We are on the wrong track with the way AI is going, and here’s why: Both the known risks and the unknown risks are not worth it given the plausible outcomes.
Both the known and unknown risks are too severe, and the plausible payoff is not worth it at all.
Let me explain.
The current, dominant type of AI is generative AI, more specifically LLMs, that are, basically, a brute-force effort. People with a much deeper understanding than me describe this approach as inelegant, inefficient, power-hungry and wasteful. That is the type of AI I’m talking about — this is important: I’m not anti-AI generally, I’m quite open to AI. What I’m not open to is shaping the world’s economy, and really the world, around generative AI.
I) The known risks of AI as it is being built right now, and more importantly the risks of the infrastructure being built for AI right now, are well established by now: Immense energy and water needs at a scale that can destabilize whole communities and ecosystems.That draw so much investment that it makes up for a significant amount of the global economy, even though those investments are built through fragile circular financing mechanics that could tear down the whole economy if one piece started wobbling a little too hard. All that hyper-focused investment comes with terrible opportunity costs: Money and effort not spent elsewhere where it might lead to more and better impact.
II) The unknown risks are, by definition, unknown. However, many of the leading researchers and most frontier labs warn of what they call existential risks of their AI, saying if the capabilities of their next models hit some critical milestones it might post a very real risks to end humanity as we know it. This might be hyperbole or even a marketing scare tactic, but for argument’s sake I want to not dismiss it as a bad faith warning and take it seriously here and say: They’re closer to subject matter, maybe they know something we don’t. If that’s the case, if (to quote a leading researcher) the risk of their AI ending humanity is at “>10%” than how could it possibly be worth it.
III) And how it might be worth it or not is maybe the most important point: The plausible outcomes are not worth it because the utopian “everyone lives better through AI though we don’t quite know how” line of argument that AI labs often bring forward seems too vague for me to put any stock in. I haven’t heard a convincing argument from the labs or anyone else that AI would plausibly lead to a future where everyone’s wealthy and has no needs. (If that was the case, if a literal utopia was in the cards, I’d be on board, probably. But it’s not.)
Instead, what we see is a historically unprecedented concentration of resources and capital and market power. AI in its current form accelerates and intensifies this even more. The “best” case scenario that is even a little bit plausible is that a handful of companies and their investors become ridiculously rich (which most of them already are). Like, multi-billionaire rich. But there is no such thing as trickle-down economics, and so this doesn’t seem a worthy goal to me.
In other words, we know that AI — in its current form — brings known, significant, guaranteed negative effects through energy and water consumption as well as opportunity costs. On top of those, it also brings unknown, unclear risks that the AI labs claim might wipe out humanity as we know it. And the plausible payoff would be a few dozen people benefiting while externalizing all these costs to the public?
To me this seems a no-brainer. This is not the path forward. There are better ways, they just aren’t getting the attention and funding they deserve as long as we continue to throw everything we have at LLMs and the infrastructure they need.
PS. I write my blog posts by hand, no AI. Yes, I use em-dashes.
