The antitrust cycle of destructive destruction, the First AI-Agent Wars and how we should remedy the Google monopoly problem

The judgement against Google is a very good thing ... but remedy design needs to think about the next war if it is to avoid the paradoxical antitrust cycle of destructive destruction. Here is how.

Tony Curzon Price, 9/08/24

"

Yet across the gulf of space, minds that are to our minds as ours are to those of the beasts that perish, intellects vast and cool and unsympathetic, regarded this earth with envious eyes, and slowly and surely drew their plans against us. — H. G. Wells (1898), The War of the Worlds

If you watch big tech regulation and aren't a shill for GAMA, you'll have spent the last 36hrs suffused by the pleasure and thrill of Judge Amind Mehta's Memorandum Opinion finding Google guilty of violation of Section 2 of the Sherman Act in its various attempts to buy protection for its search business monopoly. Three loud cheers for the Department of Justice and US antitrust.

But as many commentators have pointed out, now comes the hard part: how to remedy the abuse. As Adam Reiter pointed out on the FT's Unhedged yesterday, investors are not yet running for the exits - Alphabet shares hardly moved around the time of the announcement. And, as pointed out by Christian Bergqvist in a very nice summary post on LinkedIn, the particular feature pinpointed in Judge Mehta's ruling, the control of search through the browser and the phone operating system, may well be yesterday's problem as the primary mode of access to the web shifts to being AI-agent based rather than browser based:

Google took two preventive steps [to the threat of competition in search]. Firstly, it paid (>26b in 2021), and still pays for default status in other browsers (i.e., Safari and Firefox) across platforms (PC, Tablets, Mac, or iPhone). Secondly, it (aggressively) rolled out the Android OS to smartphones, Smartwatches, cars, etc., for “free”, only conditioned upon the pre-installation of, i.e., Google Search Engines. Google even added a cherry by throwing in cash. Why? Because Google, like Bill Gates, feared consumer migration. Initially, to smartphones, now to the Internet of Things, and soon AI-Personal Assistance's. With each migration, an opportunity to dethrone the incumbent emerges, translating into what I refer to as Browser War II, III, and IV. [...]

I don't expect the ruling to change much as it pertains to Browser War III [i.e., Smartphone search & associated data access]. Google won by surviving the migration to smartphones. Next is Browser War IV (and V), centered on AI and ChatGPT. Against whom? No, other than Microsoft, having lost Browser War II to Google.

Bergqvist's argument points to a depressing prospect for antitrust enthusiasts (and I think of myself as being one): we take 15 years to rule against a practice, just as the major new innovation takes hold and renders a remedy based in the law otiose. Perhaps, in a nod to Schumpeter's optimism about technological change, and Marx's pessimism about institutional change keeping up with it, we could call this the paradox of destructive destruction: the old monopoly gets swept aside by innovation just in time to destroy the impact of anti-monopoly remedies…

So, in light of the cycle of destructive destruction, is there any chance that we can get ahead of it? find some remedies that have a chance of solving the upcoming wars? Yes, but sadly it is unlikely to come out of the US antitrust system - getting ahead of the bad cycle requires a wider regulatory solution that focuses on data access and control. Let me explain.

How does online advertising generate the hundreds of billions of margin for the GAMA gatekeepers? Remember, every household in the UK pays a subscription of about £1000 per year per household for access to the so-called open web. How so? That is the amount marketers spend on online advertising, and that businesses recoup by adding the cost to the price of goods sold. Our free-at-the-point-of-use web is financed by a stealth tax that none of us can avoid. More than half that tax goes to GAMA, and GAMA's (true) margins are very high. Very roughly, a lower bound on the amount of pure profit going to the GAMA gatekeepers - and most of that to Google - will be of the order of £300-£400 per year per UK household. (I provide a longer version of this argument over here).

Google collects that eye-watering rent because its big properties - search foremost amongst them, but YouTube and Gmail also being large contributors - do two things very valuable to marketers: a) it is quite good (and better than anyone else) at predicting what paid for advertisement each web viewer is likeliest to click on and b) it can often sell access to that web viewer, though sometimes it has to share that honour with others - publishers like newspapers, or Apple, as emphasised by Judge Mehta, which gets paid about $15 per iPhone user by Google to make Google the default search in Safari, or any number of others who can pass eyeballs on to Google for them to monetise them.

As we move to the “Browser wars IV and V”, the wart of the AI assistants, Google, Microsoft, Amazon and Apple will all try to develop a new version of the same game: a) understand each of us such that they can predict better than others our next influenceable purchase, and b) deliver to sellers the ability to influence our AI agents to push them towards their offering. This new game will be so far from the original “Browser War” that I think we need a new terms for it - perhaps the first AI Agent War is more like it. This war is what LLM competition at the moment are really about, and why GAMA are investing so much in the technology: this really is a phase of creative destruction, at least from the perspective of investment in LLMs. (And by the way, I am not at all panglossian about that creative destruction: LLMs are getting innovation dollars because they are the path to this pot of gold, whilst I am almost certain that actual progress for human society is much more likely to come from more innovation in AI robotics - yes, even the amazing table-tennis playing robot has more potential for good in it than the kind of LLM innovation we are getting at the moment.)

