Collection of papers and articles that I’ve spotted since my previous links post that seem interesting.

  • Nuclear fusion is coming. Back in 2014, I was showing optimism about fusion, and the deadlines given by some of those companies aiming to get at commercial power plants were somewhat too optimistic. The revised dates, four years later don't look that awful though: fusion is not always 20 years into the future anymore:

    • General Fusion wanted to have their reactor in 2020 (2014), now they claim commercialisation within ten years (<2028)
    • TAE was aiming for 2020-2025, and they seem to be on track, saying they have plans to begin commercialisation in 2023.
    • Commonwealth Fusion Systems is the one 20 years in the future, but wasn't around when I wrote the original post.
    • LPP has made some progress, but their unique approach doesn't seem close to commercial grade yet.

  • New paperargues that the decline in "dynamism" of the US economy can be explained quite well by lower labour force growth, without need to resort to a market power story.

  • How much is your data worth to large corporations like Facebook? Basically nothing. As you might expect.

    • This is why I think the whole "data as labor" side of the Radical Markets framework is by far the weakest. Sure, in the aggregate data may be very valuable, but this is the diamonds/water paradox again, and while one can exploit corporations by means of having user cartels (what Lanier calls MIDs), why would you want to do that?
    • Eline Chivot has a good writeup of the flaws of this proposal.

  • Not many critiques of Stubborn Attachments have been penned, but here is one. (Not that I agree with all of it, overall I like SA a lot!) Relevant quote:

    • There is a problem that I haven’t seen directly addressed by political philosophers, but which I suspect is a difficult and important one (caveat: I’m far from an expert on political philosophy. It’s entirely possible that it has been addressed and I just haven’t seen it. If so, oops on this whole section). Suppose I am asked for my political philosophy and I say: “I’m an Awesomist. I believe in a political system in which everyone has what they need to live awesome lives”. Intuitively, we don’t want to disagree with this position; rather, we want to say that it isn’t a serious position at all (if it were, we should all be Awesomists!). But if we’re going to say Awesomism isn’t a real position, we need a reason. What requirements for being a serious position does it fail to meet?

    • Or to put it in other words: The Stubborn Attachment frameworks is heavily underdetermined. For example, if one considers inequality, how much inequality should there be? My answer to this, regarding economic growth: it doesn't matter. True it is that, say, redistribution has been found to be detrimental in some studies, but I wouldn't be surprised that there are ways to do it that are growth enhancing or growth neutral.

    • Tyler has often referred readers to his blog for specific policy proposals. Noah Smith read the book as implying that there should be more government investment in basic research and green tech.

    • Growth policy has to be seen in the light of the fact that economic growth seems quite constant over the long term, once you have your economy going. I mentioned in my presentation on Mazzucato's entrepreneurial state that after WWII, there was indeed a massive increase in funding of US basic science. What did that do to growth, or for that matter improvement in specific technologies? Nothing. Isn't this surprising? I looked at the literature on science funding and various outcomes years ago but I think I didn't produce a post about that specific narrow issue. I should. In the meantime, you can have my summary of the handbook of the economics of innovation

  • Sabine Hossenfelder gets an opinion piece at the NYT, argues against CERNpropaganda. Arguing against funding science, even if it's a very specific case (a new supercollider) is highly unpopular, on par with arguing for less healthcare or less education funding, yet this is the correct view:a new collider should be put on holdfor now.

  • A discussion of optimal taxation: When some economists talk about X being the optimal tax rate, where does that come from?

