AI is a Collective Disaster

It takes two incomes to buy a home, my kids can’t play on the street any more, and I never get invited to dinner parties. Each of these is an example of a collective action failure. Put simply, individuals optimising for their best outcome leads to a collective failure: life becomes worse for everybody because we each do what is best for us individually.

To take the classic example, it’s safer (and cheaper) for all of us when cars aren’t two-tonne monstrosities. But you are individually safer if you’re in one. So we all start buying them, and then streets and roads become more dangerous for everybody. I made myself safer, at everybody’s expense.

Widespread LLM use is going to be the next big example.

Individual Benefit

Writing, researching, coding, and otherwise working with AI is unambiguously a time-saver. I can create documents, presentations and code in a fraction of the time it used to take me. Sharpening my ideas and discovering new ones is quicker and easier than it ever has been before. In the short term at least, using LLMs is a no-brainer. Letting your employees use LLMs is similarly beneficial.

Surely then, as LLM use increases, we should all be awash with free time and life should be improving for everybody? In the 1930s, Keynes predicted a 15-hour work week. As the AI revolution looms large on the horizon, what does our new future hold?

Collective Disaster

Collective failures usually result from either finite resources being allocated between increasingly able competitors, or the disappearance of a ‘good’ that was previously not thought of as consumable.

Competition

It takes two incomes to buy a home because we have a finite amount of land and as people earn more, we simply increase the cost of buying it. We are forced to spend time curating our children’s extracurriculars because other parents are doing the same and a ‘normal’ childhood of the 1970s puts our children at a disadvantage today.

The most obvious finite resource that LLM-assisted humans are competing for is attention. Whether that be through advertising and marketing, video games and apps, or even through poky little blog posts…we’re all increasingly able competitors in the global fight for attention. Being able to use AI effectively will briefly give you a competitive advantage, but soon it’ll be the price of entry, which is why the time we save won’t become leisure.

Public Goods

Our children lost the use of our streets for many reasons, but one clear one was that we didn’t realise that open playable streets were a common good that could be lost. Dark night skies were free for everybody, until, lightbulb after lightbulb, they were extinguished. More recently, phone calls were the way of speaking to people, until spam calls made answering the phone a chore that many seek to avoid. Once calls cost nothing to make, answering the phone to strangers became too costly to be worthwhile.

The public common that LLMs are eroding, I believe, is trust in the digital world. The relationship between amount of work, time spent, and importance has been irrevocably broken.

Previously, if a colleague sent me a document and asked for review, I’d be confident that they had poured substantial time and effort into that document and that it was important to them. That gave me confidence to spend my time looking at it, too. Now, a lengthy document gets a cursory glance at best, or a thoughtless call to an LLM to review.

The UK planning system allows locals to object to developments in their area. It used to be that a reasoned and lengthy letter detailing harms would have to be the product of time spent. Now, I can build an app that reads every planning application, probes for weak points, and auto-objects in 10,000 beautifully crafted words, all in an afternoon. Council officials can’t possibly keep up with the AI slop heading their way, and so stop paying attention.

My thoughtful colleagues who write for themselves, and the local homeowner whose life will be permanently blighted by a development lose out, all because I act rationally. For examples of a system getting weaker, while individuals optimise for themselves, look no further than the current state of academic publishing.

Conclusion

I wish I had an answer for how to avert this collective disaster. There’s an arms race in spam detection for phone calls, but detection always lags creation. Rather than a technical solution, we decided answering the phone to strangers was too costly. Maybe our digital future involves only reading from trusted sources. The internet of the early twenty-first century will be looked back upon fondly, as a time when you could meaningfully engage with strangers, confident in little else except their humanity. What a shame.