Thoughts in Between

by Matt Clifford

TiB 183: AI's dead end; chip shortages; against remote work; and more...

Welcome new readers! Thoughts in Between is a newsletter (mainly) about how technology is changing politics, culture and society and how we might do it better.

It goes out every Tuesday to thousands of entrepreneurs, investors, policy makers, politicians and others. It’s free.

Forwarded this email? Subscribe here. Enjoy this email? Forward it to a friend.

Is AI marching down a dead end?

We've talked a lot about the power of what some are calling "foundation models" (see TiB 179) in artificial intelligence - that is, vast neural network models like GPT-3 (see previous coverage) that have generated increasingly impressive results as the data and computational power used to train them scales up dramatically. We will doubtless return to them frequently, so it's worth noting that they're controversial in some AI circles.

It’s not just the concern that as "stochastic parrots" (Timnit Gebru's wonderful phrase) they sometimes spout bigoted or false statements (but see also this paper), but a more fundamental critique that they represent a dead end for AI. The best-known academic proponent of that view is perhaps Gary Marcus, who has a new essay with Ernest Davis arguing that AI needs a radical research agenda beyond foundation models. Marcus and Davis say that foundation models lack important pillars of "true" intelligence and, unless and until we find ways to integrate these into our models, our efforts to build more general intelligences will falter.

Examples of the gaps include the ability to form and reuse knowledge about the world; "common sense" knowledge of categories like space and time (see also this paper); and the capacity to reason about and represent values. Perhaps this is true - I am certainly not qualified to evaluate. But I do note that with every scaling up of foundation models, they have tended to display at least some capabilities that seemed unlikely at the previous step. I highly recommend Gwern's definitive write up. Foundation models are doubtless limited, but they are strange and wonderful things.

Who benefits from remote work?

The new(ish) online publication Works in Progress continues to impress; every issue so far has had multiple fascinating pieces. If you like Thoughts in Between you will almost certainly like Works in Progress. I could have written today's entire newsletter riffing on essays from their new edition, which touch on many favourite TiB topics, from José Luis Ricon's piece on applying the scientific method to science itself to Neil Hacker's discussion of "buyers of first resort" - that is, the importance of procurement for advancing technology.

Caleb Watney has an excellent essay in the last edition on innovation clusters (like Silicon Valley) and remote work - a topic we've looked at multiple times, most recently last week. Watney makes two points that merit particular attention. First, that there is a geopolitics of technology clusters, of which the US has been the primary beneficiary for the last half century. It's a huge strategic advantage to physically house the preeminent clusters of silicon, software or AI talent assets. If remote work erodes this (Watney is more skeptical than I am), the consequences won't be purely economic.

Second, Watney argues that remote work might change who benefits from innovation. Historically, startups have gained a great deal from the talent and knowledge spillovers from physically local established firms. Remote innovation might be good for intra-firm innovation, but have fewer spillovers. This echoes our experience at EF during COVID: if entrepreneurs had even a couple of weeks to build relationships in-person, they could sustain them online. But in programmes that we had to run fully remote, the rate and robustness of team formation was significantly lower. I'm a remote work optimist, but it's worth considering that it may be a pro-incumbent force.

Chips chips everywhere... except where they're needed

A big TiB theme over the last 18 months has been the stubborn persistence of the physical world (see TiB 109, 161, 162, 164, 181). Nowhere is this clearer than when you look at supply chains, which are currently jammed up across a wide range of products. If you read just one Twitter thread this week, make it this one, which is a superb whistle-stop tour through the challenges afflicting the supply and movement of everything from giraffe toys(!) to fertiliser. It's a bit of a rabbit hole if you follow the links, but it's fascinating.

Matthew Hockenberry, the thread's author, notes that the problem has multiple causes: labour shortages in global shipping; COVID surges in key bottleneck countries and firms; rapid changes (many COVID-induced) in consumer behaviour; failures of prediction; the prioritisation of efficiency over resilience (see TiB 112); and more. Moreover, it's only likely to get more challenging to keep things moving, as we see the second order effects of climate change and geopolitical tension.

Long-time readers won't be surprised that the industry with perhaps the most acute problem is semiconductors. I recommend this Odd Lots podcast episode on the "Global Chip Crisis" in which analyst Stacy Rasgon runs through the litany of challenges that are making semiconductors scarce. The most striking thing perhaps, at least for those of us who live increasingly digitised lives, is just how constrained by physics the sector is. It takes months to spin up dormant capacity, years to create it from scratch. You may soon live in the Metaverse, but it will run on chips that most certainly do not.

Quick links

  1. "College and the culture war". What does attending college do to your political beliefs?
  2. Three strikes? Fascinating and disturbing thread on drone strikes from the Pentagon's former head of High Value Targeting
  3. The sun is shining, out comes the term sheet. Interesting study on the non-rational drivers of VC behaviour
  4. The real estate theory of value? Provocative study on what's happened to labour's share of value since 1989 and who benefited
  5. Good news! We probably won't go extinct in the way the dinosaurs did

BONUS: If you're in London and are interested in the intersection of tech and policy, this looks a good event/group

How to help

Thanks for reading Thoughts in Between. Social media shares and forwarding to friend helps grow the readership, for which I'm always grateful.

And I’m always happy to get comments, questions and recommendations from readers - just hit reply.

Until next week,

Matt Clifford

PS: Lots of newsletters get stuck in Gmail’s Promotions tab. If you find it in there, please help train the algorithm by dragging it to Primary. It makes a big difference.