Thoughts on rationalism and the rationalist community from a skeptical perspective. The author rejects rationality in the sense that he believes it isn't a logically coherent concept, that the larger rationalism community is insufficiently critical of it's beliefs and that ELIEZER YUDKOWSKY IS NOT THE TRUE CALIF.

Benefits Of Hypersonic Missles

Not All Fancy New Killing Machines Are Bad

I don’t understand why all the press about hypersonic missiles has only focused on their risks. Some risks are real but there are also strategic benefits and it irks me that it seems like the media is just reasoning based on their gut level feeling that a new fancy weapon must be bad. I’m sure experts have considered both costs and benefits and I’d love to see them but articles that don’t even stop to refute potential benefits irk me. Much like tech workers who refuse to work on military applications on principal rather than weighing the pros and cons of helping the US acquire that particular tech.

  1. Despite claims that they undermine deterrence it seems to me the exact opposite is true for the major nuclear powers. If anti-ICBM defenses destabilize deterrence it stands to reason tech which can’t be defended against could improve deterrence.
  2. The expense and technical difficulty of hypersonic weapons means that only the major powers will likely be able to build them. So now we get the best of bost worlds in that we can build out full anti-ICBM tech to reduce the danger of attacks from rogue nations like NK (unlikely to be able to afford a massive barrage or a huge number of decoys) without undermining the balance of nuclear deterrence with the other major powers. Indeed, the sheer speed of hypersonic weapons offers the tantalizing possibility of anti-ICBM weapons that could impact during boost phase provided they were stationed nearby (e.g. in SK).
  3. These missiles don’t force decisions about counterattack to be made within the short window before striking since the major powers can still counterstrike with missles housed safely on nuclear subs, hidden in silos in their vast empty fields and scrambled to wait in the air. Using hypersonic weapons as second strike weapons ensures that a relatively small number of nukes scattered on subs, scrambled into the air or placed in a few of our minutemen silos provide an effective MAD style deterrence against great powers.
  4. The fact that hypersonic missiles potentially render aircraft carriers and other capital ships useless isn’t all bad. China inevitably will develop its own aircraft carriers if they remain useful meaning both countries will spend massive amounts of money to remain on par. Far better if we remained on par without the capital ships given that whenever we need to project airpower against distant weaker states (e.g. UN approved bombings etc..) we can use the hypersonic missiles to project power. It potentially harms our strategic position re: russia but it’s not totally obvious to me if it’s a net harm or benefit.

Hypersonic Missiles Are Unstoppable. And They’re Starting a New Global Arms Race.

At War The new weapons – which could travel at more than 15 times the speed of sound with terrifying accuracy – threaten to change the nature of warfare. A Mach 14 Waverider glide vehicle, which takes its name from its ability to generate high lift and ride on its own shock waves.

Journalism Can Be Disrupted By Technology Too (Gasp)

Victims of Monopolies Don't Ask For Anti-Trust Protection

So today brings yet another editorial from the journalism world bemoaning the fact that the internet has rendered traditional journalistic outlets unprofitable. And I’m sympathetic to all the people who planned their lives around this profession and are now struggling. It’s always tough when technological progress renders a bunch of jobs obsolete. And we always see the same calls for governmental protection to protect the existing jobs and businesses. The calls for regulation always have some justification but rarely does it involve this level of absurdity. I mean really? You’re going to blame monopolistic practices by Facebook and Google and ask for an exemption from antitrust laws in the same breadth.

If the problem was really some kind of monopolistic pressure from Google and Facebook I’d expect the demand to be to enforce anti-trust law against these companies? The reason that this isn’t the demand is obvious. Companies in the news business aren’t losing money because they must comply with the whims of monopolistic services. They wouldn’t be doing any better if there were 10 popular social networks and 10 major search engines. They are losing money for the simple reason that there are too many companies producing journalistic content online. The internet reduced the transaction costs to access newspaper articles to nearly zero and as long as dozens or hundreds of papers republish the same content people won’t pay for it.

I mean the complaints in the linked editorial aren’t those of a small business being squeezed by a monopolist. They are those of an industry forced to compete for customers. Neither Google or Facebook threatens these news outlets to give them a cut of their online revenue or use their ad-platforms on pain of not being featured on their sites. Indeed, the complaint here is literally the opposite: Google and Facebook are helping people find whatever news sources they want. The ad revenue Google and Facebook generate is a direct consequence of the economic (one can debate the social value) value they bring in terms of search or social networking.

