As we have this on-going discussion about AI, I want to return to the question of scale and externalities and how they are the defining feature of the Silicon Valley economy. We can’t make any sense of today’s discussion of LLMs and AI without placing it in that context.
“Externalities” are all the byproducts and secondary effects of, in this case, a business enterprise. If you make paint or industrial solvents you create various kinds of pollution in the process. How is that dealt with? Specifically, are you the paint or solvent maker responsible for dealing with those consequences? More specifically do you have to pay for dealing with those consequences or are they absorbed by the society at large while the profits of the enterprise go exclusively to you? “Absorbing” those consequences can mean accepting levels of illness, diminished quality of life or simply having society at large pick up the tab for various kinds of remediation. The externalities are really just costs, but often semi-hidden ones. A huge amount of public policy deals with properly or improperly bringing costs and externalities into alignment.
Starting in the very first years of the 21st century the core Silicon Valley business model was to fund a lot of start up companies, see most of them fail, but have a small handful experience “hockey stick growth”. Imagine a hockey stick pointing to the left. The blade is initial profitability and growth followed by the shaft which is growth and profitability going almost straight up. You get that kind of growth by moving quickly to massive scale, usually through network effects. (You’re on Facebook because everyone else is on Facebook. Everyone uses Stripe as a payment processor because everyone else does and that’s allowed Stripe to get their prices very low and their technology extremely efficient and performant.) The key to the scale game, the key to hockey stick growth is that you find a business that scales super fast but in which cost growth remains arithmetic but revenue grows exponentially. Most of Silicon Valley over the last three decades has been based on finding that one thing.
What gets less attention is the way this relates to the question of externalities. The Silicon Valley model is one in which you quickly realize that hockey stick growth and operating profitably at scale means that managing externalities is simply not economic. It conflicts with the business model – and in ways that go beyond the simple reality that almost every business wants to socialize costs and privatize gains. The following example isn’t about externalities directly but it can help illustrate the issue. When most products and services you buy don’t work or you have a question, there’s someone you can call. Yes, you have to wait on hold. They keep trying to pipe you through to an automated system. When you get someone they’re from a foreign country and it can be hard to understand their accent. But you can usually get a person to talk to. Have you ever wondered why there is no call line for Facebook or Twitter or any other social platform? Everything is automated – not just help with logging in but the same with adjudicating a copyright dispute if someone files a frivolous copyright dispute about something you posted. It’s also why it’s pretty simple for a swarm of people or even a swarm of bots to get your accounted suspended, at least temporarily, by filing a slew of rapid fire complaints about you. If the automated system can’t deal with the problem well … it sucks to be you.
Why is this? People will often say it’s because you’re the product, not the customer. And that’s true. But it misses the real point. If Facebook had call center facilities to deal with 3 billion active accounts and 2 billion people using the product every day it would go under. It’s not economic. One of the things that led to the decline of Facebook/Meta’s reputation in the mid-teens, coming off the controversies of the 2016 election was the mass expulsion of Rohingya Muslims from Myanmar in 2017. If you don’t remember the details, the idea was that Facebook had both amplified the campaign of hate, disinformation and incitement that led to the expulsions as well as served as a tool for it. I don’t want to get too far deep into that question. And my point here isn’t really to beat up on Facebook/Meta. I’m making a structural point. Facebook is far too big, far to embedded in the global propagation of ideas, imagery and messaging for it ever to really be able to grapple with all the things it causes or is involved with. It’s totally beyond the capacity of the business model.
This isn’t to excuse anyone. My point here is to illustrate how not accounting for externalities is foundational to the Silicon Valley business model. You probably heard about the recent settlement where Meta agreed to pay $18 billion in damages related to child safety and social media addiction as well as adding fairly dramatic new restrictions on use of their platforms by minors. $18 billion is still a pretty small sum relative to Meta’s scale. But the point is simple: if Meta is actually liable for the potential harms of their products to minors it’s just bad business to let minors keep using them. The universe of potential liabilities is simply too great.
This is a pretty lengthy explanation of a point that hasn’t yet touched in any meaningful way on AI. But it has everything to do with AI and all the companies now working at the forefront of the technology. Because they’re all parts of the same economic culture and the same vision of scale, hockey stick growth and the basic principle that operating at scale and accounting for externalities, liability and all the rest is just not economic. So this post is basically a point of reference that I will refer back to in future posts. Because this framework, understanding it, perceiving its path-dependent outcomes is kind of central to understanding everything about the AI economy.
