- Statistics prof Olle Häggström collects his blog posts about Superintelligence.
- “Governing cognitive biases: case studies of the use of… behaviorally informed policy tools.”
- Haidt & Tetlock & company, “Political diversity will improve social psychological science” (forthcoming in BBS).
Peter McCluskey pointed me to a nice explanation by Brian Greene of an experiment that could theoretically distinguish the Copenhagen and Many Worlds interpretations of quantum mechanics. This is from The Hidden Reality, ch. 8, endnote 12:
Here is a concrete in-principle experiment for distinguishing the Copenhagen and Many Worlds approaches. An electron, like all other elementary particles, has a property known as spin. Somewhat as a top can spin about an axis, an electron can too, with one significant difference being that the rate of this spin—regardless of the direction of the axis—is always the same. It is an intrinsic property of the electron, like its mass or its electrical charge. The only variable is whether the spin is clockwise or counterclockwise about a given axis. If it is counterclockwise, we say the electron’s spin about that axis is up; if it is clockwise, we say the electron’s spin is down. Because of quantum mechanical uncertainty, if the electron’s spin about a given axis is definite—say, with 100 percent certainty its spin is up about the z-axis—then its spin about the x- or y-axis is uncertain: about the x-axis the spin would be 50 percent up and 50 percent down; and similarly for the y-axis.
Imagine, then, starting with an electron whose spin about the z-axis is 100 percent up and then measuring its spin about the x-axis. According to the Copenhagen approach, if you find spin-down, that means the probability wave for the electron’s spin has collapsed: the spin-up possibility has been erased from reality, leaving the sole spike at spin-down. In the Many Worlds approach, by contrast, both the spin-up and spin-down outcomes occur, so, in particular, the spin-up possibility survives fully intact.
To adjudicate between these two pictures, imagine the following. After you measure the electron’s spin about the x-axis, have someone fully reverse the physical evolution. (The fundamental equations of physics, including that of Schrödinger, are time-reversal invariant, which means, in particular, that, at least in principle, any evolution can be undone. See The Fabric of the Cosmos for an in-depth discussion of this point.) Such reversal would be applied to everything: the electron, the equipment, and anything else that’s part of the experiment. Now, if the Many Worlds approach is correct, a subsequent measurement of the electron’s spin about the z-axis should yield, with 100 percent certainty, the value with which we began: spin-up. However, if the Copenhagen approach is correct (by which I mean a mathematically coherent version of it, such as the Ghirardi-Rimini-Weber formulation), we would find a different answer. Copenhagen says that upon measurement of the electron’s spin about the x-axis, in which we found spin-down, the spin-up possibility was annihilated. It was wiped off reality’s ledger. And so, upon reversing the measurement we don’t get back to our starting point because we’ve permanently lost part of the probability wave. Upon subsequent measurement of the electron’s spin about the z-axis, then, there is not 100 percent certainty that we will get the same answer we started with. Instead, it turns out that there’s a 50 percent chance that we will and a 50 percent chance that we won’t. If you were to undertake this experiment repeatedly, and if the Copenhagen approach is correct, on average, half the time you would not recover the same answer you initially did for the electron’s spin about the z-axis. The challenge, of course, is in carrying out the full reversal of a physical evolution. But, in principle, this is an experiment that would provide insight into which of the two theories is correct.
I’m not a physicist, and I don’t know whether this account is correct. Does anyone dispute it?
Further references on the subject are at Wikipedia.
In any case, such an experiment seems far beyond our reach. But since I’m Bayesian rather than Popperian, I put substantially more probability mass on MWI than Copenhagen even in the absence of definitive experiment. 😉
My apologies in advance to the computer science journalists I haven’t found yet, but…
Why is there so little good long-form computer science journalism? (Tech journalism doesn’t count.)
