Bryan Caplan is an economist known for being a free-thinker, having contrarian – though compelling – ideas, and being uninhibitedly direct in their communication style. If you have not heard of them before, I recommend reading up.
Today, Caplan released a blogpost critiquing effective altruism. I identify as an effective altruist and most of the critiques apply directly to me as well, so it is worthwhile that I reflect on them. Standard caveat that I do not represent effective altruists, and it would be interesting to know where opinions differ.
Bryan starts by complementing EAs on their work on global health and development:
This, in turn, has led to large cash infusions for three charities with demonstrably high benefit/cost ratios: deworming, anti-malaria, and Give Directly. While the typical college EA meeting is just a social club, these social clubs have ultimately made philanthropy better, especially by inspiring tech donors.
As an economist, it is no surprise that Bryan approves of the analysis used to identify these charities as being effective.
Bryan then moves on to the criticisms.
EAs are sadly short on common sense. As a result, many have basically lost interest in philanthropy that demonstrably saves many human lives at low cost, and started obsessing about AI risk.
- My initial response was to challenge the claim that common sense is robust, e.g. I think many of Bryan’s views go against common sense, at least how I perceive common sense.
However, it is better to consider the steelman of the claim, which is that EAs have a bad balance of using fresh thinking versus defering to ‘common sense’ or existing norms. I think this is strong critique, but not for the same reasons as Bryan. E.g. I think many effective altruist organizations are each learning the same basic lessons of how to do the day-to-day running of an organization, instead of adopting existing good practice.
many have basically lost interest in philanthropy that demonstrably saves many human lives at low cost, and started obsessing about AI risk
- The word ‘basically’ is doing a lot of work here. I think many EAs do still have interest in such philanthropy. E.g. I have taken the 10% pledge and continue to donate, even though my main worry nowadays is AI risk. My current split is 50% to animal welfare charities and 50% to global health and development.
I’ve already explained at length why I think [AI risk] is silly.
- I need to watch this to respond. I suspect this will be our core crux. I will pre-emptively say that the core question is whether catastrophic scenarios are feasible. I would be surprised if Bryan makes a compelling case that the odds of a global catastrophe from AI is proportionately low, say 0.001%, but am open to being wrong here. In fact, I want to be wrong here!
But let me add that AI doomsters’ own behavior shows that deep down, they know they’re being silly. Only two doomsters have been willing to bet me; and yes, you absolutely can bet on the end of the world. The optimist prepays the pessimist, end of story.
- I am actually tempted to make a bet with Bryan. But terms might need to be adjusted? I think that literal human extinction is unlikely, because a powerful misaligned AI system does not need to wipe out literally every human in order to do what it ‘wants’. The same way humans have not caused the extinction of animals we dislike. I do not have any sense at all how likely a global catastrophe is, e.g. at least on the scale of covid-19, but am worried it is feasible. I am not sure what a suitable definition would be for a global catastrophe, to operationalize the bet. Also, there is a decent chance that the world wakes up in time and prevents such a global catastrophe.
if you really thought the world was likely to end, you should be going into massive debt, secure in the knowledge that you won’t have to repay it.
- The counter to this is simple: I am highly uncertain on whether the end of the world is likely or not. So in the case where things do turn out well, I do not want to have made myself bankrupt. Also, personally, I am highly risk averse, and likely irrationally so.
AI is such a promising technology that EA-inspired efforts to curtail it are likely to outweigh all of the good that EAs achieved back when they sensibly focused on deworming, anti-malaria, and Give Directly. That’s very sad, showing that without common sense, even high-IQ good intentions still pave the path to bad outcomes.
- Ironically I agree with the conclusion that EAs best intentions could curtail all the good EAs have done, but for the exact opposite reason. EAs have played a significant role in making AI progress as fast as it has – e.g. being early funders and employees at OpenAI – increasing the risk of global catastrophes. Alternatively, if things do go well, then EAs can take some credit, in which case Bryan should be thankful. ;)
But EAs have another set of views which, if taken seriously, wouldn’t merely strangle progress, but make humanity massively worse off: their extreme moral commitment to animal welfare. Ten years ago, I wrote a reductio ad absurdum of the pro-animal welfare position. If all animals morally matter, what about bugs? At the time, the main response was, “Bugs don’t feel pain.” A decade later, however, plenty of EAs are just embracing the reductio.
- I need to read Bryan’s backlog to understand their position here and give an adequate response. I will just note that I am partial to the idea that bugs can ethically matter, and effort should be put to reduce uncertainty on this question.
The point of an argument is to reason from highly obvious premises to not-so-obvious conclusions. But if your conclusion is extremely non-obvious, then its denial is correspondingly extremely obvious. “If A, therefore B” only works if “Not-B, therefore not-A” isn’t more plausible. How do you find maximally obvious premises? Not via argument, which is circular, but via common sense.
- This is interesting. Suppose that all three of these feel obviously true: A, A→B, and, Not-B. What should you do? I have not thought enough about philosophy or reasoning under such uncertainty. In this particular instance of animal welfare, I personally think that A will be more obvious to me than Not-B. I imagine Bryan’s response is I lack common sense. But I can just say that Bryan lacks common sense, and we have made no progress.
Writing this all out, I think the core crux is that I believe Bryan has too high confidence in the belief that AI does not pose catastrophic risks or that insects have no ethical weight.
