Monthly Archives: September 2024

The welcome application of good sense to AI hype

Posted on by
Reply

Summer over in a flash, autumn wind and rain outside – perhaps cosy evenings will speed up both my reading and review-posting.

I just finished AI Snake Oil by Arvind Narayanan and Sayash Kapoor, having long been a fan of the blog of the same name. The book is a really useful guide through the current hype. It distinguishes 3 kinds of AI: generative, predictive and content moderation AI – an interesting categorisation.

On the generative AI so much in the air since ChatGPT was launched in late 2022, and the persuasive debunking here is of the idea that we are anywhere close to ‘general’ machine intelligence, and of the notion that such models pose existential risks. The authors are far more concerned with the risks associated with the use of predictive AI in decision-making. These chapters provide an overview of the dangers: from data bias to model fragility or overfitting to the broad observation that social phenomena are fundamentally more complicated than any model can predict. As Professor Kevin Fong said in his evidence to the Covid inquiry last week, “There is more to know than you can count.” An important message in these times of excessive belief in the power of data.

The section on the challenges of content moderation were particularly interesting to me, as I’ve not thought much about it. The book argues that content moderation AI is no silver bullet to tackle the harms related to social media – in the authors’ view it is impossible to remove human judgement about context and appropriateness. They would like social media companies to spend far more on humans and on setting up redress mechanisms for when the automated moderation makes the wrong call: people currently have no recourse. They also point out that social media is set up with an internal AI conflict: content moderation algorithms are moderating the content the platform’s recommendation algorithms are recommending. The latter have the upper hand because it doesn’t involve delicate judgements about content, only tracking the behaviour of platform users to amplify popular posts.

There have been a lot of new books about AI this year, and I’ve read many good ones. AI Snake Oil joins the stack: it’s well-informed, clear and persuasive – the cool breeze of knowledge and good sense are a good antidote to anybody inclined to believe the hyped claims and fears.

Screenshot 2024-09-29 at 17.31.57

Techno-financial imperium

Posted on by

Underground Empire: How America Weaponized the World Economy by Henry Farrell and Abraham Newman is a rip-roaring read and simultaneously terrifying. The reader is left with the clear impression of fragility in the world financial system – bad enough in itself – that could tip over into armed conflict. Eek.

The book is about the intersection of the discovery by the US of physical choke-points in the global internet with legal choke points in the world economy and financial system, globalized as it has become. Specifically, the NSA started to find it convenient, post 9/11, to start gathering data on a massive scale on financial movements through the internet. Just seven sites in the US act as bottlenecks for internet traffic, while in other countries there are a similarly small number of such sites. This realisation became the Underground Empire of the title, the deployment of a set of technical and financial mechanisms using these choke points, allowing the US “to cut off businesses and even whole countries from access to key global networks that wove the world’s economy together.”

As the book goes on to describe, the initial post-9/11 responses merged into a haphazard but ever more extensive set of policies and legal tools that enabled the US to extert increasing leverage over extra-territorial entities – not only businesses but also governments. These have included using the power and reach of the US dollar as a reserve currency and currency of trade to “persuade” the Swift network into applying US requirements to exclude for example North Korean and Iranian entities from international payments, and more recently sanctions on US chip technology or IP being exported. The authors write: “From outside, the underground empire seems like a relentless machine of domination, the product of decades of engineering. From inside, it looks quite different, a haphazard construction lashed together from ad hoc bureaucratic decisions and repurposed legal authorities.”

And yet it is a powerful empire and has contributed to increasing geopolitical tensions, and in particular US-China rivalry – although it has also, thanks to its increasingly evident power, prompted countervailing efforts to escape imperial power. But the book concludes “There is no visible exit from the underground empire. … The roots of imperium go far too deep ever to be fully torn out.” The globalisation dream of peace through markets has turned into a nightmare of potential conflict thanks to the realisation elsewhere that markets are built on the foundations of American physical, financial and intangible infrastructure. The book concludes that it is up to the US now to turn the empire into more of a commonwealth, serving the interests of others as well as itself – something hard to see happening in the current political moment.

Screenshot 2024-09-16 at 16.45.55

Where did summer go?

Posted on by

Already September 1st, and a tang of autumn in the morning air. I had a two-week holiday with no laptop or emails, and lots of reading. Then a two-week scramble to catch up with the accumulated email and work. So here is a summary of the books I got through.