Several others sciences attract plenty of writing talent as well. Physics has Sean Carroll, Stephen Hawking, Brian Greene, Kip Thorne, Lawrence Krauss, Neil deGrasse Tyson, etc. Psychology has Steven Pinker, Richard Wiseman, Oliver Sacks, V.S. Ramachandran, etc. Medical science has Atul Gawande, Ben Goldacre, Siddhartha Mukherjee, etc.
Outside Aaronson and Hayes, I mostly see tech journalism, very brief CS news articles, mediocre CS writing, and occasional CS articles and books from good writers who cover a range of scientific disciplines, such as
- The Proof in the Quantum Pudding by Erica Klarreich
- Approximately Hard by Erica Klarreich
- The Future Fabric of Data Science by Jennifer Oullette
- The Mathematical Shape of Things to Come by Jennifer Oullette
- The last few chapters of The Code Book by Simon Singh
- Turing’s Cathedral by George Dyson
- Nine Algorithms That Changed the Future by John MacCormick
- The Golden Ticket by Lance Fortnow
Maybe CS is too mathematical to attract general readers? Too abstract? Too dry? Or simply not taught in high school like the other sciences? Or maybe there are problems on the supply side?
- The Center for Effective Altruism reports on outcomes from their 10+ meetings with UK policymakers so far.
- Pinker on Ivy League education (very good).
- A profile of Martine Rothblatt: “Futurist, pharma tycoon, satellite entrepreneur, philosopher. Martine Rothblatt, the highest-paid female executive in America, was born male. But that is far from the thing that defines her. Just ask her wife. Then ask the robot version of her wife.”
- Okay, good, so I won’t read the new Fukuyama books.
- AI evaluation: past, present, and future.
- Google just got serious about building a quantum computer. (Martinis, not D-Wave.)
- Chalmers’ guidelines for constructive debate and discussion. I’ve been wanting to write a post about the epistemic virtue of kindness, but this will do for now.
Lobbying and Policy Change by Baumgartner et al. is the best book on policy change I’ve read. Hat tip to Holden Karnofsky for recommending this and also Poor Economics, the best book on global poverty reduction I’ve read.
LaPC is perhaps the most data-intensive study of “Who wins in Washington and why?” ever conducted, and the data (and many follow-up studies) are available from the UNC project website here. One review summarized the study design like this:
To start, [the researchers] sample from a comprehensive list of House and Senate lobbying disclosure reports to identify a random universe of participants. After initial interviews with their sample population, the authors assemble a list of 98 issues on which each organizational representative had worked most recently [from 1999-2002, i.e. during two Presidents of opposite parties and two Congresses]. These range from patent extension to chiropractic coverage under Medicare, some very broad and some very specific. Interviewers endeavored to determine the relevant sides of each issue and identify its key players. Separate subsequent interviews were then arranged where possible with representatives from each side of the issue…
With this starting point, the researchers followed their sample of issues for several more years to track who got what they wanted and who didn’t.
Note that their issue sampling method favors issues in which Congress was involved, so “issues relating to the judiciary and that are solely agency-related may be undercounted.”
LaPC is a difficult book to summarize, but below is one attempt. Some findings were surprising, others were not.
- One of the best predictors of lobbying success is simply whether one is trying to preserve the status quo, and in fact the single most common lobbying goal is to preserve the status quo.
- Some issues had as many as 7 sides, but most had just two.
- Most lobbying is targeted at a small percentage of issues.
- Very few neutral decision-makers are involved. Where government officials are involved, they are almost always actively lobbying for one side or another. 40% of advocates in this study were government officials; only 60% were lobbyists.
- Which kinds of groups were represented by the lobbyists? 26% were citizen groups, 21% were trade/business associations, 14% were corporations, 11% were professional associations, 7% were coalitions specific to an issue, 6% were unions, and 6% were think tanks.
- The most common lobbying issues were, in descending order: health (21%), environment (13%), transportation (8%), science and technology (7%), finance and commerce (7%), defense (7%), foreign trade (6%), energy (5%), law, crime, and family policy (5%), and education (5%).
