Yudkowsky's friendly AI, and signalling.

 Here's Eliezer Yudkowsky, a somewhat well known blogger, outlining his beliefs about the artificial intelligence.

The author's unawareness of just how much he falls into the categories that he himself rejects is rather amusing. For instance, at one point he says that good guys would not implement communism into an AI.

Yet, what's about him? He grew up in a democracy, and with parents that thought they knew better than him what he'd want when he grows up (I think most of us can recall parents say that you'll appreciate something they force you to do, when you grow up). And his FAI concept follows mixture of those principles, with addition of very simplistic form of utilitarianism. Politically, it is of course about as neutral as hypothetical Chinese perfect socialism AI. Most people at that point would notice that they are implementing a political system, and try to argue why their political AI is better than hypothetical socialist AI; not Yudkowsky. This guy appears blissfully unaware that his political views are at all political; he won't cooperate with your argument by defending his political views prematurely.

More of that here , where he disses with great contempt other AI researchers of comparable competence to what would naturally be assumed of him.

This appears to be a kind of signalling. Dissing a category that you naturally belong to is a very easy and dirty way to convince some of the listeners and perhaps yourself that you do not belong to this category. Think how much hard work someone would have to do to positively convince you that he is above typical AI crackpot, let alone a visionary with a valid plan for a safer AI! He'd have to invent something like an algorithm for a self driving car, a computer vision system, or something else that crackpots can't do. He'd have to have a track record as a visionary. Something that actually works, or several such things. That level of ability is quite rare, and that's why it is hard to demonstrate. But one could just bypass this and diss typical AI researchers, and a few followers will assume exceptional competence in the field of AI.

Very simple way to control robot with smartphone

A very simple robot controlled from on-board Samsung Galaxy S2 (using my girlfriend's Galaxy Note as remote control. Todo: write some autonomous software using the phone's camera, sensors, etc)


The robot is controlled by drawing white rectangles on the smartphone screen, which activate photodiodes (I used red LEDs as photodiodes).

The robot is built entirely out of various trash and spare components I had:

Close-up on the wheel:

The blue thing on the motor shaft is insulation from some wire, used to increase grip. The wheel is two bottle caps and insulating tape, spinning freely on a thin wire used as a shaft. The weight of the robot presses the wheel and motor shaft together.

Circuit board:

Circuit diagram (I built 2 of this):

Circuit notes: I used red LED as a photodiode, BC337 for t1 and t2, and IRFZ44N for the mosfet, because that's what I had laying around. I've built 2 controllers for both wheels. If you are buying components for that project, I heavily recommend making some different circuit and using a motor controller IC that would allow you to reverse the motor. You can use the D1, T1, T2, R1 combination with pretty much anything. Also, you may want to connect the lower leg of the photodiode D1 to +9 rather than the emitter of T1 (reverse bias the diode) . IRFZ44N is an enormous overkill for these puny motors - it could switch 50 amps of current, at 55 volts.

Cineplex "Escape from this world"

Full stereoscopic 3D, at 4K resolution, playing in Cineplex theaters in Canada. See a video here.
Clouds were rendered using customized version of Volumetrics .

A brief note on different types of probability

(Typing it up to reference in discussions; it is by no means a detailed overview)

It is unanimously agreed that statistics depends somehow on probability. But, as to what probability is and how it is connected with statistics, there has seldom been such complete disagreement and breakdown of communication since the Tower of Babel. Doubtless, much of the disagreement is merely terminological and would disappear under sufficiently sharp analysis.

Leonard Savage, The foundations of statistics, 1954.

The origin of the classical probability
When you throw a symmetrical 6-sided die, after sufficiently many bounces, the probability of getting one specific face is 1/6 .

We arrive at this number by considering the symmetry of the die and the physical process of the bouncing of the die. We can not predict which side the die will land at - due to the extreme sensitivity to initial conditions - but the symmetry permits us to conclude something about the way the die will land on average if we are to perform many trials. It is not a philosophical stance that this probability represents the frequency of occurrence in an infinite number of trials - it is just really what it is - in our partial model of the die physics.



The probability as observer's belief
The observer's probability that a die on the table has rolled some number may have to be further adjusted based on extra knowledge. For instance you can look at the die, and see that it rolled 5; now your probability that the die has rolled 5 is nearly 100%.

