Open thread, Jul. 11 - Jul. 17, 2016

Summary after your explanations: Yes, I'm really interested in an open-source packaging of your library. Preferably on github.

I read a bit of what you previously wrote about your approach but I didn't read your full book.

I think a bunch of Quantified Self applications would profit from good compression. It's for example relatively interesting to sample galvanic skin response in very short time intervals of 5ms. Similar things go for accelerometer data. It would be interesting what kind of data you can draw from the noisy heart rate data on smartwatches with shorter time intervals.

Smart watches could easily gather that data with shorter time accuracy than they currently do but they have relatively limited space.

In practice I think it will depend a lot on how easy it is to use your software.

Maybe you could also have a gamified version. You have a website and every week there's a dataset that get's published. Only have of the data is released. Every participant can enter their own model via a website and the person who's model compresses the unreleased part of the data the best wins.

Is it something like this? http://www.cipr.rpi.edu/research/SPIHT/EW_Code/FastAC_Readme.pdf

Which language?

Just post it on github with no effort. If you start getting pull requests or issues logged, you'll have your answer.

Two basic questions:

(1) What are the immediate practical applications?

(2) How qualified must the user be? (The "all you have to do is supply a probability model" part is worrying :-/)

Adrenaline can improve function in a number of domains, so it might be that anyone who has test anxiety or some other performance anxiety could in certain situations that they perceive as threatening have some sweet concentration of adrenaline in their bodies such that they have improved performance rather than impaired performance, and this is never recognized as occurring by the same mechanism as test or performance anxiety unspecified because the amount of adrenaline causes symptoms that aren't detrimental. Conceivably the amount of adrenaline and the effects thereof could be different across different domains.

Maybe that happens to you. What do you think?

EDIT: This might lead to empirical evidence. Anxiety may decrease performance when attention has to be switched between tasks, but may improve performance when the task is difficult and singular. Think social situations vs. exams.

See: Yerkes-Dodson law and research on "optimal level of arousal".

I know regard reading a book a not so trivial investment of time and energy, given the huge quantity of possible books I could be reading right now.
Is there any particular reason to believe Hanson's beliefs? So that it makes sense to anticipate the future the way he does?

The reason parties are oversupplied with food is because the incentives are asymmetrical. Specifically, the loss from having too much food is considerably smaller than the loss from having too little food.

Having insufficient food is a significant loss of status since you failed as a host to provide proper hospitality. There are a bunch of obvious historical and cultural reasons why not being able to feed your guests is a bad thing, status-wise.

Having too much food is just a matter of some wasted money and/or having to eat leftovers for few days. Not a big deal at all nowadays.

For most problems like this, it's worth solving once or twice at small scale before you look for general solutions. How many parties have you thrown (or guided the food procurement for), and what have you found that makes for better estimation of needs?

Have you talked with caterers or other experts in such estimation? It would be interesting to learn how they decide when to risk too little vs too much, and the clever tricks they have to control consumption (which will make the estimates more accurate). For instance, having lots of cheap starches and limited meat, along with explicit or subtle rationing, can lead to high waste measured by weight or calories, but fairly low waste measured by cost.

Just to provide a data point: I'm hosting get-togethers of friends that you might call parties regularly and usually nothing or a very small amount is thrown away.

I'm wondering whether this might be specific to Germany. Here there is some social pressure to avoid wasting stuff (together with a strong trend for sustainability).

"why not keep track of how much food my guests actually ate, and try adjusting the amount of food at my next party to match?"

Because the amount of food that people eat is not an absolute value, but a function of how much is there. If you do that adjustment, and then continue to do that adjustment, you will end with a situation without any food. That is true both at parties and in any other situation, like meals served to people who otherwise will have nothing to eat, at least to a first approximation -- if the last situation is absolute, you will get people eating some food, but it will not be enough to live on.

why not keep track of how much food my guests actually ate, and try adjusting the amount of food at my next party to match?

I guess there is not a fixed amount of food brought per guest, but rather a random distribution. The host's goal is not to make sure that the average "food brought" equals the average "food desired", but rather that with, say, 95% probability the current "food brought" is at least 90% of "food desired" (feel free to change the numbers to fit your experience). Also, the host is hedging against the possibility that the few guests who usually come with hands full of food, suddenly can't come or for some random reason come empty-handed.

There are charities that focus on "recycling" food waste

I guess the best way to improve the world is to have a list of such charities in your neighborhood ready in a printed form, and give it to the host if they are interested.

This will not have any practical consequences whatsoever, even in the long term. It is already possible to perform reversible computation (Paper by Bennett linked in the article) for which such lower bounds don't apply. The idea is very simple: just make sure that your individual logic gates are reversible, so you can uncompute everything after reading out the results. This is most easily achieved by writing the gate's output to a separate wire. For example an OR gate, instead of mapping 2 inputs to 1 output like

(x,y) --> (x OR y),

it would map 3 inputs to 3 outputs like

(x, y, z) --> (x,y, z XOR (x OR y)),

causing the gate to be its own inverse.

