Isn’t a key difference that in Terminator the AI seems incredibly incompetent at wiping us out? Surely we’d be destroyed in no time — to start with it could just manufacture a poison like dioxin and coat the world (or something much smarter). Going around with tanks and guns as depicted in the film is entirely unnecessary.
Robert_Wiblin
Karma: 3,376
[Question] I’m interviewing Max Tegmark about AI safety and more. What shouId I ask him?
If it’s just a form where the main reason for rejection is chosen from a list then that’s probably fine/good.
I’ve seen people try to do written feedback before and find it a nightmare so I guess people’s mileage varies a fair bit.
“However, banking on this as handling the concerns that were raised doesn’t account for all the things that come with unqualified rejection and people deciding to do other things, leave EA, incur critical stakeholder instability etc. as a result. ”
I mean I think people are radically underestimating the opportunity cost of doing feedback properly at the moment. If I’m right then getting feedback might reduce people’s chances of getting funded by say, 30%, or 50%, because the throughput for grants will be much reduced.
I would probably rather have a 20% chance of getting funding for my project without feedback than a 10% chance with feedback, though people’s preferences may vary.
(Alternatively all the time spent explaining and writing and corresponding will mean worse projects get funded as there’s not much time left to actually think through which projects are most impactful.)
It would be very surprising if there weren’t an opportunity cost to providing feedback. Those might include:
Senior management time to oversee the project, bottlenecking other plans
PR firefighting and morale counselling when 1 in ~100 people get angry at what you say and cause you grief (this will absolutely happen)
Any hires capable of thinking up and communicating helpful feedback (this is difficult!) could otherwise use that time to read and make decisions on more grant proposals in more areas — or just improve the decision-making among the same pool of applicants.
That there’s an opportunity cost doesn’t show it’s not worth it but my guess is right now it would be huge mistake for Future Fund to provide substantial feedback except in rare cases.
That could change in future if their other streams of successful applicants dry up and improving the projects of people who were previously rejected becomes the best way to find new things they want to fund.
I find these arguments intellectually interesting to a degree.
But like you, my aesthetic preference is just that people who personally feel like having kids should have kids, and those who personally don’t feel like having kids shouldn’t.
If we followed that dollar-store rule of thumb I expect things would go roughly as well as they can, all things considered.
[Question] What should I ask Lewis Dartnell (author of ‘The Knowledge’ and ‘Origins’)?
My guess is this would reduce grant output a lot relative to how much I think anyone would learn (maybe it would grantmaking in half?) so personally I’d rather see them just push ahead and make a lot of grants then review or write about just a handful of them from time to time.
[Question] Next week I’m interviewing Will MacAskill — what should I ask?
[Question] I’m interviewing political scientist Chris Blattman about his new book ‘Why We Fight’. What should I ask him?
[Question] I’m interviewing Nova Das Sarma about AI safety and information security. What shouId I ask her?
Here you go: https://www.stitcher.com/show/80k-after-hours
(Seems like Stitcher is having technical problems, I’ve contacted their technical support about it.)
For the 10⁄10 criteria do you mean a $50k hiring bonus, or a $50k annual salary?
“creating closed social circles”
Just on this my impression is that more senior people in the EA community actively recommend not closing your social circle because, among other reasons, it’s more robust to have a range of social supports from separate groups of people, and it’s better epistemically not to exclusively hang out with people who already share your views on things.
Inasmuch as people’s social circles shrink I don’t think it’s due to guidance from leaders (as in a typical cult, I would think) but rather because people naturally find it more fun to socialise with people who share their beliefs and values, even if they think that’s not in their long-term best interest.
Cool yeah. I just want to provide another more boring reason a lot of us have piled on to bioethics that doesn’t even require ingroup-outgroup dynamics.
Basically all of the people you’re citing (like me) have an amateur interest in bioethics as it affects legal policy or medical practice or pandemic control (the thing we actually follow closely).
You and I agree that harmful decisions are regularly being made by IRBs (and politicians), often on the basis of supposed ‘bioethics’. We also both agree there are at least a handful of poor thinkers in the field who do offer up low quality moral philosophy to support these bad decisions. It’s only natural then for me and my fellow travelers to see these bad decisions, and these writings classified as bioethics justifying them, and suppose that the latter are an important cause of the former.
And these decisions come week after week for years, progressively infuriating me more and more.