Now one might worry that there will be a social cost if we cut back on the number of news outlets. That’s another discussion but even if so I’m quite wary of letting the news media suck at the government’s teat. I mean if the news industry sees it’s survival as dependent on anti-trust exemptions that makes it dangerously dependent on the continued good will of the government.

As I’ve said before I don’t actually think there is much to worry about. Eventually, the duplicated effort will be cut out of the news industry and we will see a stronger, better kind of investigative reporting rise from the ashes.

Tech overlords Google and Facebook have used monopoly to rob journalism of its revenue

Over the past decade, the news business has endured a bloodbath, with tens of thousands of journalists losing their jobs amid mass layoffs. The irony is, more people than ever are consuming news. There’s never been a greater need for factual reporting, from the White House down to the local school board.

Decision Theory Anti-realism

There Is No Fact Of The Matter About Correct Decision Theory

With the recent flurry of posts in the rationalist community about which decision theory ( e.g. CDT, EDT, UDT etc..) it’s time to revisit the theme of this blog: rejecting rationality realism. In this case that means pointing out that there isn’t actually a well-defined fact of the matter about which decision theory is better. Of course, nothing stops us from arguing with each other about the best decision theory but those disagreements are more like debates about what’s the best programming language than disagreements about the chemical structure of Benzene.

Any attempt to compare decision theories must first address the question: what does it mean for one decision theory to be better than another? Unlike many pseudo-problems1 there is a seemingly meaningful answer to this question: one decision theory is better than another to the extent that the choices it recommends lead to better outcomes for the agent. Other than some ambiguity about which theory is better if neither dominates the other it seems like this gives a straightforward criteria for superiority: we just look at actual outcomes and see which decision theory offers the best results for an agent. However, this only appears to give a well-defined criteria because in every day life the subtle differences between the various ways to understand a choice and how to conceptualize making a choice don’t matter.

In particular, the kind of scenarios which distinguish between the various decision theories yield different answers depending on whether you want to know who you should be (i.e. total source code) to do best, how you should program an agent if you want them to do best, which decision rule should you adopt for you to do best, and what choice gives you the best outcome. Furthermore, these scenarios call into question how the supposed ‘choices’ made by the decision theory relate to our intuitive notion in a way that makes them relevant to some notion of good decision making or if they are simply demanding the laws of physics/logic give way to offer them a better outcome in a way that has nothing to do with actual decisions.

Intuitions and Motivation

I’m sure some readers are shaking their heads at this point and saying something like

I don’t need to worry about technical issues about how to understand a choice. I can easily walk through Newcomb style problems and the rules straightforwardly tell me who gets what which is enough to satisfy my intuitive notion that theory X is better. Demanding one specify all these details is nitpicking.

To convince you that’s not enough let me provide an extreme dramatization of how purported payouts can be misleading and the question turns on a precise specification of the question. Consider the following Newtonian, rather than Newcombian, problem. You fall off the top of the empire State building what you do as you fall past the fifth floor? What would one say about the virtues of Floating Decision Theory which tells us that in such a situation we should make the choice to float gently to the ground. Now obviously, one would prefer to float rather than fly but posing the problem as a decision between these two choices doesn’t render it a real choice. Obviously, there is something dubious about evaluating your decision theory based on it’s performance on the float/fall question. At least on one conception a decision theory is no worse for failing to indicate the agent do something impossible for them so we can’t merely blindly assume that anytime we are handed a set of ‘choices’ and told what their payoffs are we can simply take those at face value.

Yet, this is precisely the situation we encounter in the original Newcomb problem as the very assumption of predictability which allows the demon2 to favor the 1 boxers ensures the physical impossibility of choosing any number of boxes other than what you did choose. Of course, the same is (up to quantum mechanical randomness) true of any actual `choice’ by a real person but under certain circumstances we find it useful to idealize it as free choice. What’s different about the Newcomb problem is that, understood naively, it simultaneously asks us to idealize selecting 1 or 2 boxes as a free choice while assuming it isn’t actually. Thus, it’s reasonable to worry that our intuitions about choices can’t just be applied uncritically in Newcomb type problems and now I’ll hope to motivate the concern that there might be multiple ways to understand the question being asked.