For example of a more involved event, you can make a robot that throws a die and tosses a coin, and if the coin has landed heads, it tells you the number that the die rolled, otherwise it tells you an uniform random number between 1 and 7 inclusive (which it obtains, say, by spinning a small roulette). Exercise for the reader: the robot gave you 6; what is the probability that the die rolled 6, and what is the probability that robot was answering using the roulette? 

Bayes rule
The Bayes theorem is the rule by which probabilities affect other probabilities in the examples such as this robot problem (which I recommend you to solve on your own).

Uses of probability theory in computer graphics (and other applied physics)

Probability theory is widely used in computer graphics; for instance to calculate illumination values by averaging the number of photons that hit specific area. Pseudo-random number generators are employed in place of the die toss; a pseudo-random number generator is actually rather similar, in essence, to the bouncing of the die. Literal computation of average number of photons is often prohibitively slow for high quality imagery as the error decreases proportionally to inverse square root of number of simulated photons; a wide variety of more advanced methods, for example Metropolis light transport, are used to improve the asymptotic convergence. Sometimes, the convergence can be improved by forcing the photons into a regular, rather than random pattern and re-regularization of the photon field. The Bayes rule also pops in once in a while.



Probability theory as logic of uncertainties
Classical logic processes certain propositions and their relations, to obtain conclusions about the world. The probability theory gives same results as classical logic in the limit of certainty, and can be used to process uncertain propositions.



Hypotheses and probability
Suppose you have a hypothesis that a coin is biased and it always lands heads. How do you test this hypothesis? You can adopt a strategy that you toss the coin 20 times and believe coin to be biased if it always landed heads. Then an unbiased coin will trick you approximately 1 time out of million. Your degree of confidence in the coin being biased is thus described as this: I assume it is biased on basis of success of an experiment which had one in a million chance of non biased coin tricking me. (You can choose the adequate number of tosses in the experiment on basis of the cost of mistake and cost of the toss). This is one of the basic concepts of the scientific method.

But wait, you say. The coin might be biased, or it might not be biased, it is uncertain if it is or isn't! How can I assume it is biased? Wouldn't it be useful to find probability that the coin is biased? If you apriori knew the probability that the coin is biased, you could calculate the probability that coin is biased after performing series of experiments, using the above-mentioned likelihood of being tricked, in combination with Bayes theorem.

So there is a strong desire to assign some more or less arbitrary prior probability for the coin being biased, and then update it using Bayes theorem. While after a multitude of updates you become more correct, it still has all the obvious disadvantages of introducing a made up, arbitrary number into your calculations.

The former representation of the partial resolution of uncertainty by experiment is often called "frequentist" , while the latter is called "Bayesian".

Length-dependent prior probabilities for hypotheses


One can assign lower prior probabilities to more complicated hypotheses. Formally, one can represent a hypothesis with a Turing machine input tape of length l, and assign it probability proportional to 2^-l . This is called 'Solomonoff prior'. You can strike out the hypotheses that do not match the observations, this is called Solomonoff Induction.

Note that it is mathematically equivalent to a belief, as a dogmatic certainty, that the ultimate physics of the world we live in is a prefix Turing machine that is fed random bits on the input tape (probability of a specific start sequence is then  2^-l) .

Got to love Torvalds

https://lkml.org/lkml/2012/3/8/495

Because that shows that they don't understand what the whole *point*
of the kernel was after all. We're not masturbating around with some
research project.  We never were. Even when Linux was young, the whole
and only point was to make a *usable* system. It's why it's not some
crazy drug-induced microkernel or other random crazy thing.

 Damn straight. I think OSS projects need a bit more of that, with all the rewriting and such going on and everything breaking all the time.

Highly exceptional cognitive test scores versus exceptional performance

I was pondering the other day why the best performers such as Nobel prize winning physicists, best mathematicians, and so on, do have somewhat high, but not truly exceptionally high performance on the IQ test and similar tests, while some of the ultra high IQ individuals, or ultra exceptional childhood SAT solvers seem to be smart, but for the lack of other word, not really be that smart or wise or capable as the scores would suggest.

The on-going Olympics provides a clear analogy. The best sprinters will not be the best marathon runners; former have large fast versus slow fibres ratio and metabolism geared more towards anaerobic performance, whereas the latter have primarily slower fibres and metabolism geared towards aerobic performance. Past certain level of exceptional performance, there could be very little overlap between the exceptional performers in those two related, but different sports. Note that the best sprinters will still run marathons a fair bit better than average man, and best marathon runners will still sprint better than average man.

