Jürgen Schmidhuber

The best explanation for this paradox and one that should be especially convincing to computer scientists is that the kolmogorov complexity of any brain aware enough to have consistent memories is higher than the universe which contains it. A Boltzmann brain might not be impossible but it is far,far less likely than arbitrary universes occurring as fluctuations. It makes sense if you think about it. It's also consistent with the fact that 17th century mathematics could make so much headway unravelling the laws of physics and we're still fumbling trying to understand bacteria. Suppose you were playing with procedural generation. Which would be harder: a procedural universe, a procedural universe with a generated AI able to interact with it or an isolated AI which has learned to act optimally in some arbitrary but specific universe? It is not clear that the last program, meant to represent a Boltzmann Brain with a memory of interactions with the universe is any less difficult than the one with the complex generated AI. The isolated AI might be even harder since a dumb algorithm could, given enough time and feedback, produce something fairly intelligent. So it would take more bits to describe a brain with your specific memories than to specify a universe which could end up with someone similar to you and easiest of all is the program which generates all possible universes. Schmidhuber goes into this here: http://www.idsia.ch/~juergen/computeruniverse.html. Even if the…

You may be interested in an even older paper: http://www.idsia.ch/~juergen/icdl2011cuccu.pdf

No citation at all to: http://www.idsia.ch/~juergen/creativity.html ?

And let's not forget Marcus Hutter and Jurgen Schmidhuber, who use compression to explain general reasoning and human curiosity http://prize.hutter1.net/ http://www.idsia.ch/~juergen/interest.html

Well we have at least a hint on how to go about it software-wise. Jürgen Schmidhuber has been working on recursively self improving universal problem solvers for decades now: http://www.idsia.ch/~juergen/goedelmachine.html

You can see some real interesting cases of genetic algorithms being used to evolve neural networks on Evolino: http://www.idsia.ch/~juergen/evolino.html

Also, take a look at some of the work by J.Schmidhuber (http://www.idsia.ch/~juergen/). For example, the LSTM architecture is quite powerful. And he very successfully applied those RNN models to a wide number of areas (handwriting recognition, robotics, ...).

By "looping", I'm assuming you mean "recurrent"? If so, you might be interested these: http://www.idsia.ch/~juergen/rnn.html http://www.lsm.tugraz.at/learning/framework.html http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.31.6787

This type of thing has been done before: http://en.wikipedia.org/wiki/FatKat_(investment_software) http://www.idsia.ch/~juergen/finance.html http://www.technologyreview.com/blog/guest/25308/ I'm sure there are other examples. I think part of the problem is this guy thinks he can get rich quickly with just the right set of programming instructions.

Why should it be this network thing? He doesn't back up his assertion that the universe is described by one of these networks. Juergen Schmidhuber lays out an interesting idea that our Universe is just one Turing machine, generated by an enumeration of all Turing machines (generating all possible Universes). http://www.idsia.ch/~juergen/everything/html.html

AIXI-tl could use a CTC computer. As well as performing arbitrary computations in one tick, we could also keep a CTC going as a "reset button": whenever we realise that an incorrect/sub-optimal decision has been made, we can go back to when we started. This lets us 'brute force' our real-world interactions (ie. our IO), as well as "merely" gaining an infinite computing speed-up. This could be exploited by something similar to the "Success Story Algorithm" http://www.idsia.ch/~juergen/directsearch/node13.html . The CTC-AI would some way of "scoring" the world (ie. a fitness function), some probability of modifying its decision-making policy, and some non-zero probability of resetting the world. If the reset probability is anti-correlated to the world's score, worlds which score higher over time are more likely, hence the machine's decisions are more likely to increase the score than decrease it, hence the policy is more likely to improve over time.