AGI/FAI Theorist for Hire
I’m nearing the end of my employment at SIAI and looking for my next gig. If all else fails I will likely move back to the Bay area (I am currently in Japan) and take a job as a programmer somewhere. However, I would prefer to focus my attention directly on developing AGI and FAI theory. In addition to my current projects (described below), I can try to answer mathematical, philosophical, or other questions for a bit of cash. For some of my previous work, see my page on the arXiv.
In the past year I’ve been involved in two major projects at SIAI. Steve Rayhawk and I were asked to review existing AGI literature and produce estimates of development timelines for AGI. My work on this project got rather bogged down and proceeded slowly, although I did learn a lot and I’ve moved in the direction of predicting AGI soonish (5-20 years). After this I tried to produce an AGI technology demo for Google’s AGI-11 conference. I was unable to finish my demo in time for the submission deadline, and shortly afterwards SIAI decided to let me go.
I have several projects that I would like to move forward with, and if I can get adequate funds (about $1000 per month to ensure my survival, or $2000 to live comfortably) I will be able to work on them.
Current project ideas:
Continue development on my incomplete AGI project (optimally, technical details not to be published).
Write a paper on AGI models that can be used as a basis for FAI research (similar to the way AIXI and its ilk are used now, but closer to reality than AIXI).
Figure out how an AI can reason formally about using objects in its environment as tools for performing computations.
I’m also interested in repurposing machine learning algorithms used for finding plausible hypotheses about data distributions into algorithms for finding action policies with high expected utility.
I’m open to suggestions for other topics. I don’t consider myself an expert at empiricism, so I prefer to work in domains where I can reason formally. Some thing I’d be up for:
If you have informal questions or concerns, I can try to think of formal mathematical questions that are similar.
Once we’re dealing with a mathematical question, I can try to answer it.
If a question looks too hard for me to answer (as will often be the case), I can try to figure out exactly what is hard about it.
I’m also interested in writing problem sets. If you want to learn about some weird domain that no textbook exists for, I’ll try to figure out what some introductory problems in that domain would look like.
Prices for any of these services are negotiable. You can contact me here or at peter@spaceandgames.com.
While I’m not in a position to hire you, I think that this is an extremely important problem. In the case where the utility function is known, I think there is lots of low-hanging fruit that will lead to progress in more “physical” application areas of machine learning like computer vision and robotics. In the case where the utility function is unknown, I think the problem is harder (at the level say of a PhD thesis), but would be a crucial step towards making progress on FAI.
If you’re interested in talking to me about either of these then I’d be happy to, assuming you have enough of a statistical background for me to get my thoughts across without too much of an explanatory burden. Assuming you haven’t already decided on a specific set of algorithms, I have some ideas here that I don’t currently have time to pursue myself that I think could lead to a publication in a good machine learning journal if done well.
This timeline is much sooner than I would predict. Could you perhaps point me to a few sources that you think would cause me to update my estimate towards yours?
There was a specific set of algorithms that got me thinking about this topic, but now that I’m thinking about the topic I’d like to look at more stuff. I would proceed by identifying spaces of policies within a domain, and then looking for learning algorithms that deal with those sorts of spaces. For sequential decision-making problems in simple settings, dynamic bayesian networks can be used both as models of an agent’s environment and as action policies.
I’d be interested in talking. You can e-mail me at peter@spaceandgames.com.
You seem to suggest that this work is incomplete, but I’m curious: is this available anywhere or is it still a work in-progress? I would be very interested in reading this, even if its incomplete. I would even be interested in just seeing a bibliography.
It is not available. The thinking on this matter was that sharing a bibliography of (what we considered) AGI publications relevant to the question of AGI timelines could direct researcher attention towards areas more likely to result in AGI soon, which would be bad.
Seen Ben and Seth’s paper on this topic? It has a bibliography.