bfinn's Shortform

I watched all 7 hours of Dominic Cummings' testimony to parliament yesterday on the UK government response to COVID. (Cummings was the Prime Minister’s top adviser.)

Key points he made that are relevant to rationalists (some hardly mentioned by the media):

• Groupthink throughout government and its SAGE committee of scientific advisers meant their initial plan - no lockdown, await herd immunity - wasn't abandoned soon enough. Psychological 'memes' - that Britons wouldn't accept lockdowns or track & trace - were believed with little basis. The groupthink was broken in part by Cummings seeking an outside view of technical documents from Demis Hassabis and Tim Gowers (interestingly).
• Institutional design failure means incompetent people get promoted to leadership and decision-making roles in UK government and political parties. Various highly competent individuals in more junior ranks were sidelined and left. People aren't incentivized to do the right things. Cummings said he himself shouldn't have been in such a high-powered job, which should have been held by someone far more intelligent & capable. (Cummings was in fact considered the smartest person in Downing St, though he lacks a technical background.)
• Weak planning for catastrophes, e.g. poor access to data, inability to circumvent slow bureaucratic procedures, lack of detailed advance plans and lines of responsibility for them. He mentioned anthrax attacks and solar flares as other potential scenarios.
• Human challenge trials should have occurred early on - perhaps even in Jan 2020.

Overall I found him a credible witness, because his testimony was very detailed (e.g. he recalled numerous dates of meetings & events), and quite self-critical, blaming himself for not forcing a lockdown sooner, and for not resigning at various points. His analysis above also seems sound.

A good, plausible movie about AI risk might be an indirect way to raise the chance of government action via public awareness (if that would be useful). Have there been such AI risk movies that I've missed? If not, how could people get one to happen?

Cf I get the impression governments take asteroid risks a bit more seriously thanks to various asteroid-strike movies over the years. Not least the recent Don't Look Up on Netflix, which despite being a comedy was probably quite thought-provoking to many viewers (e.g. showing how there could well be government & public denial, resulting in disaster).

When reading Superintelligence by Nick Bostrom it struck me parts of it would make excellent movie plots. Eg the AI's ingenious ways of escaping an AI box by social engineering. (Though, googling, I see that a dire-looking romcom called Superintelligence, perhaps inspired by but very far removed from the book, was released in 2020.)

'Up to' is a misleading rhetorical device widely used in marketing, politics, etc. It precedes a statistic that is an impressive or scary maximum, where some kind of average would almost always be the relevant figure instead. So I'm immediately suspicious of any claim containing the phrase 'up to'.

Eg broadband providers that quote maximum speeds (that only a few optimally-located customers might get in the early hours of the morning) instead of averages. IIRC the UK's advertising regulator has banned this particular trick.

Indeed maxima are often nearly meaningless, such as 'up to x% of COVID cases result in death'. Why not a single figure? In practice it may mean something like: among various studies, the highest estimate produced was x%. (So why not give the range, or just the average?) Or it means something almost irrelevant, like: among various unstated categories of people, the highest figure is x% - likely to be a small, unrepresentative subgroup (e.g. aged over 90, or those with BMI > 30).

I can't be the only person who's noticed this rhetorical device everywhere (I haven't checked), though I've never read/heard anyone else mention it.

There seems a big contradiction in the position of environmentalists who support animal rights (as almost all do). They say we shouldn’t eat meat because animals, and animal feed, contribute to CO2 emissions. But if we didn't eat meat, the animals we breed for that purpose wouldn’t exist.

Surely animal rights include the right to life. So denying billions of cows & chickens any life at all seems a strongly anti-animal position.

While meat production involves killing animals, at least they get some life. And if (a big 'if') they were well treated, so they'd enjoy the life they do get, that's a good deal for the animals. Because they'd be much better off than in the wild - a grim existence of starvation, disease, fear, and being torn to bits by predators. And they'd also be better off than not existing.

(The environmentalist assumption that life in the wild is good - because 'nature is good' - doesn't follow. Life in the natural state - e.g. the Stone Age - is nasty, brutish, and short. Human history is the struggle to escape the natural state.)

So, CO2 aside, my position is that meat eating is fine as long as the animals are well treated. Which would mean more expensive meat, and less of it, but that's OK. Happy cows are better than no cows.

(At this point some people object that 'the demand for meat is so high that it can't be met while treating the animals well' - but that ignores that demands don't have to be met; if supply is restricted by a welfare requirement, prices will rise accordingly, so only those who can afford to pay for happy cows will eat steak. It's like saying 'billions of people would want a private jet, so unless we ban private jets they'll all get one'. Well, they would if they cost $10, but not if they cost $10 million.)