- When lobbying, it’s better to be wealthy than poor, but there’s only a weak link between resources and policy-change success.
- Policy change tends not to be incremental except in a few areas such as the budget. For most issues, a “building tension then sudden substantial change” model predicts best.
- There is substantial correlation between electoral change and policy change, and advocates have increasingly focused on electoral efforts.
If you’re interested in this area, the next book to read is probably Godwin et al’s Lobbying and Policymaking, another decade-long study of policymaking that is largely framed as a reply to LaPC, and was recommended by Baumgartner.
- Max Tegmark and Eliezer Yudkowsky on AI goal retention / ontological crises.
- Superintelligence is out in the USA, and in audiobook and Kindle formats. MIRI is hosting an online reading group for the book.
- Nature on brilliant scientists who leave academia.
- FP provides an update on Honduras’ charter cities.
Some decent books:
- Greenwald, No Place to Hide
- Isaacson, Benjamin Franklin: An American Life
- Levinson, The Box
- Bloom, Just Babies
I also went through Gendlin’s Focusing and Cornell’s The Power of Focusing, which describe the latest unsupported “woo” technique for self-help that I’m experimenting with because Try Things. I don’t know yet how helpful this method is for me.
- “Exploratory engineering in artificial intelligence” for Communications of the ACM (w/ Bill Hibbard)
- Two talks for the 2014 Effective Altruism Summit: “MIRI Intro” and “Steering the Future of AI.”
- Aaronson, “Could a quantum computer have subjective experience?”
- Against Empathy, a debate between Paul Bloom, Peter Singer, Sam Harris, Jesse Prinz, and others.
- Obstacles to self-driving cars.
- “Physical principles for scalable neural recording” maps the physical limits on brain activity mapping for several different methods. Potentially informative for whole brain emulation forecasting.
- Humor: Thanks, Textbooks is a Tumblr for amusing, weird textbook writing. For example, here’s one textbooks “colloquial” explanation of the 2nd law of thermodynamics: “You can’t shovel manure into the rear end of a horse and expect to get hay out its mouth.”
- Academic urban legends, a study.
- Markov, “Limits on fundamental limits to computation.”
- Funny: An honest political ad for Gil Fulbright.
- Well-produced explanatory video on automation and technological unemployment. Not enough caveats, but of course it’s more engaging that way.
- Molyneux, “How the (hard) problem of consciousness could emerge in robots.”
- And, an overview of reductionist approaches to consciousness, by James Reggia.
- Bensinger, Groundwork for AGI Safety Engineering.
- The Good Country Index.
- The Economist reviews Superintelligence.
- New from Tetlock et al., “Forecasting tournaments: tools for increasing transparency and improving the quality of debate.”
Poor Economics is the best book I’ve read on poverty reduction. The book ends with a summary of its key lessons. Here they are:
As this book has shown, although we have no magic bullets to eradicate poverty, no one-shot cure-all, we do know a number of things about how to improve the lives of the poor. In particular, five key lessons emerge.
First, the poor often lack critical pieces of information and believe things that are not true. They are unsure about the benefits of immunizing children; they think there is little value in what is learned during the first few years of education; they don’t know how much fertilizer they need to use; they don’t know which is the easiest way to get infected with HIV; they don’t know what their politicians do when in office. When their firmly held beliefs turn out to be incorrect, they end up making the wrong decision, sometimes with drastic consequences — think of the girls who have unprotected sex with older men or the farmers who use twice as much fertilizer as they should. Even when they know that they don’t know, the resulting uncertainty can be damaging. For example, the uncertainty about the benefits of immunization combines with the universal tendency to procrastinate, with the result that a lot of children don’t get immunized. Citizens who vote in the dark are more likely to vote for someone of their ethnic group, at the cost of increasing bigotry and corruption.