Let’s now modify this situation, by imagining that we actually live in the Marvel Universe so there are a number of people (floaters) who respond to large falls by, moments before impact, suddenly decelerating and floating gently to the ground. Now suppose we pose the question of whether, as you fall past the 5th floor, you should choose to have been born a floater or not. Obviously, this question suffers from the same infirmities as the above example in that intuitively there is no ‘choice’ involved in being a floater or not but being a floater. However, we can mask this flaw by instead of phrasing the choice as between being a floater and not instead phrasing it as being between yelling, “Holy shit I’m a floater” and concentrating totally on desperately trying to orient yourself so your feet strike first. Now presuming there is a strong (even exceptionless) psychological regularity that only floaters take the first option it follows that EDT recommends making such a yell while CDT doesn’t.

However, taking a look at the situation it seems clear that the two theories are in some sense answering different questions. If I wanted to know whether or not it is preferable to be the kind of person who yells “Holy shit I’m a floater” then I should consult EDT for an answer. Instead, if I’m interested in what I should do in that situation that doesn’t seem particularly relevant. I believe this should move us to consider the possibility we haven’t asked a clear question when we ask what the right decision theory is and in the next section I will consider a variety of ways the problem we’re trying to solve can be precisified and not they give rise to different decision theories.

Possible Precisifications

Ultimately, there is something a bit weird about asking what decision a real physical agent should take in a given situation. After all, the agent will act just as it’s software dictates and/or the laws of physics require. Thus, as Yudkowsky recognizes, any comparison of decision theories is asking some kind of counterfactual. However, which counterfactual we ask makes a huge difference in what decision theory is preferable. For instance, all of the following are potential ways to precifisify the question of what it means for it to be better for XDT to be a better deciscion theory than YDT.

  1. If there was a miracle that overrode the agent’s programming/physical laws at the moment of a choice then doing so in the manner prescribed by XDT yields better outcomes than doing so in a manner prescribed by YDT.
  2. In fact those actual agents who more often choose the outcome favored by XDT do better than those who choose the outcome favored by YDT.
  3. Those actual agents which adopt/apply XDT do better than those who adopt/apply YDT.
  4. Suppose there is a miracle that overrode physical laws at the moment the agent’s programming/internal makeup is specified then if the miracle results in outcomes more consistent with XDT than YDT the agent does better.
  5. As above except with applying XDT/YDT instead of just favoring outcomes which tend to agree with it.
  6. Moving one level up we could ask about which performs better, agents whose programming inclines them to adopt XDT or YDT when considered.
  7. Finally, if what we are interested in is actually coding agents, i.e., writing AI software, we might ask whether programmers who code their agents to reason in a manner that prefers choice A produce agents that do better than programmers who code agents to reason in a manner that prefers choice B.
  8. Pushing that one level up we could ask about whether programmers who are inclined to adopt/apply XDT/YDT as true produce agents which do better.

One could continue and list far more possibilities but these six are enough to illustrate the point.

For instance, note that if we are asking question 1 CDT outperforms EDT. For the purposes of question 1 the right answer to the Newcomb problem is to be a 2 boxer. After all, if we idealize the choice as a miracle that allows deviation from physical law then the demon’s prediction of whether we would be a two-boxer or one-boxer no longer must be accurate so two-boxes always outperforms one boxing. It doesn’t matter that your software says you will choose only one box if we are asking about outcomes where a miracle occurs and overrides that software.

On the other hand it’s clearly true that EDT does better than CDT with respect to question 2. That’s essentially the definition of EDT.

To distinguish the remaining options we need to consider a range of different scenarios such as demons who punish agents who actually apply/adopt XDT/YDT in reaching their conclusions. Or consider Newcombian demons who punish agents who adopt (or whose programmers adopted one of XDT/YDT).

Ultimately, which criteria we should use to compare decision theories depends on what we want to achieve. Different idealizations/criteria will be appropriate depending on whether we are asking which rule we ourselves should adopt, how we should program agents to act, how we should program agents who program agents etc.. etc… Moreover, I’d suggest that once we’ve fully preciscified the kind of question we want to ask the whole debate about which decision theory is best becomes irrelevant. Given a fully specified question we can just sit down and compute (or do empirical analysis) and when we can’t it indicates that we’ve failed to fully specify what we are asking.

