2026-03-10
Self-replicating factories are real
Disclaimer
Main
- Reacting to this IMO bad take by Dario Amodei that even if you have a "country of geniuses in a datacentre", it could still take multiple years to cure all diseases using this.
- He still imagines biological labs similar to today, with a regulatory process similar to today, and supply chains for reagents and cultures similar to today.
- I am like, no, all of this will be self-replicating. You will have AI control the transport and supply chains, you will have AI control the power plants, you will have AI manufacture all the bio research equipment, and you will have AI sitting on top making high-level decisions on whether to build more power plants or purchase more reagents or similar. That's what a "country of geniuses in a datacentre" lets you do.
- If all of this is AI-operated, none of this has to be at human scale. Your transport can be drones that are milimetres length wide, your reagents can literally be manufactured on site if that's cheaper, you can miniaturise a lot of equipment now that humans don't operate it. The end state is an entire industrial power and supply and manufacturing chain sitting inside one building, instead of spanning a few continents.
- So much of the reason we build anything - cars, factory floors, house, etc - on the scale of metres instead of milimetres is because we want humans to operate all that. If AI operated it, this constraint isn't as tight.
- Physical constraints are still real. Like maybe you do need a minimum of 10 ml of some reagent to get a statistically significant result, and maybe manufacturing 10 ml of reagent requires a metre-sized distillation chamber or something. But again, the constraints are so much tighter.
- If you miniaturise even harder than that, you get Yudkowsky's scenarios of artifically engineered biological organisms doing all the work, or even nanotech that eats the planet.
- More importantly than this, curing all diseases is a very conservative goal compared to what you could do if you did feel you had a country of geniuses of datacentre in your control.
- Like, yeah, why waste compute on curing all human diseases, when you invest that compute in researching how to kick off an RSI loop instead, or tiling the planet with solar farms (or whatever other energy source the AI decides on), or inventing a religion to mind control humanity into getting out of the AI's way, or researching how to invent nanotech instead.
- I am not saying nanotech will get instantly invented. I am asking why will a country of geniuses in a datacentre bother to cure human disease if they genuinely feel inventing nanotech is only 10 years of expected research time away.
Subscribe
Enter Signal username, email address or whatsapp phone number to subscribe. You will receive atmost one update per month.
Alternatively, you can subscribe via RSS, or a third-party service like blogtrottr
Comment
Enter comment