I think we recognize that humans are extremely versatile. Show them how to do x and in no time they can repeat the steps. Give them a while and they will probably learn to do it faster and better.
A key question is can AI make that kind versatility and learning possible for humanoid robots. That kind of learning could be the gold standard for humanoid robots.
Will AI make that possible? Maybe one day but probably not any time soon!!
it’s a question of numbers. Once data centers have a number of transistors and connections comparable to human brains, AI will outpace HI (Human Intelligence). Humans take years to learn everything. Robots can download anything whenever it’s needed from the moment they come out of the production canal.
AI is a hive mind often found in Sci-fi.
The Captain
This is quite a long way off, I suspect. Those robots can’t just download “a solution”, because what makes us human is devising multiple solutions in real time and then choosing the one that seems most appropriate. (That’s why different people solve different problems differently, to overuse a word.)
Tesla can’t just download a path from A to B, because some days there’s an Amazon van double parked. It can’t even have “in reserve” an “Amazon van is double parked” solution because it might be a pothole the opened, or a kid running into the street after a ball. So it needs not just “the solution” but all of the possible variable inputs that can lead to multiple solutions, along with the intelligence to sort and decide which is preferable.
This example is just “driving”. If you’re going to have a robot fully competent to handle multiple kinds of tasks in a factory setting, then there are hundreds of other kinds of parameters that must be “learned,” the kinds of things that humans intuitively already have. Sight, sound, coordination, smell, pressure of fingertips, knowing torque, and so on.
And this is where I wax poetic about the astonishing capabilities of the human brain: I can remember where I put a screwdriver 6 months ago (sometimes), but forget where my hat is. OTOH, I can watch a movie from 20 years ago and say “Oh, yeah, I’ve seen this.” Somewhere, buried deep down, is the memory of that movie - and all the other movies I saw 20, 30, and 40 years ago. (No, I might not remember every jot and tittle of the dialog, but it’s astonishing that I remember any of it at all, isn’t it?)
Consider the stimulation we get daily on a drive to the store, the billboards we pass, the trees we see - and remember so much of it somewhere, deep down there. It’s mind blowing. I would hazard a guess to say that given the compactness of the space (of the brain) and the size of data centers (with racks upon racks of servers and giant hvac appliances) that we are decades and perhaps more decades away from replicating jthe capabilities of the human mind with the use of silicon (or whatever other artificially manufactured devise we invent.)
Note: I’m not saying that Tesla (for example) can’t download the basic capabilities of FSD (or successors) onto multiple cars, obviously they can. I am saying that it’s going to take a very very long time before the sort of multitasking capability of the human mind is approached and available for replication to other devices at the push of a button.
Yes, very much that, but also the simple fact that we humans are NOT solely (nor even mostly) individuals, but rather our “selves” via social connection and language are shmeered out across time and space to ancestors and colleagues and wise but nevertheless imaginary beings in songs and literature giving us advice.
Yes
That’s the output, what is seen. What is not seen is the input. No two individuals, maybe twins excepted, have the same inputs, different parents, parents genes, different religions, different geography, different age, etc., etc. Each one has a supercomputer for a brain vastly more powerful that the current mightiest datacenters. Each one creates a personalized world view. This is what allows “humans to devise multiple solutions in real time.” Right now all the Tesla EVs feed just one or a few datacenter brains.
Once humanoid robots multiply there will be more diversity if they feed multiple datacenter brains. To replicate humans they would need individual datacenter brains. That’s a longtime off!
The Captain
Yep. These are some of the things that will make a robot that can “do all” something way in the future. A trivial example is running a factory motor. The human running it, will check the oil, will check the bearings, will check power consumption, and with all those things will determine the heath of the motor. The robot can do all that. However, the human, maybe even unconsciously sometimes after 10+ years of experience, ALSO uses their sense of smell to determine if the motor might be “burning” a bit of the oil, and then might suspect that the bearing are a little out of whack, or maybe a seal has sprung a tiny leak, or maybe someone forgot to change the oil, or maybe someone changed the oil but inadvertently used old oil, or maybe one of 100 other things. The robot can’t do that. Even if the robot has odor sensors added, even the MOST sensitive odor sensors, there is no way for the robot to “learn” what a suspicious odor is, and what is nothing to worry about. That’s because when training on videos of humans, or even when given human instruction, there’s no way to describe that odor. It is something innate in humans that live for years, and experience odors all their lives, and then experience odors and isolate a specific odor and then make the connection to something in the motor. Now someday a robot may be able to do so, but not anytime soon. Now, don’t get me wrong, they do have odor sensing devices, even very good ones, but describing odors is very difficult - usually you need to cause the odor to happen and then see the sensor reaction and record that as “odor 34652”. But that’s not enough for that faint whiff of motor oil burning ever so slightly at the very beginning of the potential issue. And as more and more burns, you get the real characteristic odor which might be picked up by robot. But that isn’t all, the human may have a “feeling” about a motor, maybe they felt an ever so slight vibration that is different than it used to be, or different than the other motors, and the human may have filed deep in the recesses of their mind, “hey, I should keep special watch over that motor because I have a feeling about it”.
That’s not to say that someday robots may learn how to do all this stuff, they probably will. But it isn’t on the order of years, it’s more like decades at best. It’s similar to all those folks that seem to think that we can switch over to all electric vehicles in 10 or 15 years. That’s not possible for many reasons, it’ll take a few decades at best.