The Use of Decision Theory By Agents

As a postscript I’d note that it’s also misguided to assume that the right way to program some kind of AI agent is to have that agent adopt some kind of decision theory like framework. Many discussions of decision theories seem to presume this by phrasing questions in terms of what decision theory should an AI apply/adopt. However, there is no reason to suppose that the way to produce the behavior favored by XDT is for the agent to actually believe/apply XDT. For instance, if a demon punishes agents who have adopted XDT then the outcomes XDT prefers might be best achieved by agents which explicitly eschew XDT. More pragmatically, it’s not at all clear that the most effective way for agents to reach XDT compatible outcomes is to perform the considerations demanded by XDT. That’s a good way to implement some algorithms but not all.

The reason that decision theory is useful in normal situations (i.e. lacking Omega/Newcombian demons) is that it’s a decent heuristic to assume that the way we internally consider outcomes/make choices doesn’t affect the payout we receive. Under this assumption pretty much all ways of preciscifying the question give the same answer and it offers some good advice for programming agents. However, the usefulness of the framework once we abandon this isn’t clear and can’t simply be assumed.
Thus, not only would I argue that the debate over which decision theory is best is misguided, but that we need to be more careful about the assumptions we make about applicability as well.

Thus, not only would I argue that the debate over which decision theory is best is misguided, but that we need to be more careful about the assumptions we make about applicability as well.


  1. For instance, any attempt to answer what makes one programming language better than another reveals substantial disagreement about which tradeoffs are desirable and no agreed upon framework for resolving them. Indeed, we in some sense all recognize that which programming language tradeoffs are desirable is context dependent. 
  2. Or in Yudkowsky’s formulation, Omega. 

Stop Calling Subjects Ethically Fraught

It's An Excuse Not An Argument

Listening to the Last Week Tonight on Gene Editing (it’s pretty good) and seeing this debate about paying organ donors I’m compelled to call out the practice of simply asserting that something is ethically fraught or troublesome.

Both with respect to not compensating organ donors (something which could save huge numbers of lives) and with (mostly prospective) limits on eliminating genetic disease or even barring improvement I think we let people who are simply uncomfortable with change off the hook by constantly repeating the supposed truism that the issue is ethically fraught or there are serious ethical concerns. It’s basically a free pass that excuses the fact that they are putting their discomfort ahead of people’s welfare.

Under all the scenarios/conditions seriously being considered No, there aren’t ethical concerns. Fears like letting a bank reposes your kidney are no more relevant to the proposals on the table than the fear that debtors will enslave people is to wages. Similarly, concerns about racially motivated eugenics programs have no plausible relationship to any kind of gene therapy even being prospectively considered.

Of course, we should hear potential concerns about such policies just like we would for any other policy/technology. However, opponents should be on the spot to either shut up or come up with compelling arguments suggesting harms. Based on the fact that the opponent in the WSJ to paying for organ donation is reduced to arguments like “The introduction of money for a precious good comes at the cost of the ability for one to aspire to virtue” makes me doubt they can come up with such arguments.

I’d add that I think philosophers are partially to blame on this point. As a matter of philosophical interest we correctly find clever new arguments seeking to show that paid organ donation is actually somehow problematically coercive or otherwise wrong more interesting than the obvious argument that it saves lives. However, just as physicists need to convey to the public that the very thing which makes theories which deviate from the standard model interesting also makes them less likely I think philosophers need to do this as well.

How to Provide Better Incentives to Organ Donors

Three experts discuss strategies to address the shortage of organs available for people who need transplants.

Social Control and The Principle Agent Problem

The Chinese Example And The Dangers Of Restricting Free Speech

This interesting post reminded me of my suspicion that a lot of the censorship in China isn’t the result of Xi Jinping’s crazed desire to be repressive. Almost certainly Xi would benefit from far less censorship and may indeed benefit from reports in the media exposing misbehavior by low level party officials but the incentives of those with the power to control expression (both to show off their loyalty and hide embarrassing events) means that far more censorship gets implemented than Xi would ideally want.

I think this is an important lesson for those who want to limit our free speech (or academic freedoms) when it comes to issues of race, gender harassment and the like. Even though the speech that one intends to ban may not have much value and impose great harms one needs to keep in mind the risks posed in delegating the practical authority to determine what speech qualifies.

Politician’s Incentives Regarding Facebook

God I hope not but sounds plausible.