  • DeepMind beatsStarcraft 2, making me lost a bet that it would not happen in 2018 (Sad!). What's next? Here I suggested years ago a few examples: Magic The Gathering, automated engineering, or entrepreneurship. But at this stage, one thing is increasingly clear: If one has access to a fixed, easily knowable, set of rules, and has a lot of computational power then the setting in question will eventually fall. This makes games easy prey for AI. In real world situations, this is harder and you can't train an agent with 200 hundred years worth of experience. If you want to, for example, produce music, what are you trying to optimize? You can try to copy and learn a distribution, like GANs do, but the ultimate value function is in the heads of people, and in what people think is good or bad music. And sure you can get input from people, but not at arbitrarily high speeds, except perhaps if one reads brainwaves as the music is being produced and adjusts the music accordingly. For example, imagine one wants to have an automated CEO. What's the simulator to plug your system to? What rules? This does not imply it can't be done, it's just to say that the current approaches wouldn't work.

    • Edit: See the comment section for more discussion

  • Book: The Revolt of the public.

  • To highlight the role that reasoning quality plays in moral judgments, we review literature that he did not mention showing that individual differences in intelligence and cognitive reflection explain much of moral disagreement.

  • What happens if you take a dataset and run 20,000 different analyses on them (different specifications, variables, etc)? What if the datasets are large, with thousands, or millions of observations? Well, you get lots of different results, of course. But this paper is interesting because the authors try to quantify and visualize the impact of a given methodological choice. This is something I'm strongly in favor of, quoting myself:

    • For most questions we ask there are many interpretations. For example, drawing from my recent posts, we could ask “Does communism work?”. And there we go into “But what do we mean by communism?”, “What do we mean by work?”. Then we get into “Which countries count as communist?”, “Which measures should we used to think about ‘working’?”, “Which time period?”. Etc. This happens because for any concept[1], there is no clear-cut definition for it (Huemer, 2015). Things have particular qualities and concepts are fuzzily defined clouds around them so that we can generalise. Sketching their boundaries and working with concepts is a true art.[...] So what I do is try to get different interpretations of the meanings and see if we can say something ‘For most plausible interpretations’ or if we have to resort to ‘In this sense, this happened, in that other sense, these other things happened’.

    • Code and data for the study available here

  • Book: The Oxford Handbook of Megaproject management. (I have not read it)

  • Gwern's list of ordinary life improvements

    • I would add: Deliveroo and similar services, Amazon, fintech banks, or social networks like twitter and facebook

  • Tokyo: Where public transport and private transportare both adequately priced by market mechanisms. For example, parking or large inner city motorways are not subsidised or publicly provided, but neither is public transit subsidised.

    • More japanese transport goodness, thread

  • Low level laser therapy: a form of alternative medicine that seems to work for some cases. It consists on shining low powered lasers on the skin. This sounds like it shouldn't work, but apparently it does? (e.g. here's an assessment of RCTs for arthritis pain relief, including placebo controls)

There are a few other links that I would normally link here, but those will show up instead in an upcoming post about blockchain and crypto.

https://medium.com/@jjmstark/the-year-in-ethereum-87a17d6f8276

Comments from WordPress

  • Rational Feed – deluks917 2019-01-27T14:02:52Z

    […] Links 24 by Artir – Links. Topics include AlphaStar, Nuclear Fusion and Stubborn Attachments. […]

  • Simon Simon 2019-01-26T22:44:03Z

    Hey! Thank you for the links. I'm pretty excited about fusion.

    About DeepMind beating StarCraft 2... that's not quite a done thing yet. Ignoring the older version of AlphaStar that beat TLO (who was playing at a high but not professional level), it's pretty clear that the AlphaStar agents that went 5-1 against MaNa won at least two of the games using inhuman mechanics rather than tactical or stategic thinking. In fact, AlphaStar didn't demonstrate much strategic thinking at all (i.e. Building unit X as a counter to seeing the opponent building unit Y).

    I recommend this analysis: https://youtu.be/sxQ-VRq3y9E?t=211

  • Artir Artir 2019-01-27T11:43:45Z

    Also this https://medium.com/@aleksipietikinen/an-analysis-on-how-deepminds-starcraft-2-ai-s-superhuman-speed-could-be-a-band-aid-fix-for-the-1702fb8344d6 I took DeepMind's assertion that the match was fair at face value, but I might have to reconsider.