The brain is a fair bit more mysterious organ than the muscle, and we understand it very poorly, but it is the case that there are several well known variables that would represent a trade-off between different types of performance. For instance, glia to neuron ratio. Glia are the support cells that provide nutrition to neurons as well as remove the metabolism by-products of the neurons; furthermore glia have been recently found to be implicated in the memory.

Glia to neuron ratio should influence the cognitive performance, and it appears highly unlikely that the optimal ratio for the tests would precisely coincide with the optimal ratio for the insight making or the real world performance. There are many other such trade-offs within the brain. Hormonal levels, thickness of myelin shealths, short range vs long range connectivity, gray matter vs white matter... you can continue this list for pages.

The short term tests consisting of large number of disjointed questions which do not involve a significant body of learned knowledge (outside verbal, which represents a highly specialized brain region and very special type of learned knowledge) seem even more distant from insight making or long term work than sprint is different from a marathon race; I would expect even less of an overlap between exceptional performers on the criteria there. The relation may be more similar to that between grip strength and biathlon.

With regards to the childhood testing, it seems clear that exceptional childhood athletes at age 10 would have variations (Various hormonal dis-balances?) that are detrimental to the adult performance.

Note that none of this argument contradicts existence of correlations. Bulk of measured correlation comes from the values close to the mean, and it is the case that all the exceptional performers at age 20 were very good at age 10. Just not as good as the phenotypes which make use of the trade-offs.

edit: See also: Spearman's law of diminishing returns ; among the high IQ range, the correlation between different traits decreases. It is thus not surprising that at high IQ range, the correlation between the skills implicated in IQ test and skills implicated in, for example, theoretical physics, would decrease.

Download pages back online

The download pages (for those who purchased Polynomial via Plimus) should be back online. Unfortunately I lost the list of manual activations for the Steam customers, but I'm working on better integration with Steam anyway and it will be possible to download Linux version if you bought Polynomial from Steam.

Ion chamber

Ion chamber is an ionizing radiation detector that detects the radiation by measuring electrical conductivity of air (or other gas). I’ve built one:


In that video, I used old radium clock to test it.

 I wrapped the clock in plastic to ensure that the radioactive dust would not get out in the event that I accidentally drop it and break the glass - cleaning up radioactive dust can spoil your entire day.

For those into electronics, I am using LMP7721 amplifier and 100 TOhm resistor (for which I thank anonymous contributor – it is very hard to find those resistors!).

The circuit is basically the one from LMP7721 example datasheet, other resistors are for input protection, etc. I’ve soldered stuff onto prototype-ish board, which I soldered to flaps bent from the can that used to contain peanuts.

The chamber is formed by the can and the wire inside. The other can that used to contain peas is used as shielding (I opened it up for the picture). I simply measure the output with digital voltmeter. Besides radium clocks, this thing can detect 1kg bag of potassium salt.

The advantage of ion chamber over Geiger counter is that it works over the entire ionizing radiation range, with the response more closely matching that of the body. (Geiger counters are very bad at counting soft x-rays.) Other advantage is that it is just a ton simpler, requires no vacuum, can test for radon directly, etc. I’m planning to build another one using a switch to keep the measurement circuitry disconnected from the ion chamber electrode except for the times when measurements are taken. That should allow higher accuracy, and permit me to use more ordinary amplifier, like TL072 .

Changed my mind about moving the blog.

It's bit of pain in the backside to keep wordpress updated and correctly backed up, and I barely do any blogging anyway so its not worth the effort, especially as the blogger has improved the customization options. So I'll be blogging here from now on. On the downloads page outage: I am fixing it now.

Website outage

My site's webserver broke down, with loss of some data. The purchase data was backed up, but not the blogger (due to minor oversight with permissions). I am in process of restoring the customer records from Plimus (the backup was somewhat out of date so I need to restore them from there). All links will be working, but if you need to re-download something right now you can mail me.

Blog moved over to wordpress installation on my site.

Now on http://dmytry.com/blog . I'll move select few pages over to new blog over the time.

edit: Or actually maybe not. The ideal would be to automatically post on both. I'll see if I can make a script to do that.