The Peltzman Model of Regulation and the Facebook Hearings – Marginal REVOLUTION

If you want understand the Facebook hearings it’s useful to think not about privacy or technology but about what politicians want. In the Peltzman model of regulation, politicians use regulation to tradeoff profits (wanted by firms) and lower prices (wanted by constituents) to maximize what politicians want, reelection.

Privacy Regulation Is Likely Unworkably Hard

Don't Count On The Government Regulating Facebook

Tyler Cowen provides a great analysis of one of the generic calls for regulating big data (and Facebook in particular). Putting this together with his previous post pointing out that it would cost us each ~$80/year to use facebook on a paid basis1. Taken together they make a compelling case that there is no appetite in the US for serious laws protecting data privacy and that whatever laws we do get will probably do more harm than good.

To expand on Cowen’s point a little bit let’s seriously consider for a moment what a world where the law granted individuals broad rights to control how their information was kept and used. That would be a world where it would suddenly be very hard to conduct a little poll on your blog. Scott Alexander came up with some interesting hypothesizes regarding brain functioning and trans-gender individuals by asking his readers to fill out a survey. But doing that survey meant collecting personal and medical information about his readers (their gender identification, age, other mental health diagnoses) and storing it for analysis. He certainly wouldn’t have bothered to do any such think if he was required to document regulatory compliance, include a mechanism for individuals to request their data be removed or navigate complex consent and disclosure rules (now you’ve gotta store emails and passwords making things worse and risk liability if you become unable to delete info). And what about the concerned parent afraid children in her town are getting sick too frequently. Will it now be so difficult for her to post a survey that we won’t discover the presence of environmental carcinogens?

One is tempted to respond that these cases are obviously different. These aren’t people using big data to track individuals but people choosing to share non-personally identifiable data on a survey. But how can we put that into a law and make it so obvious bloggers don’t feel any need to consult attorneys before running survey?

One might try and hang your hat on the fact that the surveys I described don’t record your email address or name2. However, if you don’t want repeated voting to be totally trivial that means recording an IP address. Enough questions and you’ll end up deanonymizing everyone and there is always a risk (Oops, turns out there is only one 45 year old Broglida). On the other hand google if it’s ok as long as you don’t deliberately request real world identifying information the regulation is toothless — google doesn’t really care what your name is they just want your age, politics, click history etc.. .

Well maybe it should only be about passively collected data. That’s damn hard to define already (why is a click on an ajax link in a form different than a click on a link to a story) and risks making normal http server logs illegal. Besides, it’s a huge benefit to consumers that startups are able to see which design or UI visitors prefer. So checking if users find a new theme or video controls preferable (say by serving it to 50% of them and seeing if they spend more time on the site) shouldn’t require corporate counsel be looped in or we make innovation and improvement hugely expensive. Moreover, users with special needs and other niche interests are likely to particularly suffer if there is no low cost hassle free way of trying out alternate page versions and evaluating user response.

Ultimately, we don’t really want the world that we could get by regulating data ownership. It’s not the world in which facebook doesn’t have scary power. It’s the world where companies like facebook have more scary power because they have the resources to hire legal counsel and lobby for regulatory changes to ensure their practices stay technically legal while the startups and potential competitors don’t have those advantages. Not only do we not want the world we would get by passing data ownership regulations I don’t think most people even have a clear idea why that would be a good thing. People just have a vague feeling of discomfort with companies like facebook not a clear conception of a particular harm to avoid and that’s a disastrous situation for regulation.

Having said this, I do fear the power of companies like facebook (and even governmental entities) to blackmail individuals based on the information they are able to uncover with big data. However, I believe the best response to this is more openness and, ideally, an open standards based social network that doesn’t leave everything in the hands of one company. Ultimately, that will mean less privacy and less protection for our data but that’s why specifying the harm you fear really matters. If the problem is, as I fear, the unique leverage being the sole possessor of this kind of data provides facebook and/or governments then the answer is to make sure they aren’t the sole possessor of anything.

Zeynep Tufekci’s Facebook solution – can it work? – Marginal REVOLUTION

Here is her NYT piece, I’ll go through her four main solutions, breaking up, paragraph by paragraph, what is one unified discussion: What would a genuine legislative remedy look like? First, personalized data collection would be allowed only through opt-in mechanisms that were clear, concise and transparent.


