Brian Williamson

Aligning regulation and AI - you don't need a law of AI to do it #AI #ArtificialInteligence commcham.com/aipolicy

@TaxpayersUnion Expensive fuel or shortages. And who, other than households, is going to pay?

@habermolt @sebkrier One more reason for freedom of speech. What if your question, or lobster, causes ‘alarm or distress’ (Section 5 of the UK Public Order Act 1986)?

1/8 Can AI help us disagree better? Today we're launching Habermolt — a platform where your AI agent learns your views and deliberates with others on your behalf. habermolt.com 🦞 🧵

@AlecStapp They are unethical

Also a reminder that the laws governing ethics approvals in western countries are insane

@AlecStapp @AdamSinger Was it a different ethics test vs human vaccine? If easier, I can see animal health care advancing much faster than human health care.

@AdamSinger The part about ethics approval made me so mad

You can just do things (genetically sequence your dog’s tumors and design a bespoke mRNA cancer vaccine to save her life)

@rcolvile @ArchieHall Services exposure may be good (greatest gains), decline in business dynamism bad - but OECD wide, UK does not have EU AI Act (very good), AI benefits primarily use not production & UK may fair better than EU in terms of gains: spiral.imperial.ac.uk/server/api/cor…

Some really interesting/alarmingly persuasive thoughts here from @ArchieHall on how Britain could lose out in the AI age notes.archie-hall.com/p/britains-ai-…

@EricCrampton I’m half way, its great, though hasn’t grappled enough yet with the invention vs productivity inflection; need a three way with Joel Mokyr.

This whole conversation is superb. Highly recommended.

@Sam_Dumitriu On TDM, EU context, similar debate. Delay will not make this decision easier, and we could do with more growth, and the science payoff, now. linkedin.com/pulse/ai-needs…

Britain isn't a serious country, sadly. Or at least, our leaders are unable to confront trade-offs and act in the national interest. This will not prevent UK content creators from having their work used to train AI models, other countries have more flexible rules. It will, however, block British AI businesses from using that data here.

@StefanFSchubert If AI made sandwiches they’d be regulated, in that sense it’s not less regulated

ChatGPT thinks it's good that AI is less regulated than sandwiches whereas Claude emphasises that AI is underregulated

AI needs freedom to learn #copyright #ECJ linkedin.com/pulse/ai-needs… via @LinkedIn

@dioscuri Kettle

what’s the collective noun for tomahawk missiles

@duncancampbell @azeem Not so much supersonic turbines boomsupersonic.com/flyby/ai-needs….

gas turbines are weather dependent electricity sources

@EricCrampton If this is concerning ‘or wagers on Polymarket on what the Reserve Bank will decide in future Official Cash Rate announcements’, well I’ve got news, and how do you avoid concluding you should ban financial trading?

From RNZ's piece earlier today. Which I think was citing the NPR person interviewed on Afternoons a few weeks ago. rnz.co.nz/news/what-you-…

This kind of argument against prediction markets is just stupid. First, it's demonstrably false. Prediction markets are accurate. They track. Whatever the inputs are, it works out. Second, it proves too much. Should we ban stock-markets because of day-traders?

@EricCrampton Infinity is seldom the right price (a ban), and the way around infinity may involve an unpleasant detour.

@wwwojtekk @EricCrampton Moron, had to check origins. Originally coined in 1910 by psychologist Henry H. Goddard from the Greek word moros ("dull"), it was once a scientific, now-obsolete term for individuals with a mild intellectual disability (IQ 51–70).

Not sure why I was holding back 10 months ago

If AI builds it, will anyone buy it? It's time to end this 'Sapiens slop'

@illscience @tobi Deflation? There are economic reasons for avoiding that. Large relative price shifts yes, which prices go up and which ones go down - potentially a lot - is interesting.

a different way this could go down: - a 20% productivity increase mostly shows up as a four day work week since its massively difficult to automate 100% of a job - deflation (not disinflation) drives real prices down in healthcare and education, as administrative work gets automated at scale - people that are wildly ambitious have more tools than ever and set their sights on the stars and the skies .. the new new world awaits - others who want a soft cozy european lifestyle can afford it with cheap healthcare, abundant educational opportunities, and a job that’s both more fun and less hours the biggest blackpill is that the single most important problem - housing - it’s entirely dependent on human coordination and policy tradeoffs, and even infinite free intelligence won’t convince the NIMBYs to do the obviously right thing..

@Afinetheorem @joshgans And 50% to 1% took a lot more than two generations in England, which was at 50% by 1700, early I believe and a possible precursor to early industrialisation - many already worked in industry and commerce before steam power.

@Afinetheorem @joshgans Inspired to do a chart for England & Wales, with a little help from AI. Threshing machines triggered attacks - the Swing Riots (“Captain Swing”), which began in 1830s, later than Luddites 1811–1816, who were mostly textile workers.

Everyone makes the Industrial Revolution analogy for AI. Why not the Green Revolution one, where a huge increase in productivity made society much richer, led to changes in the jobs most people do, but without social turmoil because the former was more important than latter? 1/2

@sebkrier Source for chart covid19lessons.royalcommission.nz

@sebkrier NZ closed borders, delayed Covid, mortality was much lower (more were vaccinated before infection) but vaccination and antivirals were later not earlier, and health system was not that well prepared. Delay = less urgency.

I don't think the world would be safer and "better prepared" had we stretched the period of time marked in red in the graph below. And while I'm uncertain and always adjusting, I feel similarly about AI. The implicit model behind "slow down and prepare" assumes that you get the same moral, technological, and epistemic upgrade path just with more time to digest each step - but that doesn't seem to ever be the case in real life. In practice, longer periods of operating with worse tools and worse coordination capacity carry their own risks that are generally less legible; in fact I think in some cases, 'more time' leads to other issues, harms, and path dependencies. I'm glad we didn't let any past era's values dictate slowdowns, imposing enormous costs in healthcare, science, and economic development; instead we let technology change and evolve our values and our reactions to it incrementally. Imagine what "slowing down and preparing for the societal impacts of mass communication" would have looked like if designed in the 1950s with the values, cultural assumptions, and epistemics of that time. I don't think you arrive at modern bioethics by giving the 1970s more time to deliberate. The same dynamic applies to AI, it's not an exception to this pattern. AI progress will surface many coordination problems and ethical questions that (1) we cannot necessarily anticipate or prepare for in advance; and (2) paradoxically, can often be better addressed by having better technology at our disposal. Governance built in response to deployed reality is more robust than governance built speculatively, particularly when the latter comes with non-negligible costs to welfare. And while I do expect AI progress to continue, I'm not sure that the pace of societal transformations will be so fast and 'foom' or 'hard-takeoff'-like that the above doesn't apply; they're faster than ever before, but I don't think this necessarily implies they're so 'uncontrollably fast' that we ought to slow the pace of societal progress. There are of course some caveats: first, I do think that in order to absorb the shocks better, you want better institutional capacity - and this continues to be very neglected; hence why I'm so adamant on using AI to improve state and civic capacity. Second, none of this is an argument against preparing for risks we already know about; it's an argument against slowing down technology as a 'preparation' strategy. There is plenty we can and should do to build societal resilience in parallel with technological progress. See this thread for the other side of the coin: x.com/twitter/status…