vetta project

– Henry Markram in this article.

Maybe, but I suspect he will be beaten by people working on ANNs.

I’ve been invited to give a talk next week at the Gatsby Unit in London. Gatsby is known around the world for its research into theoretical neuroscience, computational neuroscience and machine learning. It’s home to famous researchers such as Peter Dayan, David MacKay and, until recently, Geoffrey Hinton. I’ll also get to spend a day hanging out in London. This is going to be fun!

Just read an interesting little article on graphene transistors. I’d never even heard of them before, apparently it’s the stuff that’s in your pencil. Anyway, with a bit of work you can make amazing transistors with it.

…electrons move through graphene with almost no resistance, generating little heat. What’s more, graphene is itself a good thermal conductor, allowing heat to dissipate quickly. Because of these and other factors, graphene-based electronics could operate at much higher speeds. … de Heer says, “I believe we can do a terahertz…”

Once again Kurzweil has hit the mainstream media: this time an article on the BBC. It’s currently the “most emailed” article on the website, so it’s getting a good share of attention. Although he influenced me strongly many years ago, I’m not a big fan of his writings since 2000. In my view his visions of the future are now so extreme that they actually create sceptics among those who didn’t really have much of an opinion previously. This is a problem because if society doesn’t take the possibility of relatively near term AGI seriously, then it’s not going to direct resources towards the extremely difficult issues that surround this.

That aside, this event provides another opportunity to sample the responses to these sorts of ideas among a wider audience. Slashdot is an interesting place to observe this because they are a group who are enthusiastic about technology, but otherwise seem pretty mixed. There are a few supportive comments, but also plenty of comments like this:

“It is not too much of an overstatement to say that the field of AI has not significantly progressed since the 1980’s.” [here]

The comment gets more positive as it goes on (he suggests evolution to be the solution), but still, it’s this often repeated statement that AI research isn’t making any progress.

I find this kind of weird, because as I read the journals and go to conferences what I see is a constant flow of new ideas and results, ranging from the philosophical and theoretical down to record breaking results on all kinds of practical problems. Of course some problems are not progressing as fast as others and you can always point to those that are not going anywhere fast. But that’s a bit like saying that medicine is not going anywhere because a cure to cancer, AIDS and the common cold still haven’t arrived at your local pharmacy.

Personally, I find the rate of progress in AI to be so fast that it leaves me feeling dizzy. I struggle to keep up with all the interesting new ideas. In my subjective judgement, my expected (in a statistical sense) date for human level AGI is 2025, and I put my standard deviation at about 7 years. This estimate has been consistent since I first started asking myself this question in 1999, shortly after joining Webmind. I expect basic AGI to exist 5 years prior to human level AGI, and a primitive proof of concept AGI to exist perhaps 5 years before that (though I think very few people will recognise it as such when it appears). It will be interesting to see whether my beliefs remain consistent over the next 5 years.

I’m getting a steady trickle of positive feed back from universities around the world about my limits of prediction paper and the paper I wrote with Marcus Hutter on universal intelligence. I had imagined that this kind of work would be more readily accepted by the online communities as I’d guessed they would be more open to these new ideas. To my surprise it’s actually the academics, some of them quite well known, who are excited about this.

My PhD thesis is now going through the final check by Marcus and should soon go to my committee. It took such a lot of time and effort to write; remind me never to write another book. When I put together a print version on lulu.com it will probably come out at around 220 pages. If I decide to include some of the derivations needed for Solomonoff induction in an appendix, it could be 250.

With that out of the way I now have time to focus on building the AGI. The team working on the core computation engine over the last year have done a great job. It’s time to start constructing the first iteration of the core learning algorithm and universal intelligence test.

From a footnote in the paper I’m currently reading:

“The cubic drift of our share process is also closely related to that of the stochastic Ginzburg-Landau diffusion used in superconductivity physics to model phase transitions.”

You know, there really should be a ban on physicists going to work in mathematical finance…

2007 was quite a year. It started with two weeks vacation and a friend’s wedding in Rarotonga in the Cook Islands. That was an amazing experience. Two weeks with nothing to do except hang out with friends, soak in the sun, swim with all the tropical fish in the light blue waters of the lagoon, relax and think. I need a vacation like that every winter to de-stress and clear my mind.

Upon returning to Switzerland it was all on. First up, personal life issues that continued on for another 8 months… but I’m not getting into that here. Then there was the whole PhD to deal with. Research to finish, papers to write, and of course the thesis to write. I ended up putting more effort into this last bit of research than I had planned. It paid off with getting a publication into one of the top conferences in the world, followed by a 50 page journal publication. In all, a total of 5 publications for the year, which is  a pretty good haul.

After finishing this work I then started on my thesis full time, however about a month into that my PhD position ended, and so I switched to a post doc in finance at the university. Moving into a new field is not easy — before doing any research you need to catch up on what everybody else has been learning in their studies for the last few years. So that kept me pretty busy during the week. Some of the math was familiar, however I’d never studied stochastic calculus before and it turns out that this is the lingua franca of finance. It’s not an easy subject. With finance taking most of my week, evenings and weekends became thesis time. This slowed down my thesis writing a lot, and it also meant that I was working 7 days a week. I went months without having a day off. Becoming single near the start of all this was probably a blessing as I could never have finished everything otherwise.

It’s now the end of the year. I’m finally starting to get some financial simulations working that should grow into my research into asset pricing models. I’m playing around with the last few pages of the last chapter of my thesis, and my PhD supervisor is currently reading the rest. Hopefully I can clean up the remaining issues in the coming weeks and then send it off to my committee. With such a good publishing record I don’t expect the defense to be any problem.

Phew… what a year. But in the end, everything I had to do I got done, moreover, I did it well.

One of the great things about guys who start tech companies that make money is that they like to share what they learnt online. Here’s one article I recently read that was quite interesting.

… you spend two hours trying to understand the correct usage of semi-colons verses colons, when to capitalise following a colon, and then fixing up your text.

For those who really must know: go here and then here.