We saw many instances in which a simple piece of information makes a big difference. However, not every information campaign is effective. It seems that in order to work, an information campaign must have several features: It must say something that people don’t already know (general exhortations like “No sex before marriage” seem to be less effective); it must do so in an attractive and simple way (a film, a play, a TV show, a well-designed report card); and it must come from a credible source (interestingly, the press seems to be viewed as credible). One of the corollaries of this view is that governments pay a huge cost in terms of lost credibility when they say things that are misleading, confusing, or false.
Second, the poor bear responsibility for too many aspects of their lives. The richer you are, the more the “right” decisions are made for you. The poor have no piped water, and therefore do not benefit from the chlorine that the city government puts into the water supply. If they want clean drinking water, they have to purify it themselves. They cannot afford ready-made fortified breakfast cereals and therefore have to make sure that they and their children get enough nutrients. They have no automatic way to save, such as a retirement plan or a contribution to Social Security, so they have to find a way to make sure that they save. These decisions are difficult for everyone because they require some thinking now or some other small cost today, and the benefits are usually reaped in the distant future. As such, procrastination very easily gets in the way. For the poor, this is compounded by the fact that their lives are already much more demanding than ours: Many of them run small businesses in highly competitive industries; most of the rest work as casual laborers and need to constantly worry about where their next job will come from. This means that their lives could be significantly improved by making it as easy as possible to do the right thing — based on everything else we know — using the power of default options and small nudges: Salt fortified with iron and iodine could be made cheap enough that everyone buys it. Savings accounts, the kind that make it easy to put in money and somewhat costlier to take it out, can be made easily available to everyone, if need be, by subsidizing the cost for the bank that offers them. Chlorine could be made available next to every source where piping water is too expensive. There are many similar examples.
Third, there are good reasons that some markets are missing for the poor, or that the poor face unfavorable prices in them. The poor get a negative interest rate from their savings accounts (if they are lucky enough to have an account) and pay exorbitant rates on their loans (if they can get one) because handling even a small quantity of money entails a fixed cost. The market for health insurance for the poor has not developed, despite the devastating effects of serious health problems in their lives because the limited insurance options that can be sustained in the market (catastrophic health insurance, formulaic weather insurance) are not what the poor want.
In some cases, a technological or an institutional innovation may allow a market to develop where it was missing. This happened in the case of microcredit, which made small loans at more affordable rates available to millions of poor people, although perhaps not the poorest. Electronic money transfer systems (using cell phones and the like) and unique identification for individuals may radically cut the cost of providing savings and remittance services to the poor over the next few years. But we also have to recognize that in some cases, the conditions for a market to emerge on its own are simply not there. In such cases, governments should step in to support the market to provide the necessary conditions, or failing that, consider providing the service themselves.
We should recognize that this may entail giving away goods or services (such as bed nets or visits to a preventive care center) for free or even rewarding people, strange as it might sound, for doing things that are good for them. The mistrust of free distribution of goods and services among various experts has probably gone too far, even from a pure cost-benefit point of view. It often ends up being cheaper, per person served, to distribute a service for free than to try to extract a nominal fee. In some cases, it may involve ensuring that the price of a product sold by the market is attractive enough to allow the market to develop. For example, governments could subsidize insurance premiums, or distribute vouchers that parents can take to any school, private or public, or force banks to offer free “no frills” savings accounts to everyone for a nominal fee. It is important to keep in mind that these subsidized markets need to be carefully regulated to ensure they function well. For example, school vouchers work well when all parents have a way of figuring out the right school for their child; otherwise, they can turn into a way of giving even more of an advantage to savvy parents.