I could see I’m making a mistake to judge bioethics as a field by sampling a representative bunch of papers (weighted by citations maybe), reading them, and deciding how reasonable they typically seem. Unfortunately that’s an involved process that few people with an amateur interest are going to have time for. Each person can only go down a few rabbit holes like that each year in between our normal work, personal commitments, staying healthy, and so on.
So I appreciate you and other people doing that heavy lifting and then sharing the results — it’s the only way it’s practical for our mistake to be corrected!
Fair enough, I’m happy to talk less about bioethicists and talk more about institutional review of research ethics.
For what it’s worth I and other critics do regularly/constantly refer people to the classic dissection of the problem caused by IRBs (The Censor’s Hand).
We also talk about the misaligned incentives faced by bureaucrats about as ad nauseam as we talk about bioethics.
And when I’ve seen IRBs in action they have worked to keep their decisions and the reasons for them secret and intimidate researchers into not speaking out, while philosophers publish their ideas in journals you can read (and their arguments can then be used as cover for IRB decisions). So as a practical matter it has been easier for folks to scrutinise bad thinking from philosophers than IRBs even if the average quality of the latter is much worse.
Exciting to see a post about this episode 5 hours after we put it out (!).
A few quick thoughts:
“Berkowitz never mentions that the median voter in most Republican primaries is currently “pro-Trump” so he leaves out the single sentence explanation.”
No but I say that. IIRC one of his responses also takes this background explanation as a given.
“Japan and New Zealand have shown that sovereign parliamentary democracies do not manifest even nascent electoral movements.”
In general I’m with you on thinking some systems of government are less conducive to populist movements, but I’m not sure one can show that by choosing two cases without checking for counterexamples.
“Berkowitz argues that Trump drives turnout. But 2016, Trump’s first term, had moderate turnout. 2020 was Trump’s second election, so why is he driving turnout in turn two.”
I mean there’s multiple factors but I’m with Berkowitz here.
In 2020 polls more people had very strong views about Trump (strong approve or strong disapprove) than is typical of a president, while I don’t think that was true in 2016. So I don’t think there’s anything strange about the idea that he raised turnout in 2020 but not 2016.
The other big factor I would think is easier postal voting.
“No it is not. People do not take on high personal costs for collective gains without a facilitating organization.”
But that just raises the question, why weren’t there better facilitating organizations? Folks expected them to be present but it seems they didn’t organize effectively.
Great to see someone giving this a crack! Let me know how it works out. :)
“The 2.16% U.S. federal funds rate in 2019 is one of the most conservative interest rates possible.”
The U.S. Federal Funds rate has been effectively 0% since April 2020 and was roughly 0% for six years from 2009 to 2015. The same is roughly true of the UK. Central banks in both countries are saying they’ll keep rates low for years to come.
I can’t immediately find a reputable business savings accounts in the UK/US that currently offers more than 1%.
Those that offer the highest rates (something approaching 1%) on comparison sites tend to have conditions (e.g. you lock the money up for a period, or have to keep depositing regularly), and usually have a maximum amount on which you can earn interest, a maximum which is low enough to be binding for these organisations.
These accounts usually offer a high rate to attract customers for a while, then dramatically reduce the interest rate and trust you won’t be bothered moving your money. I think that’s their basic business model.
Opening bank accounts for non-profits, at least in the UK, is a pain — something that will take a few weeks, and some time/attention from the operations team, management and trustees (who are needed for e.g. security checks). It looks like you usually won’t be able to put in more than a million dollars/pounds in any given account, often less.
So you’d need to open many accounts, keep track of them, secure the chequebooks, have them audited annually, integrate them into your bookkeeping system, change the signatures when staff turn over, figure out the idiosyncratic requirements to pull out money when you need to, and so on.
This may sound simple but if you’ve worked in operations you’ll know it’s actually a big hassle.
In return, for each account opened you make <£10k a year, and probably need to keep closing accounts and moving your money into new ones every few years, as the teaser rate used to draw you in is removed.
This may all be worth it, but it’s far from a no-brainer, as these organisation have other fruitful projects they could be using staff to pursue.
I interpreted them not as saying that Terminator underplays the issue but rather that it misrepresents what a real AI would be able to do (in a way that probably makes the problem seem far easier to solve). But that may be me suffering from the curse of knowledge.