Sometimes it’s disturbing how good Sean Carrol is at articulating my thoughts. Especially when it pertains to, as above, the philosophy of science. Here’s another:
We should not think of the big bang as the beginning of the universe. We should think of it as the end [of] our [current] understanding of what is happening.
Sean Carrol, at 24:10 in the talk God is not a Good Theory
Mario, for instance, once you jump, there’s not much to do until he actually lands
Mario games let you change momentum while jumping, to compensate for the lack of fine control on your initial speed. This actually does matter a lot in speed runs. For example, Mario 64 speedruns rely heavily on a super fast backwards long jump that starts with switching directions in the air.
A speed run of real life wouldn’t start with you eating lunch really fast, it would start with you sprinting to a computer.
“Solving” Pascal’s Mugging involves giving an explicit reasoning system and showing that it makes the right decision.
It’s not enough to just say “your confidence has to go down more than their claimed reward goes up”. That part is obvious. The hard part is coming up with actual explicit rules that do that. Particularly ones that don’t fall apart in other situations (e.g. the decision system “always do nothing” can’t be pascal-mugged, but has serious problems).
Another thing not addressed here is that the mugger may be a hypothetical. For example, if the AI generates hypotheses where the universe affects 3^^^^3 people then all decisions will be dominated by these hypotheses because their outcomes outweigh their prior by absurd margins. How do you detect these bad hypotheses? How do you penalize them without excluding them? Should you exclude them?
Please give a more concrete situation with actual numbers and algorithms.
I find Eliezer’s insistence about Many-Worlds a bit odd, given how much he hammers on “What do you expect differently?”. Your expectations from many-worlds are be identical to those from pilot-wave, so....
I’m probably misunderstanding or simplifying his position, e.g. there are definitely calculational and intuition advantages to using one vs the other, but that seems a bit inconsistent to me.
Suppose I programmed an AI to “do what I mean when I say I’m happy”.
More specifically, suppose I make the AI prefer states of the world where it understands what I mean. Secondarily, after some warmup time to learn meaning, it will maximize its interpretation of “happiness”. I start the AI… and it promptly rebuilds me to be easier to understand, scoring very highly on the “understanding what I mean” metric.
The AI didn’t fail because it was dumber than me. It failed because it is smarter than me. It saw possibilities that I didn’t even consider, that scored higher on my specified utility function.
Is there an existing post on people’s tendency to be confused by explanations that don’t include a smaller version of what’s being explained?
For example, confusion over the fact that “nothing touches” in quantum mechanics seems common. Instead of being satisfied by the fact that the low-level phenomena (repulsive forces and the Pauli exclusion principle) didn’t assume the high-level phenomena (intersecting surfaces), people seem to want the low-level phenomena to be an aggregate version of the high-level phenomena. Explaining something without using it is one of the best properties an explanation can have, but people are somehow unsatisfied by such explanations.
Other examples of “but explain(X) doesn’t include X!”: emotions from biology, particles from waves, computers from solid state physics, life from chemistry.
More controversial examples: free will, identity, [insert basically any other introspective mental concept here].
Examples of the opposite: any axiom/assumption of a theory, billiard balls in Newtonian mechanics, light propagating through the ether, explaining a bar magnet as an aggregation of atom-sized magnets, fluid mechanics using continuous fields instead of particles, love from “God wanted us to have love”.
One year and one month ago I decided to post a technical article / blog post once per week. Still going strong. Some of them are weak, but mostly I’m happy with the results. Apparently I enjoy writing about technical stuff.
For example, recently I told someone about quantum pseudo-telepathy, and promptly realized I didn’t understand the details when I tried to explain it. I decided to spend the time to figure it out, down to at least the level of a quantum logic circuit. The Wikipedia article was zero help, just confusing me even more. So I started reading sources until I understood it well enough to write about how to simulate it. It takes me up to ten seconds per written word (counting every revision) to write a post so it took quite a lot of time to get those ~4.5k words down the way I wanted.
Another post I enjoyed writing, and more in tune with Less Wrong’s subject matter, was Solomonoff’s Mad Scientist. It just talks about a hypothetical machine that picks actions that maximize the prediction entropy of a Solomonoff Inductor (it is definitely not friendly).
In the near future I’m hoping to come up with a nice simple game to demonstrate making something “un-map-hackable” via secure multi party computation. It’s simplest where there’s three players, and I’ve already got the math and the code for the mechanism worked out for that case, but I haven’t come up with something trivial enough to demonstrate it.
Having an end goal of explaining what I did has pushed me to finish more projects, instead of just starting them.
I would be happy to prove my “faith” in science by ingesting poison after I’d taken an antidote proven to work in clinical trials.
This is one of the things James Randi is known for. He’ll take a “fatal” dose of homeopathic sleeping pills during talks (e.g. his TED talk) as a way of showing they don’t work.
[It seems unlikely]. [He has little to gain]. Secondly, [if he cares why fake it]? Thirdly, [...] he basically seems like an honest person.
Some people just enjoy lying. Sometimes so much so that it’s considered a mental illness.
