indeed.
Essentially, an air molecule doesn't have enough energy to register at your hearing sensors, that is, to move your eardrum (or cochlear hairs).
Though, now that I'm thinking about it, if the white noise generator I bought to help me sleep is really good at producing white noise with uniform power at high enough frequencies, an air molecule would have enough energy to move my eardrums. I would also be on fire.
And if my white noise generator is really really good at producing white noise with power uniform across all frequencies, the noise's mass-energy will cause my bedroom to collapse into a black hole and I will be unable to leave a 5 star review on Amazon.
Do you happen know a back-of-the-envelope way to get that 30 THz figure?
See my edit
Why can I hear noise (white noise / pink noise / brown noise), but not hear temperatures?
EDIT FOR CLARIFICATION Air temperature is caused by air molecules moving randomly at high speed, white noise is caused by air molecules moving randomly at high speed, what's the difference? Why does white noise fill the room with sound instead of just raising the temperature slightly?
My hand-wavy-sounds-like-science-technobable guess is that temperature does fill the air with sound, but most of the energy of that sound is at far too high frequencies for my eardrums to detect (in part because my eardrums are emitting noise at a those frequencies). Maybe the average wavelength of thermal noise is roughly the mean free path length of the air molecules, so the average frequency of the noise is roughly 5 GHz. But I just making stuff up and really don't know.
Well, I don't know. Some of the US is near Mexico, but most of it isn't. In Europe the farthest you can get from a border to foreign speaking country is perhaps southern Italy. The four US states which border Mexico are each bigger than Italy. Germany is a bigish country in Europe area-wise, but it's less than 3.7% the size of the US. The Mercator projection makes an optical illusion -- the US is huge.
Yeah, your explanation sounds absolutely correct. But before you think "silly monoglot Americans", remember that London is closer to Istanbul than New York is to Mexico. Countries where people don't mostly speak English are thousands of kilometers away from most Americans.
Asset allocation (what portion of your money is in stocks and bonds) is very important, depends on your age, and will get out of whack unless you rebalance. So use a Vanguard Target Retirement Date fund.
Although the Linux kernel and modern CPUs are piecewise-understandable, whereas neural networks are not.
Is it for reasons similar to the Strawman Chompsky view in this essay by Peter Norvig?
Here's how I read your question.
- Many machine learning techniques work, but in ways we don't really understand.
- If (1), I shouldn't study machine learning
I agree with (1). Could you explain (2)? Is it that you would want to use neural networks etc. to gain insight about other concrete problems, and question their usefulness as a tool in that regard? Is it that you would not like to use a magical back box as part of a production system?
EDIT I'm using "machine learning" here to mean the sort of fuzzy blackbox techniques that don't have easy interpretations, not techniques like logistic regression where it is clearer what they do
It sounds like you do not understand what your experiments are doing. That's pretty much why I'm not studying electromagnetism.
-- Letter from James Clerk Maxwell to Michael Faraday, in the setup of a Steam Punk universe I just now invented
The Snowden / Manning leaks (from what I've heard) suggest this issue is the third or forth priority of US intelligence organizations. One presumes that the US intelligence organizations do not consider it in their interests to advertise this fact.
I think the article was making a stronger statement: programming isn't the best way to learn programming (at least at first). Sounds bonkers to me, but I don't trust my pedagogic intuitions.
If you're at the state where the worst thing about a proof is that it relies on the axiom of choice, you're practically at the finish line (at least compared to here). Once proofs has been discovered, mathematicians have a pretty good track record of whittling them down to rest on fewer assumptions. From my (uninformed dilettante's) perspective, it's not worth limiting your toolset until you've found some solution to your problem. Any solution, even ones which rest on unproven conjectures, will teach you a lot.
I would think it faster to search for proofs of any kind, then simplify to an elementary/constructive/machine verifiable proof.
Do you know, offhand, if Baysian networks have been extended with complex numbers as probabilities, or (reaching here) if you can do belief propagation by passing around qubits instead of bits? I'm not sure what I mean by either of these thing but I'm throwing keywords out there to see if anything sticks.
I don't think the consensus of physicists is good enough for you to place that much faith in it. As I understand modern day cosmology, the consensus view holds that universe once grew by a factor of 10^78 for no reason. Would you pass up a 1 penny to $10,000,000,000 bet that cosmologists of the future will believe creating 10^100 happy humans is physically possible?
what criteria do you use to rule out solutions?