Secondly, I understand that the Landauer bound is so extremely small that worrying about it in practice is like worrying about the speed of light while designing an airplane.

Finally, I don't know how controversial the Landauer bound is among physicists, but I'm skeptical in general of any experimental result that violates established theory. Recall that just a while ago there were some experiments that appeared to show FTL communication, but were ultimately a sensor/timing problem. I can imagine many ways in which measurement errors sneak their way in, given the very small amount of energy being measured here.

As far as I can see, the experiment has shown that what was considered to be the lower bound is actually not.

However I don't understand how the claim of "no lower bound at all" necessarily follows. For all we know there is just a different, lower (lower bound).

I've read (mostly things by Ron Maimon) that marijuana* can actually impair your ability to do calculations (and in extent, I'd also assume your ability to make decisions) and I'm curious if there's any truth to that.

• Is there a difference between marijuana, medical marijuana, weed, instert_name_here? They seem to be used interchangeably. At least they seem to cause a similar if not the exact same effect.

past = outside view

For example, if in the past people have repeatedly suggested a plan to create a paradise on Earth, and the plan, when realized, repeatedly ended with bloodshed and poverty, and now someone suggests the same plan again... I guess that's a reason to suspect it probably wouldn't end well. At the very least, the proponent should explain why exactly the previous instances have failed and what exactly they are planning to do differently today to avoid that specific failure.

But there is a difference between using the past as an outside view, i.e. conservatism; and worshipping the "past as my modern mind imagines it", i.e. neoconservatism / neoreaction. The latter is, ironically, in some aspects similar to the progressives who are worshipping the fictional future -- similar approach to modelling society, different aesthetics (or as you called it "positive / negative affiliation with the passage of time").

I think a lot of political questions hinge on what's possible, and also what the consequences of policies are. If someone says "I think we should arrange marriages instead of letting individuals pick," then the immediate questions to settle are 1) will people allow such a policy to be put in place / comply with it, and 2) what will the consequences be?

(There's also the "does this align with principles" deontological question, but this is relatively easy to answer without looking at the past or present so I'll ignore it.)

And the past provides our primary data source to answer those sorts of questions. Yes, we can imagine multiple different causal effects of attempting to arrange marriages, but how those interplay with each other and shake out is hard to know. But other people tried that for us, and so we can investigate their experiments and come to a judgment.

It doesn't matter if it comes from the past, the Bible, Lord of the Rings or utopian literature. Just discuss the model! It's mostly fiction anyway.

I very much prefer people who base their political beliefs based on empirics about the real world compared to people who just base their political beliefs on made-up fantasy. I don't think there a good reason to treat both the same.

Try us. Are you selling to us? If yes then maybe not so great to do (however Squirrelinhell released hasteworm just recently and no one complained). If no, then idea sharing is good.

This is further complicated by the fact that even the anticipation-beliefs are probabilistic. So you can have a "belief" that says "I love my sports team", and a belief that says "I expect my team to win (probability 80%)".

So in both cases it is possible for the team to lose and the person to keep their belief.

I don't think the test does what you propose. I can both strongly identify with a belief and at the same time make anticipations based on the belief.

This paper frames the problem as "look at a program and figure out whether it will be harmful" and correctly observes that there is no way to solve that problem with perfect accuracy if the programs being analysed are arbitrary. But its arguments have nothing to say about, e.g., whether there's some way of preventing harm as it's about to happen; nor about whether it is possible to construct a program that provably does something useful without harming humans.

E.g., imagine a world where it is known that the only way to harm humans is to press a certain big red button labelled "Harm the Humans". The arguments in this paper show that there is no general procedure for deciding whether a computer with the ability to press this button will do so. But they don't rule out the possibility that you can make a useful machine with no access to the button, or a useful machine with a little bit of hardware in it that blows it up if it gets too close to the button.

(There are reasons to be concerned about such machines because in practice you probably can't causally isolate them from the button in the way required. The paper's introductory material discusses some such reasons. But they play no role in the technical argument of the paper, at least on the cursory reading I've given it.)

Assuming that a superintelligence will contain a program that includes all the programs that can be executed by a universal Turing machine on input potentially as complex as the state of the world

What is the notion of "includes" here? Edit: from pp 4-5:

This means that a superintelligent machine could simulate the behavior of an arbitrary Turing machine on arbitrary input, and hence for our purpose the superintelligent machine is a (possibly identical) super-set of the Turing machines. Indeed, quoting Turing, “a man provided with paper, pencil, and rubber, and subject to strict discipline, is in effect a universal machine”