AlphaZero was one of the bigger headlines recently. Google’s new chess AI taught itself for 4 hours starting from a blank slate—no opening or endgame tables—and crushed Stockfish, the world’s previous best computer. See chess website articles here and here, a lichess.org collection of the games here, and the original research paper here via arXiv. This obviously has lots of real-world implications.
The most interesting thing is the way it won games. Ever since the early days of chess programming, we thought that chess computers could understand basic tactics but never deep positional play. Even in the pivotal 1997 Kasparov vs Deep Blue match, the human world champion famously said that Deep Blue must have been getting help from human grandmasters as it was playing non-computer-like moves.
Watching two chess AI’s play each other is typically a boring feat. But AlphaZero plays in a very human-romantic style, at least in the games that were revealed (and there’s definitely some selection bias there). AlphaZero often gave up lots of material for tempo, and it worked. One of the most talked-about positions is the following, where AlphaZero (white) abandons the Knight on h6 and plays Re1. It went on to win the game.
There’s lots of caveats in terms of how “real” of a result this is. Namely, the example games had Stockfish set on suboptimal settings. But still, it increases my opinion of the complexity of chess. As computers have gotten better, the way they play chess became more and more boring. But maybe the curve is not monotonic and we might have a stage where the game becomes more interesting again. Though I fear that eventually it will degenerate into optimal play from move one.
People have been talking about the Sam Altman blog post.
Earlier this year, I noticed something in China that really surprised me. I realized I felt more comfortable discussing controversial ideas in Beijing than in San Francisco. I didn’t feel completely comfortable—this was China, after all—just more comfortable than at home.
That showed me just how bad things have become, and how much things have changed since I first got started here in 2005.
It seems easier to accidentally speak heresies in San Francisco every year. Debating a controversial idea, even if you 95% agree with the consensus side, seems ill-advised.
More recently, I’ve seen credible people working on ideas like pharmaceuticals for intelligence augmentation, genetic engineering, and radical life extension leave San Francisco because they found the reaction to their work to be so toxic. “If people live a lot longer it will be disastrous for the environment, so people working on this must be really unethical” was a memorable quote I heard this year.
I don’t have any experience with the San Francisco discussion climate, but this seems weird. The fact that someone felt the need to write this post is a sign about the culture.
I’m probably way more in favor of politically incorrect ideas, mainly since I think the world vastly overvalues traditional ideas, and ironically because there is so much that you can’t say in China. Tyler Cowen points out, “…your pent-up urges are not forbidden topics any more. Just do be careful with your mentions of Uncle Xi, Taiwan, Tibet, Uighur terrorists, and disappearing generals.”
So Altman’s general point about politically incorrect ideas is probably correct. I don’t have any problem with discussing unpopular ideas. But I just don’t see people moving form San Francisco to China as a reasonable solution. There are certain topics that we might be overly sensitive to, but the overall level of idea tolerance would seem very tilted in favor of the US.
Obligatory shout out to 80000 Hours’ extinction risk article. The idea was to discuss various sources of extinction and estimate their chances of occurring.
What’s probably more concerning is the risks we haven’t thought of yet. If you had asked people in 1900 what the greatest risks to civilisation were, they probably wouldn’t have suggested nuclear weapons, genetic engineering or artificial intelligence, since none of these were yet invented. It’s possible we’re in the same situation looking forward to the next century. Future “unknown unknowns” might pose a greater risk than the risks we know today.
Each time we discover a new technology, it’s a little like betting against a single number on a roulette wheel. Most of the time we win, and the technology is overall good. But each time there’s also a small chance the technology gives us more destructive power than we can handle, and we lose everything.
An informal poll in 2008 at a conference on catastrophic risks found they believe it’s pretty likely we’ll face a catastrophe that kills over a billion people, and estimate a 19% chance of extinction before 2100.
As a trader, the first thing that comes to mind is to create some betting markets on such events happening and have a bunch of people trade, but this leads to weird selection effects and the payout is too long-term. So looking at some polls and mentally adjusting is probably right.
In addition, their ordering of what to prioritize is interesting:
- AI safety
- Global priorities research
- Building the effective altruism community
- Pandemic prevention
- Improving institutional decision-making
- Nuclear security
I should maybe have a recurring Twitter section. Anyway, here is a tweet by Julia Galef, and I’ve also wondered about this topic a lot.
The thought experiment I want to run is to throw together a racially diverse set of kids in a bubble, and expose the kids to roughly no knowledge of real world history or any hints of racism outside, and otherwise act like everything is normal. In this bubble world, would they start becoming racist against each other? I would guess no.
I think an underrated explanation in general of why people do something is because everyone else around them does it or that parents or teacher early on in their life do it. Social/cultural norm is a really strong incentive/disincentive for activities.
Cryptocurrencies and Electricity
There are definitely people worrying about the massive amount of world electricity consumption from bitcoin mining. Newsweek extrapolates that bitcoin will take up the world’s electric output by 2020. It’s currently at 0.15% according to some website. This is not small, giving how quickly it has been growing. Wired worries it will become the paperclip machine:
That’s bad. It means Bitcoin emits the equivalent of 17.7 million tons of carbon dioxide every year, a big middle finger to Earth’s climate and anyone who enjoys things like coastlines, forests, and not dying of mosquito-borne diseases. Refracted through a different metaphor, the Bitcoin P2P network is essentially a distributed superintelligence utterly dedicated to generating bitcoins, so of course it wants to convert all the energy (and therefore matter) in the universe into bitcoin. That is literally its job. And if it has to recruit greedy nerds by paying them phantom value, well, OK. Unleash the hypnocurrency!