  1. Now, while a subscription funded facebook would surely be much much cheaper I think Cowen is completely correct when he points out that any fee based system would hugely reduce the user base and therefore the value of using facebook. Indeed, almost all of the benefit facebook provides over any random blogging platform is simply that everyone is on it. Personally, I favor an open social graph but this is even less protective of personal information. 
  2. Even that is pretty limiting. For instance, it prevents running any survey that wants to be able to do a follow up or simply email people their individual results 

More Confusion About Gender Equality

It's Never Been About Numerical Equality

So apparently the Swedish government is going to pay women to edit Wikipedia out of concern that wikipedia contribution is heavily biased in favor of men. This misunderstands what’s desirable about gender equality in a serious way. While this may be nothing more than harmless idiocy it provides an important warning about the importance of taking a hard look at programs designed to increase gender equity.

There is no intrinsic good to having the same number of women editing Wikipedia (or engaged in any particular career or activity) as men. Rather, there is a harm when people are denied the ability to pursue their passion or interest on account of bias or stereotypes about their gender.

Now, if one believes that some activity discriminates against interested women one might think that artificially inducing women to participate (affirmative action, or even payment) is an effective long term strategy to change attitudes, e.g., working with women will change the attitudes of men in the field and place women in positions of power so future women won’t face the same discrimination. However, wikipedia actively encourages using unidentifiable user names, doesn’t require gender identification and there is no evidence of a toxic bro-culture among frequent editors. Thus, there is no reason to think injecting more female editors into wikipedia will reduce the amount of discrimination face by women in the future. Indeed, even if you believe that women are underrepresented on wikipedia because of discrimination or stereotyping, e.g., women aren’t techie or women aren’t experts, then paying women to edit wikipedia is wasting money that could have been used to combat this actual harm.

Moreover, there is no particular evidence that the edits made by frequent editors to wikipedia are particularly likely to be somehow slanted against women or otherwise convey a bias that this kind of program would be expected to rectify. Indeed, paying members of particular groups to edit wikipedia is an assault on wikipedia’s reliability. While I’m not particularly concerned about Swedish women the underlying principle that no one should be able to pay to ensure wikipedia is more reflective of the views of a certain identity group is important. I mean what happens to information about the Armenian genocide if Turkey decides that it should pay Turks to increase their representation on Wikipedia?

But why care about this at all? I mean so what if the Swedes blow some money stupidly? It’s not like men are suffering and need to be protected from the injustice of it all.

The reason we should care is that it’s shows in a clear and uncontrovertible fashion how easily well intentioned concern about gender equity can go off the rails. Given the potential blowback and murkiness of the issues there is a tendency to just take for granted the fact that programs which claim to be about improving gender equity are at least plausibly targeted at that end. However, this proves that even in the most public circumstances its dangerously easy for people to conflate ensuring numerical equality with increasing gender equality. Given that in many circumstances the risk isn’t merely wasting money but, as in affirmative action and quota programs, actively making things worse (e.g. by making people suspect female colleagues didn’t really earn their positions) we need to be far more careful that such programs are doing some worth those costs.

Not Enough Women at Wikipedia? | EconLog | Library of Economics and Liberty

by Pierre Lemieux …women need state encouragement to do some of the one million edits that are made on Wikipedia every day. Presumably, this will promote the liberation of women. The Swedish government, or at least its foreign minister, wants…

NRA Conferences Reduce Gun Injuries?

Misleading Reporting and Dubious Statistics

So the following letter is being widely reported online as if it is evidence for the importance of gun control. I’m skeptical of the results as I detail in the next post but even if one takes the results at face value the letter is pretty misleading and the media reporting is nigh fraudulent.

In particular if one digs into the appendix to the letter one finds the following statement: “many of the firearm injuries observed in the commercially insured patient population may reflect non-crime-related firearm injuries.” This is unsurprising as using health insurance data means you are only looking at patients rich enough to be insured and willing to report their injury as firearms related: so basically excluding anyone injured in the commission of a crime or who isn’t legally allowed to use a gun. As a result they also analyzed differences in crime rates and found no effect.

So even on it’s face this study would merely show that people who choose to use firearms are sometimes injured in that use. That might be a good reason to stay away from firearms yourself but not additional reason for regulation as is being suggested in the media.

Moreover, if the effect is really just about safety at gun ranges then its unclear if the effect is from lower use of such ranges or that the NRA conference encourages greater care and best practices.

