Posts
Comments
Would you classify MC-AIXI as a General AI?
I think a better example than frictionless surfaces and no air resistance would be idealized symmetries. Once something like Coulomb's Law was postulated physicists would imagine the implications of charges on infinite wires and planes to make interesting predictions.
We use the trolley problem and its variations as thought experiments in order to make predictions we can test further with MRIs and the like.
So a publication on interesting trolley problem results would be like theoretical physics paper showing relativity predicts some property of black holes.
I would compare the trolley problem to a hypothetical physics problem. Just like a physicist will assume a frictionless surface and no air resistance, the trolley problem is important because it discards everything else. It is a reductionist attempt at exploring moral thought.
I think you are looking at the Trolley Problem out of context.
The Trolley Problem isn't suppose to represent a real-world situation. Its a simplified thought experiment designed to illustrate the variability of morality in slightly differing scenarios. They don't offer solutions to moral questions, they highlight the problems.
Didn't Harry also swear to keep what he and Draco experiment with secret? This is why he never told her about the magic gene either, unless I am misremembering things.
From Ch. 23
There's something called the Interdict of Merlin which stops anyone from getting knowledge of powerful spells out of books, even if you find and read a powerful wizard's notes they won't make sense to you, it has to go from one living mind to another
The Lazy Student, The Grieving Student, The Sports Fan: make the deadline for reports a curve instead of a cliff. Each day of delay costs some percentage of the grade.
I've always liked the "drop the n lowest scores" strategy. For example, 10 assignments given with the lowest 2 scores ignored.
You are pre-committing to a set of rules, where any excuse would have a much lower probability of being true. Any excuse would need to include 3 excuses. Combining the probabilities of each of the excuses will likely bring the total under your acceptable threshold. Basically, it's lowering the likelihood that you will want to violate the rules.
You can also look at like this. Your model of people predicts that they are scoundrels, and will try to violate the rules, maximizing their utility at your expense. So build a system where procrastinators can maximize their utility at no expense to you.
Hi, I'm Matt Stevenson. 24 yr old computer scientist. I work on AI, machine learning, and motor control at a small robotics company.
I was hooked when I read Eliezer on OvercomingBias posting about AGI/Friendly AI/Singularity/etc...
I'd like to comment (or post) more, but I would need to revisit a few of the older posts on decision theory to feel like I'm making an actual contribution (as opposed to guessing the karma password). A few more hours in the day would be helpful.
Even if it is a gut feeling and not an explicit lie, he is still showing that his facts are weak since he's resorting to emotions.
I think this is a problem that applies to a lot of people who are socially dysfunctional, not just those who are high intelligence. Generalizing from one example?
It seems that if there were karma transfers in place, betting against a funny picture post would be an almost guaranteed loss.
We could set up some ground rules that would exclude you specifically from starting the post, but I don't see how any rules set up in advance would prevent collusion to create the thread. Also, any karma lost for making an explicitly bad thread would be more than made up for with the 500 karma win.
I'm not sure what you are trying to argue here? I am saying that trying to use a reference class prediction in a situation where you don't have many examples of what you are referencing is a bad idea and will likely result in a flawed prediction.
You should only try and use the Outside View if you are in a situation that you have been in over and over and over again, with the same concrete results.
... then the data is most likely insufficient for reasoning in any other way If you are using an Outside View to do reasoning and inference than I don't know what to say other than, you're doing it wrong.
If you are presented with a question about a post-singularity world, and the only admissible evidence (reference class) is
the class of instances of the human mind attempting to think and act outside of its epistemologically nurturing environment of clear feedback from everyday activities.
I'm sorry, but I am not going to trust any conclusion you draw. That is a really small class to draw from, small enough that we could probably name each instance individually. I don't care how smart the person is. If they are assigning probabilities from sparse data, it is just guessing. And if they are smart, they should know better than to call it anything else.
There have been no repeated trials of singularities with consistent unquestionable results. This is not like procrastinating students and shoppers, or estimations in software. Without enough data, you are more likely to invent a reference class than anything else.
I think the Outside View is only useful when your predictions for a specific event have been repeatedly wrong, and the the actual outcome is consistent. The point of the technique is to correct for a bias. I would like to know that I actually have a bias before correcting it. And, I'd like to know which way to correct.
Edit: formatting
It seems like the Outside View should only be considered in situations which have repeatably provided consistent results. This is an example of the procrastinating student. The event has been repeated numerous time with closely similar outcomes.
If the data is so insufficient that you have a hard time casting it to a reference class, that would imply that you don't have enough examples to make a reference and that you should find some other line of argument.
This whole idea of outside view is analogous to instance based learning or case based reasoning. You are not trying to infer some underlying casual structure to give you insight in estimating. You are using an unknown distance and clustering heuristic to do a quick comparison. Just like in machine learning it will be fast, but it is only as accurate as your examples.
If you're using something like Eigenfaces for face recognition, and you get a new face in, if it falls right in the middle of a large cluster of Jack's faces, you can safely assume you are looking at Jack. If you get a face that is equally close to Susan, Laura, and Katherine, you wouldn't want to just roll the dice with that guess. The best thing to do would be to recognize that you need to fill in this area of this map a little more if you want to use it here. Otherwise switch to a better map.
Edit: spelling
Being Watched +4-7 - This can depend on who the other person is and the situation. I don't like paired programming since I'm an introverted thinker, and I find it really distracting. When there is someone else in the room doing work, it motivates me to do more work. I find the reverse can be true as well. If I'm around a bunch of people who are slacking off, I become less motivated.
Cripple your Internet +5 - This is a pretty effective technique, but I have a hard time being consistent with this at all.
One thing I've noticed is that my akrasia, as well as productivity, seem to have an inertia. Its not necessarily that I always have an urge to procrastinate, but if I am not being productive, it is always easier to open a browser and start procrastinating than to start working.
I try to arrange my tasks in a way that I can keep a productive inertia. Make sure all tasks are well defined with clear starting points and goals. Try to break a task up into smaller tasks that can be completed in less than a day. Arrange tasks so they build off of the work I just completed, I try not to switch between unrelated work more than a few times in a single day. When starting a project, I start with a few easy tasks that I know can be completed quickly. This helps frame my mind for work.
Similarly, before I stop working for the night, I try to set leave myself an easy problem to solve than would lead into something larger. When I know exactly how to accomplish something, there is less resistance to starting the task. A lot of times it is easier to continue working than to start working.
By far, what I've found to be most motivating is to be working with people who themselves are very motivated about the work. I am a very competitive person. This can be a good thing with the right group.
Here you are relying on omega using two ordering systems that we already find highly correlated.
What if Omega asked you to choose between a blegg and a rube instead of A and B. Along with that, Omega tells you that it did not necessarily use the same ordering of blegg and rube when posing the question to the copy.
EDIT: More thoughts: If you can't rely on an obvious correlation between the player labels and choices, why not have a strategy to make a consistent mapping from the player labels to the choices.
The key to winning this game is having both parties disagree. If both parties know the goal and have a consistent mapping process, it would be trivial for them to arrive at different choices.
A simple mapping would be alphabetize the player labels and the choice labels. Player(1) => choice(1), Player(2) => choice(2), Player(n) => choice(n).
Now, to return my procastinating energies to hacking out an improved ruby based bash replacement. rush just doesn't have the tab-completion I need to give myself the illusion of smooth productivity.
If you were to include a history of your commands across sessions, and maybe an option to dump all the commands from the current session into a .rb file, I would love you forever.
solve?
This is wonderful. I'm rather new to LW/OB and I've been reading through chains of posts.
I was about to start working on something just like this to help myself and other new readers.
Thank you.