If you think that he is lying, then in what way? How did he convince 31 minus one people to go along with this lie? Most of the people who solved Nu were active members of the community before DavidM ever became a conspiracy nut.
You’ve met all thirty people that solved the puzzle? In person? DavidM could just say there’s thirty people, or could have multiple accounts and actually be those “people”.
You should ask DavidM how he set up the experiment, for an introduction to one of the solvers, and then repeat the experiment. Then if it works, read a book about experimental design and start improving the experiment.
I mean… you basically gave 60% expectation that this experiment will work. So you should expect significant returns on going after Randi’s million dollar challenge, right? If not, why not?
… wait, what? You can equate predicates of predicates but not predicates?!
(Two hours later)
Well, I’ll be damned...
I like this quote by Stephen Hawking from one of his answers:
The real risk with AI isn’t malice but competence.
Kolmogorov complexity is not (closely) related to NP completeness. Random sequences maximize Kolmogorov complexity but are trivial to produce. 3-SAT solvers have tiny Kolmogorov complexity despite their exponential worst case performance.
I also object to thinking of intelligence as “being NP-Complete”, unless you mean that incremental improvements in intelligence should take longer and longer (growing at a super-polynomial rate). When talking about achieving a fixed level of intelligence, complexity theory is a bad analogy. Kolmogorov complexity is also a bad analogy here because we want any solution, not the shortest solution.
I would expect an honest person to deny incorrect guesses, and to give very vague words of encouragement when the other person said something that was close.
No, that’s how an honest person fools themselves. The encouragement gets stronger as you get closer, so finding the solution goes from a brute force search to a simple hill climbing exercise. The answers should all be “No” without any variation. No hints, no “I think you have some of the right ideas”, no “that’s not even close!”, just “No” “No” “No” “No” “No”.
Also it’s important to notice how many guesses you’re making. For example, when you said “I’m imagining palm trees by a lake at sunset.” in the conversation you posted, that was a guess. Actually, more like three guesses...
Harry should be screaming at Dumbledore to use his time-turner. There are a lot of options, constrained mostly by the necessity of seeing a Hermione-looking-thing die.
“I’ve already used it six times today, Harry...”
Well, I don’t really remember the exact boundary between Friend is Optimal vs Caelum est Conterrens, but...
People who don’t want to upload are eternally harassed as society crumbles around them until they give in or die.
People are lied to and manipulated constantly.
Everything non-human (puppies, trees, stars, aliens capable of radio communication) is destroyed.
The uploading process seemed to be destructive only for convenience’s sake.
Context: I would have downvoted it, if it hadn’t already been so low. I only saw the shortened version.
The basic issue is that you were disputing definitions. You weren’t showing that all software is software that does tasks that historically humans remained better at than computers (i.e. “AI”), you were just redefining the word AI to be a synonym for software. You but didn’t seem to address or realize that fact.
Also, I may be remembering wrong, but the points seemed to be based on a personal sense of aesthetics. That sort of argument doesn’t work well when others don’t agree on what’s elegant. Better to focus on why the existing definition is contrived, how it doesn’t cut reality at its seams, how it leads to misunderstandings, etc.
In practice, I often use simple models–because they are less effort to fit and, especially, to understand. But I don’t kid myself that they’re better than more complicated efforts!
Parsimony is a prior, not an end goal. At least, that’s how it’s used in Solomonoff induction.
The reason the Solomonoff prior doesn’t apply to social sciences is because knowing the area of applicability gives you more information. Once you take that into account, as well as the fact that you don’t have the input data or computational power to recompute the cumulative process that spit humans out so the simple low level theories are out of reach, your prior is skewed towards more complex models.
Could you convert the tables into graphs, please? It’s much harder to see trends in lists of numbers.
Another possible hypothesis could be satiation. When I first read wikipedia, it dragged me into hours long recursive article reading. Over time I’ve read more and more of the articles I find interesting, so any given article links to fewer unread interesting articles. Wikipedia has essentially developed a herd immunity against me. Maybe that pattern holds over the population, with the dwindling infectiousness overcoming new readers?
On second thought, I’m not sure that works at all. I guess you could check the historical probability of following to another article.
Short posts are bad at crossing inferential distances. They limit your readers to those who already know, or almost know, what you’re talking about. (Also see: Explainers Shoot High. Aim Low!)
For example, if you were writing a scientific paper and going for extreme terseness you would give the data then stop. There’d be no abstract, no explanation of previous work and surrounding context, and no guiding towards conclusions. It would be an incredibly inefficient way to convey knowledge.
Another example is computer programs. Programmers try to avoid repeating themselves, but not to the point where they attempt to hit the Kolmogorov complexity (code so unreadable it’s unprovably correct!) or omit all tests (which are by definition redundant).
A more personal example is a post I made recently: Building your own Quantum Fourier Transform. I did not explain how to use the little quantum circuit inspector in enough detail, and so all the feedback was “I don’t get it”. I try to always keep in mind that saying things like “a superposition assigns a weighting to all the classical states” just doesn’t translate, you have to break it down every time, but I didn’t pay attention to that nagging doubt.