Scattered notes and afterthoughts.
Spending two days with an elite crowd of thinkers is extremely inspiring. Just spending time talking to my personal heroes Eliezer and Robin justified the ticket cost. Met some wonderful youngsters. Highlights:
- Laughing about how incredibly nice it will be when augmented reality glasses can identify people you recognize but have forgotten with a friendly fellow who does machine learning on sensor data at a company he started in San Diego
- Admiring the stained glass in Grace Cathedral with several people my age while Fermi estimating the sun’s power output in kilowatts (hint: don’t start from the power/sq ft of sunlight that hits earth)
- Receiving a primer in how timeless decision theory defeats causal decision theory in the iterated prisoner’s dilemma from a London-accented lad who promotes efficient altruism in Toronto
- Having a dialect concerning the best mathematicians of all time, based on their contribution to the mathematics involved in technological acceleration (Turing was up there) with an Australian kid no older than myself who was visiting to teach some math to some of the researchers.
Best part, taken from the pre-summit party on Thursday:
Scene: 9:30pm, Daniel, Eliezer Yudkowsky, Robin Hanson, and three other LB/OWs standing on a damp porch outside a suburban house in Berkeley. It has just stopped drizzling.
The group has just finished considering a theory of one very emphatic talker concerning the cryptography involved in bitcoins. We get to discussing the gold standard.
“The way to maximize the value of precious metals has been well established.” I said dismissively. “You dig a huge hole and bury it.” I proceed to get laughs from both Eliezer and Robin.
Ok, you had to be there.
And the talks. Incredible line-up. TL;DR Jaan Taallinn > Peter Norvig > Robin Hanson > the rest
Ray Kurzweil: This talk was remarkable mostly due to the fact that his theory of how the neocortex works – sparsely distributed, layered, hierarchical – was indistinguishable from Jeff Hawkin’s theory as outlined in his great book On Intelligence. That is a good sign in my book. He also showed a lot of exponential curves with some really fun-looking acceleration. He emphasized many times that his famous projections from thirty years ago have been proven accurate.
Peter Norvig: The last of the conference, this talk was certainly the most inspiring. All he did was go over his predictions from his talk at the 2007 Singularity Summit and outline the progress he’s made with the Google superteam of AI hotshots. He was correct on 5 predictions out of 6, being wrong only in forecasting that the Internet would become a significant factor in AI research. He explained the basic science behind Google’s famous unsupervised youtube image classifier. During the Q&A he mentioned that his team trains the programs via batch, i.e. one long training session on a fixed data set, but that they would ideally prefer it to continually crawl the web and learn, all the while deciding how to expend resources between searching, storing, and classifying. Peter’s calm optimism was enchanting. He spoke as if saying, “Artificial intelligence? Oh, we’re working on it. Yeah we’ve had some major breakthroughs in the past few years. We’ll have more soon. Want to see the cool stuff we made?”
Robin Hanson: This was the talk I was most excited to see, having missed him present this material in Redwood City a few months ago. He did a very basic, but extremely compelling, microeconomic analysis of the advent of simple whole brain emulations, or “ems” for short. He predicts increased urbanization, as the information delay due to distance becomes severe on very quickly operating ems. He predicts a rapid reduction in wages, but at the same time an economy that at least doubles every few years. It was left unclear which of these two factors was stronger, and weather wealth/capita would increase. He predicts massive change to company structure with the elimination of the constraint that managers cannot work more than 100 subjective hours a week. He acknowledges that these predictions apply only to the short period of time after the invention of whole brain emulation and before brains can be algorithmically improved. I especially enjoyed this talk because, coming from a neuroscience background, I think that Whole Brain Emulation will come much before Artificial General Intelligence.
Jaan Taalinn: Jaan completely outdid the other speakers in his presentation. He told, via an animated comic strip, the story of Fred, a software engineer who goes on a quest to understand how it is that he is helping to usher the universe into a new era of intelligence. Along his way he meets philosophers and even a cosmologist who present new information and theories to our hero. The general point made to Fred is that he shouldn’t be surprised that he’s inventing the future, based on bizarre arguments that draw heavily on the anthropic principle and the possibility of living in a simulation. I don’t generally understand anthropic reasoning and this was no exception. It did however accomplish the goal of inspiring forward thinking and planning, and I think inspiration was second only to networking as the primary value of this event.
Some of the other talks were excellent as well, but I didn’t find them particularly relevant to thinking about the singularity. Steven Pinker’s talk about the decline of violence was wonderful, though I couldn’t figure out why he was delivering it to us. Maybe he just likes us and wanted to raise the profile of the event? I’d like to think he chose to speak at the summit because he knew just how many game-changers were in the audience. Vernor Vinge, Melanie Mitchell, Carl Zimmer, and Daniel Kahneman were the remaining celebrity speakers, and I think only Peter Thiel (who helped found the summit in 2006) and Elon Musk could have improved the lineup. Actually, I would have loved to see John Conway. One can always dream for next year.
All told it was a delightful weekend. It is marvelous to see the people who seriously address rapid technological advancement all under one roof. Granted, the work of the Singularity Institute has the potential to be worthless just as it has the potential to save our species. It comes down to having the comfort of knowing that if there is something to be done, namely developing Friendly Artificial Intelligence in advance of other types of AGI, many of our best and brightest are working on it. Now, if only I can donate enough money for them to hire one or two more mathematicians…