I don't know :-(. Certainly I like physics as a darn good heuristic, but I don't think I should reject bets with super-exponentially good odd based on my understanding of physics. A few bits of information from an expert would be enough to convince me that I'm wrong about physics, and I don't think I should reject a bet with a payout better than 1 / the odds I will see those bits.
Relativity seems totally, insanely physically impossible to me. That doesn't mean that taking a trillion to one bet on the Michelson Morley experiment wouldn't have been a good idea.
In the U.S., what happens to people who cannot survive without assistance (like people without the dexterity to feed themselves, or even people without the ability to keep a job or fill out paperwork) who do not have a family or the means to trade for help? What, physically, keeps them from starving to death?
Thank you for writing this post. I've had the same intuition as you about stopping rules for over a year now, but have never taken time to sit down and work it out. I look forward to working through the comments!
What's a network topology based sandboxing mechanism?
Question: I have a strong sense of a "dominant" direction (often, but not always, west). This direction is self-apparent in virtually every memory or mental visualization of any location I can think of. So, for example, captain's chair on the USS enterprise "obviously" faces "east", and the library on Myst island is obviously on the north side. I'm not going to forget which direction is down, and I'm not going to forget which direction is (usually) west.
Does anyone else here have oriented spacial memories?
My understanding is that a Snowden-leaked 2008 NSA internal catalog contains airgap-hopping exploits by the dozen, and that the existence of successful attacks on air gapped networks (like Stuxnet) are documented and not controversial.
This understanding comes in large measure from a casual reading of Bruce Schneier's blog. I am not an security expert and my "you don't understand what you're talking about" reflexes are firing.
But moving to areas where I know more, I think e.g. if I tried writing a program to take as input the sounds of someone typing and output the letters they typed, I'd have a decent chance of success.
Oops! I misremembered. So8res' second post was for that tournament, but his first was two weeks earlier. Shouldn't have put words in his mouth, sorry!
I am not close to an expert in security, but my reading of one is that yes, the NSA et. al. can get into any system they want to, even if it is air gapped.
Dilettanting:
- It is really really hard to produce code without bugs. (I don't know a good analogy for writing code without bugs -- writing laws without any loopholes, where all conceivable case law had to be thought of in advance?)
- The market doesn't support secure software. The expensive part isn't writing the software -- it's inspecting for defects meticulously until you become confident enough that defects which remain are sufficiently rare. If a firm were to go though the expense of producing highly secure software, how could they credibly demonstrate to customers the absence of bugs? It's a market for lemons.
- Computers systems comprise hundreds of software components and are only as secure as the weakest one. The marginal returns from securing any individual software component falls sharply -- there isn't much reason to make any component of the system too much more secure than the average component. The security of most consumer components is very weak. So unless there's an entire secret ecosystem of secured software out there, "secure" systems are using a stack with insecure, consumer, components.
- Security in the real world is helped enormously by the fact that criminals must move physically near their target with their unique human bodies. Criminals thus put themselves at great risk when committing crimes, both of leaking personally identifying information (their face, their fingerprints) and of being physically apprehended. On the internet, nobody knows you're a dog, and if your victim recognizes your thievery in progress, you just disconnect. It is thus easier for a hacker to make multiple incursion attempts and hone his craft.
- Edward Snowden was, like, just some guy. He wasn't trained by the KGB. He didn't have spying advisors to guide him. Yet he stole who-knows-how-many thousands of top-secret documents in what is claimed to be (but I doubt was) the biggest security breach in US history. But Snowden was trying to get it in the news. He stole thousands of secret document, and then yelled though a megaphone "hey everyone I just stole thousand of secret documents". Most thieves do not work that way.
- Intelligence organizations have budgets larger than, for example, the gross box office receipts of the entire movie industry. You can buy a lot for that kind of money.
In addition to current posters, these tournaments generate external interest. I, and more importantly So8res, signed up for an account at LessWrong for one of these contests.
In fairness to Hal Abelson, the pontification I remember isn't in the lecture in question, and my annoyance is more directed at pretentious classmates and some other things edit and aimed at marketing style, rather than substance.