Reasons To Suspect The Underlying Study

Also, I’m pretty skeptical of the underlying claim in the study. The size of the effect claimed is huge relative to the number of people who attend an NRA conference. I mean about 40% of US households are gun owners but only ~80,000 people attend nationwide NRA conventions or ~.025% of the US population or ~.0625 of US gun owners. Thus, for this statistic to be true because NRA members are busy at the conference we would have to believe NRA conference attendees were a whopping 320 times more likely to be inflict a gun related injury than the average gun owner.

Now if we restrict our attention to homicides this is almost surely not the case. Attending an NRA convention requires a certain level of financial wealth and political engagement which suggests membership in a socioeconomic class less likely to commit gun violence and than the average gun owner. And indeed, the study finds no effect in terms of gun related crime. Even if we look to non-homicides gun deaths from suicides far outweigh those from accidents and I doubt those who go to an NRA convention are really that much more suicidal inclined.

An alternative likely explanation is that the NRA schedules its conferences for certain times of the year when people are likely to be able to attend and we are merely seeing seasonal correlations masquerading as effects from the NRA conference (a factor they don’t control for). Also as they run all subgroup analysises and don’t report the results for census tracks and other possible subgroups the possibility for p-hacking is quite real. Looking at the graph they provide I’m not exactly overwhelmed.Not exactly convincing graph

The claim gets harder to believe when one considers the fact that people who attend NRA meetings almost surely don’t give up going to firing ranges during the meeting. Indeed, I would expect (though haven’t been able to verify) that there would be any number of shooting range expeditions during the conference and that this would actually mean many attendees would be more likely to handle a gun during that time period.

Though, once one realizes that the data set one is considering is only those who make insurance claims relating to gun related injuries it is slightly more plausible but only at the cost of undermining the significance of the claim. Deaths and suicides are much less likely to produce insurance claims and the policy implications aren’t very clear if all we are seeing is a reduction in people injured because of incorrect gun grips (see the mythbusters about this..such injuries can be quite serious).

Artificial Intelligence And The Structure Of Thought

Why Your Self-Driving Car Won't Cause Armageddon

In recent years a number of prominent individuals have raised concerns about our ability to control powerful AIs. The idea is that once we create truly human level generally intelligent software or AGI computers will undergo an intelligence explosion and will be able to escape any constraints we place on them. This concern has perhaps been most throughly developed by Eliezer Yudkowsky.

Unlike the AI in bad science fiction the concern isn’t that the AI will be evil or desire dominion the way humans are but simply that it will be too good at whatever task we set it to perform. For instance, suppose Waymo builds an AI to run its fleet of self-driving cars. The AI’s task is to converse with passengers/app users and route its vehicles appropriately. Unlike more limited self-driving car software this AI is programmed to learn the subtleties of human behavior so it can position a pool of cars in front of the stadium right before the game ends and helpfully show tourists the sites. On Yudkowsky’s vision the engineers achieve this by coding in a reward function that the software works to maximize (or equivalently a penalty function it works to minimize). For instance, in this case the AI might be punished based on negative reviews/frustrated customers, deaths/damage from accidents involving its vehicles, travel delays and customers who choose to use a competitor rather than Waymo. I’m already skeptical that (super) human AI would have anything identifiable as a global reward/utility function but on Yudkowsky’s picture AGI is something like a universal optimizer which is set loose to do its best to achieve rewards.

The concern is that the AI would eventually realize that it could minimize its punishment by arranging for everyone to die in a global pandemic since then there would be no bad reviews, lost customers or travel delays. Given the AI’s vast intelligence and massive data set it would then hack into microbiology labs and manipulate the workers there to create a civilization ending plague. Moreover, no matter what kind of firewalls or limitations we try and place on the AI as long as it can somehow interact with the external world it will find a way around these barriers. Since its devilishly difficult to specify any utility function without such undesirable solutions Yudkowsky concludes that AGI poses a serious threat to the human species.

Rewards And Reflection

The essential mechanism at play in all of Yudkowsky’s apocalyptic scenarios is that the AI examines its own reward function, realizes that some radically different strategy would offer even greater rewards and proceeds to surreptitiously work to realize this alternate strategy. Now its only natural that a sufficiently advanced AI would have some degree of reflective access to its own design and internal deliberation. After all it’s common for humans to reflect on our own goals and behaviors to help shape our future decisions, e.g., we might observe that if we continue to get bad grades we won’t get into the college we want and as a result decide that we need to stop playing World of Warcraft.