A couple were directly about AI: I, Warbot: The dawn of artificially intelligent combat by Kenneth Payne and The Age of AI by Henry Kissinger, Eric Schmidt and Daniel Huttenlocher. This is not at all my territory but it seemed a good idea to learn something about AI and conflict as we have a relevant project at the Bennett Institute. Both left me pretty terrified, particularly the former: “The warbot doesn’t want anything and it isn’t troubled by our moral dilemmas,” writes Payne. The delegation of decisions from the localized (is that person spotted by facial recognition a valid target?) to the global (is it time to launch a major attack) to AI seems inevitable; the book ends calling for serious reflection. Indeed. As for the other, I certainly know less about geopolitics than do the authors, but some of the claims on which the geopolitical analysis is constructed seem over the top. The book suggests AI is changning the human relationship with reason, challenging the possibility of a free society and free will, and that we will be letting AI decide for us when we have ambiguous contexts or goals. “The boundary between AI and humans is strikingly porous,” they write. The book concludes AI is significantly changing human consciousness and moving us “close to knowing reality beyond the confines of our own perception.” Hm. It’s doing all sorts of things but I’m not sure about this kind of claim. I like better my colleague Neil Lawrence’s reflections in The Atomic Human on humanity in the light of AI.

Screenshot 2024-09-01 at 15.02.03My favourite read about tech was Scott Shapiro’s Fancy Bear Goes Phishing: The dark history of the information age in five extraordinary hacks. The book pulls off the impressive feat of being a cracking good read with compelling stories and a super-clear technical explanation of why the internet is so easy to hack. (Incidentally, the name Fancy Bear for the notorious Russian state hackers was devised by the co-funder of CrowdStrike, the company that became famous for triggering Microsoft’s global Blue Screen of death meltdown while I was away. And ironically, the book explains that Microsoft had already improved its safeguards so that developers accessing the kernel would be far less likely to trigger a Blue Screen event.)  One thing I particularly liked about the book was its designation of ‘downcode’ and ‘upcode’, the former being the computer code of all the various types, the latter the societal laws, regulations and norms within which the internet operates. “Change the upcode and you will change the kind of downcode produced.” Highly recommended.
Screenshot 2024-09-01 at 15.04.40I also enjoyed Shane O’Mara’s Talking Heads: How Conversation Shapes Us, which the publisher kindly sent to me. A book about neuroscience, it argues that thinking requires memories, and memories are collectively constructed through conversation. Our individual memory systems contribute to a cultural memory repository. “A society without memory has no capacity to repair itself, it has no capacity to evolve, it has no capacity to adapt, it has no capacity to remember or learn.” There is a two-way process of individual thoughts shaping the collective and in turn being shaped by them. Conversations or narratives or other cognitive artefacts enable co-ordination and collective action. (This reminded me of my friend Celia Heyes’ concept of Cognitive Gadgets.) Thus nationalism is a cognitive concept and needs a psychological understanding. The book doesn’t discuss social media directly but clearly the way we have conversations about the world has changed as a result, and not in a good way.

On with the gloomy theme. I was looking forward to Abby Innes’s book Late Soviet Britain: Why materialist utopias fail not least because the title is so brilliant. It has three parts. The first argues that central planning and the neoclassical competitive economy are equivalent – as indeed formally they are, one where quantities adjust and one prices. Francis Spufford’s superb book Red Plenty dramatizes this equivalence. (This isn’t cited here – I think the author would enjoy it.) I was mildly annoyed by the book’s referring to the ‘materialism’ of both – the former Soviet bloc certainly measured material product but the western metric of domestic product is 80% non-material. But maybe there’s a technical meaning of materialism I don’t know.

The second part of the book describes the growing dysfunctions of neoliberal regimes in the west, whereby the concepts of neoclassical economics and market primacy were translated into political practice. For instance, New Public Management, the book argues, has failed just as central planning failed in practice, by incentivising individuals to deliver to target rather than the genuine aim (and just as machine learning systems fail due to their ‘alignment problem’, automated New Public Management if we’re not careful). The third, short, section is a brief history of Britain’s political evolution – why we did Brexit and why the Conservatives are in a mess.

The whole didn’t quite come together for me because I couldn’t follow the step from the formal theoretical equivalence of the two ‘ideal’ economic models to the argument that “actually existing neoliberalism” was as doomed to failure as central planning. Neither system in practice was very similar to the theoretical ideal. For instance, New Public Management was arguably a quantity-planning mechanism rather than a market mechanism. But the book’s central construct is a stimulating idea and I enjoyed reading it.

Screenshot 2024-09-01 at 15.07.08Finally in the non-fiction, Anne Applebaum’s short and compelling

, pointing out the emergence of autocracy as a global, internet-enabled business model. Time to pull the plug on the enabling industry of lawyers, accountants and bankers. What will it take to get other nation-state governments to stop supporting hostile autocrats by permitting their business interests to carry on unimpeded?

Screenshot 2024-09-01 at 15.09.05Along with all these I read a ton of fiction, from detective novels (Elly Griffiths, Donna Leon, C.J. Sansom) to Sarah Perry’s terrific Enlightenment, Atsuhiro Yoshida’s sweet Goodnight Tokyo (a taxi driver version of my favourite TV programme Midnight Diner*) all the way to the seriously serious Solenoid by Mircea Cartarescu (a gift from a Romanian former student).

* A good illustration of why Netflix is a poor substitute for domestically-produced programmes, if you compare the 1st series with the Netflixed later ones.