Carving Up Reality.

In a wonderful choice of words, the Wall Street Journal yesterday mentioned in an article that Agilent Technologies Inc. "[...] makes machines to analyze DNA, chemicals, sound waves and other items [...]".

Not to mention that *Kolmogorov complexity is completely irrelevant to intelligence*.

Michael Vassar, the President of the Singularity Institute for Artificial Intelligence recently gave an interview on Accelerating Future where he favorably mentions Marcus Hutter's work on AI:

AF: Why should someone regard SIAI as a serious contender in AGI?

Vassar: The single biggest reason is that so few people are even working towards AGI. Of those who are, most are cranks of one sort or another. Among the remainder, there is a noticeable but gradual ongoing shift in the direction of provability, mathematical rigor, transparency, clear designer epistemology and the like, for instance in the work of Marcus Hutter and Shane Legg. To the extent that SIAI research and education efforts contribute to rigorous assurance of safety in the first powerful AGIs, that is a victory as great as the creation of AGI by our own researchers.

Now that's an interesting contrast with earlier statements by Eliezer Yudkowsky and Ben Goertzel, co-founder of and Director of Research at the SIAI, respectively:

I seriously do NOT think there is any practical value to be gotten out of trying to create a pragmatic AGI system by "scaling AIXI down." Ben Goertzel, 2007 http://www.mail-archive.com/singularity@v2.listbox.com/msg00509.html 

To sum up: (a) The fair, physically realizable challenge of cooperation with your clone immediately breaks the AIXI and AIXI-tl formalisms. (b) This happens because of a hidden assumption built into the formalism, wherein AIXI devises a Cartesian model of a separated environmental theatre, rather than devising a model of a naturalistic reality that includes AIXI. (c) There's no obvious way to repair the formalism. It's been diagonalized, and diagonalization is usually fatal. The AIXI homunculus relies on perfectly modeling the environment shown on its Cartesian theatre; a naturalistic model includes the agent itself embedded in reality, but the reflective part of the model is necessarily imperfect (halting problem). (d) It seems very likely (though I have not actually proven it) that in addition to breaking the formalism, the physical challenge actually breaks AIXI-tl in the sense that a tl-bounded human outperforms it on complex cooperation problems. (e) This conjectured outperformance reflects the human use of a type of rational (Bayesian) reasoning apparently closed to AIXI, in that humans can reason about correlations between their internal processes and distant elements of reality, as a consequence of (b) above. Eliezer Yudkowsky, 2003 http://www.mail-archive.com/agi@v2.listbox.com/msg00862.html 

AIXItl is a different story. It's computable, and is vastly less useful than Novamente. It's a ridiculous algorithm really, since at each time step it searches an infeasibly large space of possible programs. It's useful purely for theoretical purposes. Ben Goertzel, 2003 http://www.mail-archive.com/agi@v2.listbox.com/msg00765.html 

Not to mention that *Kolmogorov complexity is completely irrelevant to intelligence*. Eliezer Yudkowsky, 2008 http://www.sl4.org/archive/0811/19505.html

Solomonoff Induction Breaks Egan's Dust Theory

Greg Egan’s 1994 novel Permutation City, which features unlikeable characters, wooden dialogue, and a depressing storyline, is one of the most thought-provoking works of science fiction ever written. It’s basically a book-length expansion of Egan’s “Dust Theory”. The related Church-Turing thesis implies that I couldn’t know whether I’m made of real atoms or just accurate computer simulations of atoms. The Dust Theory expands this to the case where the output of the atom-simulation undergoes a permutation – I still couldn’t tell what’s happening in the “basement”. Since any pattern of sufficient length can be permuted to a simulation of my atoms, and therefore my subjective experience, I can never discern from the “inside” whether I’m made of atoms, of simulated atoms, or of a random pattern of black-and-white flowers in a field on a small planet orbiting Betelgeuse.

My argument against the dust theory is that it does not explain anything. I believe I’m made of atoms because that explains a lot, that is, it compresses a description of my perceptions given my actions. (This is an informal paraphrasing of Solomonoff induction.) In fact, I believe I’m Manuel, who is such-and-such a type of guy, because it explains an awful lot of the stuff I’m perceiving, and doing. A world model with a “basement” not of physical atoms, but simulated atoms on a small turing machine, has about the same Kolmogorov complexity as the original model, so my take on that is “who knows?”. But if a theory makes it necessary to specify an extra permutation in the end ... if the permutation is to be Martin-Löf random, its complexity is to be about equal to the length of the string to be scrambled. Whoah, that’s a lot of extra bits! Each extra bit reduces the theory’s prior probability by 50%, so that’s pretty much off the table.

That’s also why I don’t buy into the “We are in Digits of Pi” theory. Granted, pi itself has a small Kolmogorov complexity, but in order to explain my perceptions and actions, in sum N bits, one would have to specify a region that lies some 2^N digits behind the comma. That’s much more costly (N bits) than the “atom” or “Turing machine” based theories above (K(N) bits), and is therefore, by virtue of Solomonoff induction, a stillborn theory.

