4 ms·
Everytime I see something like this I am more convinced that deep learning/AI focus should go away from language into other domains where it can make a bigger i
by amunozo 2y ago
Everytime I see something like this I am more convinced that deep learning/AI focus should go away from language into other domains where it can make a bigger impact on less compute.
- rapsey 2y agoOf course it will. AI is driving a flood of development in robotics. A home robot is definitely coming in the next 10 years and an assembly line worker is going to be an extinct job.
- nradov 2y agoWe already have home robots in the form of vacuum cleaners. If you're referring to humanoid robots for domestic chores then forget it, 10 years is way too optimistic. The safety problems alone of having a powerful heavy robot moving around with people and pets will take decades to solve. AI in the form of LLMs is effectively impossible to test for safety critical applications with so few system constraints. The risk of product liability lawsuits will prevent this technology from being sold anytime soon. Some limited industrial use may be feasible because there the robots can be carefully monitored and have limited interactions with humans. Plus there are remaining hardware problems in terms of power and actuators that will take longer than 10 years to solve in a cost effective way.
- JoeAltmaier 2y agoSeems pessimistic. I already have an oven that can burn the house down, a dishwasher that might flood out my kitchen, a refrigerator that can fail and ruin a fortune in perishable foods. Hell, my lawn mower might remove a foot! Why is a robot that might step on the cat, somehow inconceivably more hazardous?
- FrojoS 2y agoThose examples are unlikely to kill you. A humanoid robot can. Let's say it quickly moves its arm on a collision course with your head. The arm can weigh around 10 kg. The impact can easily kill an adult. I'm a bit less pessimistic than OP. It's possible to run safety checks (e.g. collision avoidance) on a lower, deterministic level, and use generative, black box, 'AI' for higher-level planning.
- postalrat 2y agoLike animals the robot arm should also be compliment and not always be moving with full force because it is going to hit things even with active collision avoidance.
- numpad0 2y agoA lead filled baseball bat flying through space is always moving at full force unless some external force is applied.
- rapsey 2y agoWhy would it move quickly? There is zero reason for a humanoid home robot to move quickly.
- usrnm 2y agoNone of those appliances are autonomous
- JoeAltmaier 2y agoPlainly false. They operate without my direct supervision, mostly.
- nradov 2y agoStatic home appliances don't move around and the space of possible dangerous failure modes is limited. This stuff is easy to test for safety. And a refrigerator failure isn't a potential safety issue. LLMs tend to fail in novel ways that are impossible to reason about. What is the expected failure rate for mistaking a human infant sleeping on the floor for a large doll that needs to be picked up and put away? You can't prove that the rate will be zero. We don't have good ways to model and estimate those failures rates in such complex systems.
- JoeAltmaier 2y agoHell, the family dog has some of the same issues. My point is, robots are not essentially new in their home-environment risks. We accept some level of risk now. The pessimistic view that home robots are impossible to deploy until their risks are reduced to zero, is nonsense. My theory: we'll have home robots as soon as they're affordable and have a market demographic. Like any other technology.
- bee_rider 2y agoWe have higher standards for designed devices than sentient ones, especially ones that have a long history like dogs. I love my dog, but appliance with the QA tolerances of dogs would not sell well. Some dogs were bred for helping you work, others were bred as guard dogs. If I went around selling something like a 75% toaster, 25% gun device, and we don’t know which parts come from which, I think I would go to jail.
- rabf 2y agoInteresting article covering humanoid robots in development now: https://james.darpinian.com/blog/you-havent-seen-these-real-humanoid-robot-demos https://james.darpinian.com/blog/you-havent-seen-these-real-...
- fnordpiglet 2y agoIt already has. Language is just the most accessible by most people. But deep learning is not new and is not newly applied to problems. But it’s unclear how much of this article is hype or new. Gigafactory was also supposed to be 100% automated in a much more difficult assembly than putting a few boards into a metal case and gluing a screen on. They failed in practice and balanced into something more practical. (I’d not even in this article they eventually nod that it’s not fully automated just the “important parts”)
- gardaani 2y agoPlaystation factory was also almost fully automated: "PlayStation 4 units come off the assembly line at a pace of about one every 30 seconds." https://news.ycombinator.com/item?id=23748664 https://news.ycombinator.com/item?id=23748664
- brookst 2y agoThere are too many people and too much investment for there to be a singular focus or any coordinated shift in focus.