Imagine a near future in which GAMA each launch a version of their AI agents, made sufficiently compelling for each of us to adopt one. I can imagine my AI agent, for the sake of argument provided by Google, offering to keep track of all my insurance policies (together with all the official paperwork that goes with them, like car registrations, MOT certificates, claims, licence endorsements etc), and, at renewal time, the agent collects offers from a range of insurance providers, The AI agent summarises each policy, tells me why I might prefer one to the other, and gets me to click on my choice. The agent then fills out the rest of the paperwork, and I know I am insured according to my wishes. Do I want this insurance agent in my life? You bet I do, and I am sure many others will be like me in this regard.

You can see the Google playbook here: given me a wallet that integrates all my official documents with a tracker of the activities that require them. Google can coax us to offering up these documents in the name of convenience and trusted data handling. Apple will clearly be in this game also. Already, each of them is pushing their “wallets” as being the nice homely analogy for these documents. Once I've stuffed my Google wallet, and given Google can scan my gmail for communications from my insurance, it can build up a pretty accurate picture of my insurance needs. It is now basically home and dry - the Google Agent can wander around the insurance market pitching me as a very well qualified lead, and can offer access to me to a handful of the highest bidders for the privilege.

If Google - or another GAMA - succeeds in this pivot, GAMA as a whole will have ensured that the destruction really is, from the point of view of society as a whole, destructive, and from the point of view of their margins, business as usual. How so?

Imagine that Google, Apple, Meta and Amazon have all created these wallets and AI agents and that they are all reasonably good at their job. Can we expect competition to deliver the rents to consumers and citizens? Will competition be the guarantor against the Doctorowian forces of (delicately labelled) enshittification this time around? Well … if GAMA has their way, putting official documents, contracts and all the data needed by agents into wallets other than theirs will be hard, and getting them out to move them to another wallet will be even harder. Try today to move passwords from Apple KeyChain to a rival provider and you'll get a sense of the future they will try to build. And citizens and consumers will quite rightly be very worried about the security of the data needed by agents, so there will be a very natural reluctance to go with unknown entrants as wallet-keepers.

All of this means that even if each GAMA has a successful line in AI Agents, each citizen/consumer will be utterly locked into one of the walled gardens. Therefore, each GAMA will have control of each Citizen/Consumer, and competition will not in fact help. This is rather like the absence of competition today between Edge and Chrome, or Chrome and Firefox: each of us usually uses just one Browser, and search advertising on Bing is essentially the same price as on Chrome - once you have a user, you are a monopolist over their attention, and you will price like a monopolist. That applies whether you have 1% or 80% share of eyeballs.

I expect that GAMA boardrooms are quietly confident that they'll win the First AI Agent War. They may worry whether they will gain market share over other GAMAs, but they are confident, I expect, that they will continue to be the primary resellers of valuable attention - even if it is now the delegated attention of an AI-Agent they are selling on. That's all pretty depressing. So how do we use a remedy in the Third Browser War to have a chance of defeating GAMA in the upcoming First AI Agent War? There are two key elements to a genuine remedy: a) mandating interoperable data standards for the kind of data that will end up in wallets (ie the ability to move wallet providers easily) and b) a legal, regulatory and governance framework to set the conditions for delegation to AI agents. (I have rehearsed the argument for these as remedies over here).

Wallets need to be built in a way very similar to email clients or the web: any email client can read emails, any web browser can read W3C compliant hypertext; on the server side of each, an email server knows how to route and send emails, and any web server knows how to respond to a web client. Wallets are the clients to what needs to be a new and augmented hypertext standard, one that has sophisticated access control, authentication and verification capabilities, as well as being well-suited to being processed by AI agents. That is the first part of the remedy, the interoperability bit.

The second, the delegation framework, has to do with the increasing attention & cognitive overload that cyberspace imposes on us in the absence of trustworthy delegation of choices. This overload leaves each of us open to exploitation through rational ignorance, and the collective effect of all that exploitation is a Doctorowian enshittified cyberspace. The only way out is to delegate some of our data choices to trustworthy institutions that are value-aligned with us - call them data unions.

These two crucial pillars of a good cyberspace would also be a remedy to the market dominance identified by Judge Mehta. With control over truly portable and interoperable personal data, a data union today would be able to offer to marketers a better source of buying intention data than Google can offer, and so would be able to collect the attention rent on behalf of the attention owner - the Citizen/Consumer. This solution, however, lies beyond the ken of antitrust law. It requires mandated standards, the redefinition of data rights, and a new sort of Power of Attorney framework fit for the AI-Agent age. A far-sighted whole-of-government approach is needed. But whoever does it well will turn the current cycle of destructive destruction to a truly creative one, in which AI-agents will be deployed for the good of Citizens and Consumers, not GAMA managers and shareholders.