One of the reasons Egan’s Dust Theory is appealing at first glance is that he introduces it through permutations of low Kolmogorov complexity which nevertheless look “complex” to the human mind. (The general case, which he – I think –doesn’t explicitely state, is known as the pseudo-random number generator.) The big step from there to arbitrarily complex permutations – almost all seemingly random patterns cannot be created with a pseudo-random number generator – is swept under the argumentative rug. I admit the sweeping is not done deliberately, as Egan doesn’t seem to know about Solomonoff induction.

For the record I do believe in Tegmark’s mathematical universe theory. I also believe my laptop’s harddisk contains mostly random data (courtesy 7zip, matroska, and others.) And, yes, I also believe a tiny fraction of myself is in a field of lowers somewhere (not Betelgeuse). More on this soon, hopefully, in a post I’ve been struggling to write for two years.

Michel Djerzinski

On March 27th, 2009, in the early afternoon, he went to the main post office in Galway. He sent one copy of his manuscript to the French Academy of the Sciences in Paris, and another one to the British journal Nature. What happened thereafter remains a mystery. The fact that his car was found close to Aughrus Point naturally lead to speculations about suicide - something that came to no surprise to Walcott and the technicians at the center. [...] Many witnesses attest to his fascination with this distant edge of the Western world, constantly bathed in a soft, shining light, where he had come so often, where, as he wrote in one of his last notes 'the sky, the sea, the light converge.' We believe that Michel Djerzinski went into the sea.

Michel Houellebecq's Atomised is one of the books most dear to me. I recognize myself in the novel's main character, Michel Djerzinski, who shares many biographical aspects and character traits with me, like, to give a few examples, our close relationships to our grandmothers, our research work in biophysics, our grim view of the human condition, and our relentless attempts to engineer a posthuman species of sentient beings.

As you can read above, today is the day where Djerzinski, having completed his breakthrough theory of SENS, disappears "into the sea". (The book was published in 1998.) I originally had plans of traveling to Galway on the occasion, maybe stay for a few days at the coastguard station outside of Clifden, where Djerzinski took residence during the last years of his life, and take in the atmosphere. Nevertheless, I decided otherwise.

For a certain kind of books, and a certain kind of young men, there exists a considerable risk that the books seriously mess with the young men's self-perception. My guess is that Atomised is one of those books. I love Atomised, and I see an almost creepy similarity between myself and the protagonist, but I clearly understand today that I am not Michel Djerzinski, and should feel no need to resemble him even more than I already do. What helped me understand this (besides from aging ten years), was coming across several other fictional works in which I could also profoundly recognize facets of myself; Thomas Bernhard's Ungenach, and, more recently, Makoto Shinkai's 5 Centimeters Per Second, are just two examples.

So rest in peace, Michel, and thanks for leading me along the way for a while.

Hibernation Redux

The local newspaper ran a story a few days ago on the accumulated hours of sunshine that we had in Wels, Austria (where I'm located) from Dec. 21st and March, 21st.

What made the fact newsworthy was that the number, due to constant fog and cloud cover, came down to seventy. The nearby city of Linz got 140 hours.

Just to put this into proper context: According to the statistics available from the BBC, the long-term average of total hours of sunshine for January, February, and March, is 180 for St. Petersburg, 210 for Rekyavik, 270 for Stockholm, 420 for Fairbanks, and 600 for San Francisco.

No surprise I've been in deep hibernation mode once again ...

Quote Of The Day

"The Japanese government outlawed the practice of self-mummification in the late 19th century."

(Found on Pink Tentacle.)

The Leibniz Drive

Every aspiring Mad Scientist must invent at least one scheme for an FTL drive. So here's mine.

My proposal does not use any speculative physics, such as wormholes or large amounts of negative mass. It is based 100% on physical laws we know today. What I propose, however, is the application of molecular nanotechnology on a very large scale.

First let's make it clear what we want:

Goal: A galactic empire in which Cpt. Cabonza and his motley crew can travel from Arghra V to Balubius II in one day.

We now add the following insight (which, as far as I know, has not yet the status of a physical law)

Insight: Information cannot travel faster than the speed of light.

We derive a conclusion from that:

Conclusion: Cpt. Cabonza's journey must not carry any information from Arghra to Balubius.

Now we know what we have to do in order to construct a scenario where FTL travel is possible:

Plan: Make physical reality an epiphenomenon of a pre-synchronyzed, parallel, redundant, computing process.
In practice that means each star system consists solely of programmable matter. The surface dynamics of that programmable matter, and that of any star system reachable by FTL drive, is pre-computed in a "hidden layer". To repeat: all connected star systems simulate the whole of the "Empire", but control only the "local" manifestations of physical reality. This includes any inhabitants, sentient or otherwise. Balubius can therefore let Cpt. Cabonza pop out of immaterial nothingness one day after Arghra disassembled him (motley crew and all).
Information coming in from outside the empire to Arghra can potentially break this pre-established harmony. It is therefore necessary to funnel any such information through "gateways", which retain the physical carrier of the information for as long as it takes to disclose the information to all connected system (in the case of a galaxy, ~50.000 -100.000 years). After this time, the carrier is released, and is allowed to interact with the "pre-informed" star systems. This does not mean the carrier has to come to a stop - it just has to cross the gateway region sufficiently below lightspeed. (I just retconned the Zones of Thought!)
I will add a FAQ here as questions come up.