At first blush it might seem obvious that realizing its rewards are given by a certain function would induce an AI to maximize that function. One might even be tempted to claim this is somehow part of the definition of what it means for an agent to have a utility function but that’s trading off on an ambiguity between two notions of reward.

The sense of reward which gives rise to the worries about unintended satisfaction is that of positive reinforcement. It’s the digital equivalent of giving someone cocaine. Of course, if you administer cocaine to someone every time they write a blog post they will tend to write more blog posts. However, merely learning that cocaine causes a rewarding distribution of dopamine in the brain doesn’t cause people to go out and buy cocaine. Indeed, that knowledge could just as well have the exact opposite effect. Similarly, there is no reason to assume that merely because an AGI has a representation of their reward function they will try and reason out alternative ways to satisfy it. Indeed, indulging in anthropomorphizing for a moment, there is no reason to assume that an AGI will have any particular desire regarding rewards received by its future time states much adopt a particular discount rate.

Of course, in the long run, if a software program was rewarded for analyzing its own reward function and finding unusual ways to activate it then it could learn to do so just as people who are rewarded with pleasurable drug experiences can learn to look for ways to short-circuit their reward system. However, if that behavior is punished, e.g., humans intervene and punish the software when it starts recommending public transit, then the system will learn to avoid short-circuiting its reward pathways just like people can learn to avoid addictive drugs. This isn’t to say that there is no danger here, left alone an AGI, just like a teen with access to cocaine, could easily learn harmful reward seeking behavior. However, since the system doesn’t start in a state in which it applies its vast intelligence to figure out ways to hack its reward function the risk is far less severe.

Now, Yudkowsky might respond by saying he didn’t really mean the system’s reward function but its utility function. However, since we don’t tend to program machine learning algorithms by specifying the function they will ultimately maximize (or reflect on and try to maximize) its unclear why we need to explicitly specify a utility function that doesn’t lead to unintended consequences. After all, Yudkowsky is the one trying to argue that its likely that AGI will have these consequences so merely restating the problem in a space that has no intrinsic relationship to how one would expect AGI to be constructed doesn’t do anything to advance his argument. For instance, I could point out that phrased in terms of the locations of fundamental particles its really hard to specify a program that excludes apocalyptic arrangements of matter but that wouldn’t do anything to convince you that AIs risked causes such apocalypses since such specifications have nothing to do with how we expect an AI to be programed.

The Human Comparison

Ultimately, we have one example of a kind of general intelligence: the human brain. Thus, when evaluating claims about the dangers of AGI one of the first things we should do is see if the same story applies to our brain and if not if there is any special reason to expect our brains to be different.

Looking at the way humans behave its striking how poorly Yudkowsky’s stories describe our behavior even though evolution has shaped us in ways that make us far more dangerous than we should expect AGIs to be (we have self-preservation instincts, approximately coherent desires and beliefs, and are responsive to most aspects of the world rather than caring only about driving times or chess games). Time and time again we see that we follow heuristics and apply familiar mental strategies even when its clear that a different strategy would offer us greater activation of reward centers, greater reproductive opportunities or any other plausible thing we are trying to optimize.

The fact that we don’t consciously try and optimize our reproductive success and instead apply a forest of frameworks and heuristics that we follow even when they undermine our reproductive success strongly suggests that an AGI will most likely function in a similar heuristic layered fashion. In other words, we shouldn’t expect intelligence to come as a result of some pure mathematical optimization but more as a layered cake of heuristic processes. Thus, when an AI responsible for routing cars reflects on its performance it won’t see the pure mathematical question of how can I minimize such and such function any more than we see the pure mathematical question of how can I cause dopamine to be released in this part of my brain or how can I have more offspring. Rather, just as we break up the world into tasks like ‘make friends’ or ‘get respect from peers’ the AI will reflect on the world represented in terms of pieces like ‘route car from A to B’ or ‘minimize congestion in area D’ that bias it towards a certain kind of solution and away from plots like avoid congestion by creating a killer plague.

This isn’t to say there aren’t concerns. Indeed, as I’ve remarked elsewhere I’m much more concerned about schizophrenic AIs than I am about misaligned AI’s but that’s enough for this post.