I also stumbled upon a more optimistic viewpoint, claiming that bitcoin mining will trigger increased development and adoption of clean energy:
But electricity costs matter even more to a Bitcoin miner than typical heavy industry. Electricity costs can be 30-70% of their total costs of operation. Also, Bitcoin miners don’t need to worry about the geography of their customers or materials shipping routes. Bitcoins are digital, they have only two inputs (electricity and hardware) and network latency is trivial as compared with a truck full of steel. This particular miner moved an entire GPU farm across the U.S. because of cheap hydroelectric power in the Pacific Northwest and, in his words, “it’s worth it!” That’s also why we see miners in Iceland. Aside from beautiful vistas you can find abundant geothermal and hydraulic power in the land of volcanoes and waterfalls.
If Bitcoin mining really does begin to consume vast quantities of the global electricity supply it will, it follows, spur massive growth in efficient electricity production—i.e. in the green energy revolution. Moore’s Law was partially a story about incredible advances in materials science, but it was also a story about incredible demand for computing that drove those advances and made semiconductor research and development profitable. If you want to see a Moore’s-Law-like revolution in energy, then you should be rooting for, and not against, Bitcoin. The fact is that the Bitcoin protocol, right now, is providing a $200,000 bounty every 10 minutes (the bitcoin mining reward) to the person who can find the cheapest energy on the planet. Got cheap green power? Bitcoin could make building more of it well worth your time.
It’s very unclear in bitcoin’s case how good the upside is for the world, but it doesn’t seem anywhere close to being an extinction risk.
Recommended is Tyler Cowen’s post on crytocurrencies and social value.
I previously wrote that we take modern life improvements for granted and sometimes erroneously yearn for the hunter-gatherer life. Well here is a Quillette article on precisely the romanticization of that. Here are some examples:
In his later work, Lee would acknowledge that, “Historically, the Ju/’hoansi have had a high infant mortality rate…” In a study on the life histories of the !Kung Nancy Howell found that the number of infants who died before the age of 1 was roughly 20 percent. (As high as this number is, it compares favorably with estimates from some other hunter-gatherer societies, such as among the Casiguran Agta of the Phillipines, where the rate is 34 percent.) Life expectancy for the !Kung is 36 years of age. Again, while this number is only about half the average life expectancy found among contemporary nation states, this number still compares favorably with several other hunter-gatherer populations, such as the Hiwi (27 years) and the Agta (21 years). Life expectancy across pygmy hunter-gatherer societies is even lower, ranging from about 16-24 years, although this may have as much to do with pygmy physiology as with the hunter-gatherer lifestyle.
11 of these 15 societies have homicide rates higher than that of the most violent modern nation, and 14 out of the 15 have homicide rates higher than that of the United States in 2016. The one exception, the Batek of Malaysia, have a long history of being violently attacked and enslaved by neighboring groups, and developed a survival tactic of running away and studiously avoiding conflict. Yet even they recount tales of wars in the past, where their shamans would shoot enemies with blowpipes. Interestingly, Ivan Tacey & Diana Riboli have noted that “…the Batek frequently recount their nostalgic memories of British doctors, administrators and army personnel visiting their communities in helicopters to deliver medicines and other supplies,” which conflicts with the idea that hunter-gatherer societies would have no want or need of anything nation states have to offer. From 1920-1955 the !Kung had a homicide rate of 42/100,000 (about 8 times that of the US rate in 2016), however Kelly mentions that, “murders ceased after 1955 due to the presence of an outside police force.”
So, what explains the popularity of this notion of an “original affluent society”? Why do people in societies with substantially greater life expectancy, reduced infant mortality, greater equality in reproductive success, and reduced rates of violence, romanticize a way of life filled with hardships they have never experienced? In wealthy, industrialized populations oriented around consumerism and occupational status, the idea that there are people out there living free of greed, in natural equality and harmony, provides an attractive alternative way of life.
I also definitely live in a bubble, as I don’t know anyone openly in favor of hunter-gatherer society.
This also reminds me of Joseph Stiglitz’s book, The Price of Inequality. Most of the book is very methodical or at least numbers-driven. Then comes this absurd passage on the Bhutanese (p. 155 of the Norton edition):
Bhutan, the remote Himalayan state to the northeast of India, for instance, is protecting its forests as part of a broader commitment to the environment. Each family is allowed to cut down a fixed number of trees for its own use. In this sparsely populated country, I asked, how could one enforce such an edict? The answer was simple and straightforward: in our jargon, social capital. The Bhutanese have internalized what is “right” when it comes to the environment. It would be wrong to cheat, and so they don’t.
I’ve been waiting for years to quote this paragraph, but here it is. There is in general some weird sacred reverence of non-Western cultures. Is this related to the Altman political correctness theme? Can I just pick a well-off small community in America and say “it would be wrong to cheat, and so they don’t”? Anyway, it’s really easy to say some society works pretty well, and then take all the modern improvements for granted.