Yeah, if they execute that idea successfully they could massively massively reduce the infrastructural costs of rolling these out.

If. But I don't think it's impossible.

Or you can just hire a housekeeper to fold your towels rather than spend $800,000 on a robot every 5 years to do it

You are making the strongest argument for generalization, I agree with the logic of your argument. There is benefit in being able to rapidly reconfigure production.

Like, the power of a 3D printer or CNC machine to customize its output to be essentially whatever the fuck you want every single time is important.

But I still disagree with humanoid robots with general purpose AI being the solution to that.

For example, I saw one Chinese robot which was a painting robot, and you could wear a sleeve and paint a thing and the robot would replicate those motions precisely. So there you have a general purpose painting-bot that can be reprogrammed for any painting-purpose within minutes.

I agree that’s powerful, and I can see how the same would be true for, like, the screw-driver-bot or the welding-bot. And each of these generalized use cases could benefit from AI (“does it need another coat of paint? Did I miss a spot?”).

So a factory composed of general purpose robots makes a lot of sense and wouldn’t even be more expensive than a factory of custom-made purpose-specific robots.

But the “does absolutely anything bot, with general purpose AI that can reason about how to solve the problem” this I absolutely don’t buy into.

In part, because I genuinely don’t see AGI ever being that good. But that’s a criticism of AI rather than these robots.

So ok assume AGI is or becomes that good, I still don’t see the “does everything” robot being useful compared to a team composed of painter-bot, welder-bot, sorter-bot, etc.

I also don’t see the real utility in the humanoid form.

The argument is that since they’re humanoid they can use any tool designed for humans.

But surely redesigning those tools for the robot itself has already been proven more effective given decades of factory automation.

And sure bipedal motion is more versatile than wheels, but a factory floor is an artificial space so you can make the factory floor compatible with wheels and have a robot with less moving parts, less moving parts being almost axiomatically the better choice.

Hell, we even redesigned our cities and planet to be suited to cars and trains. So the idea that bipedal motion is more versatile, while absolutely true, is still not very strong to me.

I agree that there are surely some use-cases for bipedal or quadruped robots. I can think of some for sure but I have to really think to discover a purpose where it’s not simply easier and more reliable to redesign the space for wheels instead.

Each capability an AI humanoid robot has seems suitable for edge-cases basically. That’s why I don’t see it. Sure, probably there will be enough of those edge-cases to make them a profitable industry but its niche.

Isn’t that the exact opposite though?

ASICs are specialized chips. They are highly use-case targeted. Microcontrollers are extremely specific, often single-task specific.

The robot that has been designed for the task of cutting metal panels exactly the same over and over seems like the equivalent of an ASIC or microcontroller here. Something designed for a specific purpose or small subset of purposes and capable of doing that purpose in an extremely optimized manner.

The AI humanoid seems more like a CPU here… capable of solving any problem but due to its general purpose nature, not optimized for any particular problem.

To continue pushing the analogy, to me humanoid AI robots seem the equivalent of trying to sell CPUs to mine bitcoin. The world already has a better solution.