Fourth, poor countries are not doomed to failure because they are poor, or because they have had an unfortunate history. It is true that things often do not work in these countries: Programs intended to help the poor end up in the wrong hands, teachers teach desultorily or not at all, roads weakened by theft of materials collapse under the weight of overburdened trucks, and so forth. But many of these failures have less to do with some grand conspiracy of the elites to maintain their hold on the economy and more to do with some avoidable flaw in the detailed design of policies, and the ubiquitous three Is: ignorance, ideology, and inertia. Nurses are expected to carry out jobs that no ordinary human being would be able to complete, and yet no one feels compelled to change their job description. The fad of the moment (be it dams, barefoot doctors, microcredit, or whatever) is turned into a policy without any attention to the reality within which it is supposed to function. We were once told by a senior government official in India that the village education committees always include the parent of the best student in the school and the parent of the worst student in the school. When we asked how they decided who were the best and worst children, given that there are no tests until fourth grade, she quickly changed subjects. And yet even these absurd rules, once in place, keep going out of sheer inertia.
The good news, if that is the right expression, is that it is possible to improve governance and policy without changing the existing social and political structures. There is tremendous scope for improvement even in “good” institutional environments, and some margin for action even in bad ones. A small revolution can be achieved by making sure that everyone is invited to village meetings; by monitoring government workers and holding them accountable for failures in performing their duties; by monitoring politicians at all levels and sharing this information with voters; and by making clear to users of public services what they should expect—what the exact health center hours are, how much money (or how many bags of rice) they are entitled to.
Finally, expectations about what people are able or unable to do all too often end up turning into self-fulfilling prophecies. Children give up on school when their teachers (and sometimes their parents) signal to them that they are not smart enough to master the curriculum; fruit sellers don’t make the effort to repay their debt because they expect that they will fall back into debt very quickly; nurses stop coming to work because nobody expects them to be there; politicians whom no one expects to perform have no incentive to try improving people’s lives. Changing expectations is not easy, but it is not impossible: After seeing a female pradhan in their village, villagers not only lost their prejudice against women politicians but even started thinking that their daughter might become one, too; teachers who are told that their job is simply to make sure that all the children can read can accomplish that task within the duration of a summer camp. Most important, the role of expectations means that success often feeds on itself. When a situation starts to improve, the improvement itself affects beliefs and behavior. This is one more reason one should not necessarily be afraid of handing things out (including cash) when needed to get a virtuous cycle started.
- Resolutions of mathematical conjectures over time.
- Peter McCluskey on Superintelligence and AI takeoff.
- Humor: Wrong Hands (visual puns).
- Michael Barr’s slides on the software defects which caused unintended acceleration in the 2005 Toyota Camry provide a good illustration of how easy it is for well-resourced projects to nevertheless fail to achieve high assurance software in numerous ways.
- Visualizing Africa’s progress.
Several decent, enjoyable books:
- Pennebaker’s The Secret Life of Pronouns
- Levy’s In the Plex
- Ellenberg’s How Not to Be Wrong
- Heller’s Ayn Rand and the World She Made
- Stone’s The Everything Store
- Stross’ The Launch Pad
- Hoffman’s The Dead Hand
Quammen’s Spillover was not particularly “enjoyable” given it’s subject matter, but it was informative and engaging.
Banerjee & Duflo’s Poor Economics is one of the most persuasive books I’ve read on the subject of poverty reduction.
Murray’s Curmudgeon’s Guide to Getting Ahead was short and entertaining but a mixed bag.
Lochbaum et al’s Fukushima was a helpful overview of exactly what happened at Fukushima, in what order, what the policy response was, and what the health and political fallout was.
- “Approval-seeking”: Christiano’s latest post on indirect normativity for AGI.
- “Accuracy of forecasts in strategic intelligence” (in PNAS).
- Some long-term effective altruism advice by Nick Beckstead that I generally agree with: Working in AI or synbio and Finding a job at an organization focused on existential risk.
- Paul Christiano & Katja Grace made some amendments to MIRI’s AI predictions dataset. They explain their amendments and compute the new statistics here.
- Orseau’s two new AGI papers: “Teleporting universal intelligent agents” and “A formal model for multiple, copiable AIs.”
- Bostrom’s new Superintelligence talk, accompanying his book.
- Lilienfeld et al – “Why ineffective psychotherapies appear to work, a taxonomy.”
- A visualization of Superintelligence.