If I were to attempt to summarize the lecture in question, it would be "The Greeks named Geometry after measuring the earth, but hundreds of years later think of them as wrestling with more fundamental ideas about space. Hundreds of years from now, people won't think of computer science as writing C programs for silicon, so much as wrestling with more fundamental ideas about ".
If you think that I'm missing the important crux, please let me know
It's a profoundish idea, and an interesting one to think about.
But (my point) I don't think it is an important misconception for the general public that computer science is about computers when it is in fact about . For 98% percent of humanity, and a good portion of computer scientists, Computer Science is a good name.
If DNA computers were big, or nanoparticle cellular automata building large structures were a thing, I would be more for separating out computers and to the general public. I hear the meme more in circumstances I interpret as trying to sounding counterintuitive and deep, which I think is the cause of my knee-jerk negative reaction.
EDIT The people I am quoting do not live in my world. They are at places like the Center for Bits and Atoms, where they really are studying the without the computers. But for the masses, today, in 2015, I do not think the distinction matters.
If you tell a pre-industrial farmer about machines, they are likely to form confused ideas. "You want theses "combine harvesters" to be born fully grown? That may sound like a time saver, but I assure you from years of experience that these combine harvesters will never be obedient unless you train them from infancy".
I think the misconceptions people have with AI stems from a lack of familiarity with any intelligent agent besides humans, not from bad terminology. You're going to have trouble talking to foragers about industrial equipment no matter what you call it.
SICP's whole "Computer science is neither about computers nor a science" pontification annoys the heck out of me, but arguments over definitions in general annoy the heck out of me. I mean, who the hell cares what CS is called? We don't title any Physics class Differencial Equations Which Exist in Their Own Platonic Sense And Only Incidentally Can Be Used To Model Our World, even though alien mathematicians whose world runs on cellular atomata might still study the heat equation. "Computer Science" tags a class as something you should study if you want to invent new things related to computers, and another name wouldn't do that much better.
Nice article. Minor note: the use of "launch a coup against the vice president" as the example of nice behavior added a surprising amount of cognitive burden for me. Stroop effect and all that.
With a 1,000 square kilometer industrial complex for the manufacture of slinkys and a million trained botanists.
So it has a current utility of (1-ε)10, and can increase this by reducing ε - hence by building even more paperclips.
I take ε to be the probability that something weird is happening like you're hallucinating your paperclips. Why would building more paperclips reduce ε? If you are dreaming, you're just making more dream paperclips.
I'm sure you'd spend your time with trying to find increasingly elaborate ways to probe for bugs in Descartes' demon's simulation. It is not clear to me why your increasingly paranoid bug probes would involve making paperclips.
For one, I cannot answer certain questions in the frame which my therapist imposes because I intellectually reject the assumptions that underlie them.
Examples? Just curious.
For another, I do not fully agree with the psychological establishment on what constitutes "healthy", adaptive, rational behaviour and would not like myself to adhere to even the closest variation on mental normality. There are areas of myself which I do not wish to display as "up for fixing", and do not allow interference other than my own in those areas. I consider myself rational enough to debug myself, if and only if I decide an intervention is warranted. Otherwise, I like and accept myself as I am, I consider most of my traits, broadly speaking, as part of my ideal self, and would like to preserve most aspects about myself.
Total armchair-psychologist kibitzing. This reads to me like someone who feels judged, or been repeatedly told that there's something wrong with them that they should fix, or for some other reason is in an emotionally defensive position. My guess is if you felt less judged or more respected, following the shrink's suggestions would feel more like giving some new habits a test drive than like ritually sacrificing parts of your identity.
If I can recover on my own and with the aid of antidepressants, without having to waste time on therapy for it, I'd gladly do it, but I have the lingering doubt that the shrink may be right after all and I might need to get my head checked.
I don't know enough about psychotherapy efficacy statistics to say, but heuristically I tend to assume that experts are better at judging these things than non-experts.
A possibility to consider: there may be behavior changes that would have a highly positive impact to your future life (avoiding arguing habits which exacerbate relational strife, to make up an example), but that aren't terribly relevant to getting out of this depressive slump. Have you tried the usual anti-depressive suspects (exercise and socializing)? When you feel better / at the beginning your next relationship, it might be worth revisiting some of the things which went wrong in the past to try to avoid them.
This comment should be read as informal musings for the purpose of collecting outside views.