1290
Rich people don't understand us at all.
(lemmy.nz)
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 26 Aug 2026
1290 points (99.5% liked)
Work Reform
17362 readers
3340 users here now
A place to discuss positive changes that can make work more equitable, and to vent about current practices. We are NOT against work; we just want the fruits of our labor to be recognized better.
Our Philosophies:
- All workers must be paid a living wage for their labor.
- Income inequality is the main cause of lower living standards.
- Workers must join together and fight back for what is rightfully theirs.
- We must not be divided and conquered. Workers gain the most when they focus on unifying issues.
Our Goals
- Higher wages for underpaid workers.
- Better worker representation, including but not limited to unions.
- Better and fewer working hours.
- Stimulating a massive wave of worker organizing in the United States and beyond.
- Organizing and supporting political causes and campaigns that put workers first.
founded 3 years ago
MODERATORS
In most demos, yes. Though they have managed to do things that vaguely seem correct, in very specific tasks in very controlled environments and often showing a behavior you wouldn't want.
Like for example, they got it trained to pull out a dish rack and put a dish in, and push the rack back in. Then a human pulls out the rack and the robot pushes it back in. Which they like to show as it responding to the environment, but in real life the human wanted to do something with the rack, why would the robot incessantly keeping it from being out be desired?
Another one where it could pick up clothes, so long as the clothes were within a set square meter and the background was solid white.
The remote control is the mechanism to try to train them to work independently, but it is a tall order. To get the relatively more straightforward movement of operating a car, they took millions of miles of data to feed it. These robots are aiming for far more intricate work and have a severe lack of access to training data.
Further, imagine any task that becomes very difficult just by putting on gloves. Congratulations, there's a task machine learning can't help with, because they don't have tactile in the input loop.
Interesting issue regarding the gloves situation. However I might add that gloves are often used for safety whereas a robot might not need that safety. For example, pulling weeds.
Though, I think you’re getting at something subtle. It’s not about tasks which require gloves. It’s about any theoretical task which would become harder if you happened to do it with gloves on.
That’s a lot of tasks when you think about it. For starters, I use my fingers in a kind of “friction test” to make sure I’ve fully cleaned a dish of any oil. If my finger slides, it needs another scrub.
It's not about the machinery struggling with gloves, it's just the difficulty of capturing the data when a human really needs to touch/feel.
Like that 'friction test', the point is when they use machine learning, they want to feed all the inputs and outputs from the human operator. They don't have a good way of capturing 'feel' (you can have a camera to capture what eyes see, you can have microphones to capture what the human hears). Similar on human inputs, for a car, they can instrument the pedals and steering and they pretty much got it all. If a human hand is doing something, well, if the human is local they might be able to make a glove to capture all that, but our sense of feeling can be fouled so easily that trying to instrument it tends to ruin it. If the human is remote then you can just get controller style inputs.
Yeah, this should be researched. How would you produce an analog signal to represent touch semantics?
You ever heard of that paper, *what the frog eye tells the frog brain” (Or something like that) ? It covers what’s effectively semantic information, not bits of data, getting sent over the central nerve. Edges, flashes, …
I wonder what the skin might tell the brain.
I also wonder if there would be a way to bounce a static field, or something, off the surface… maybe vibrate the material, etc,,, to approximate these sort of details.
I'll bet that all the data being harvested by smart glasses is now going to train those models. Basically it provides a first-person view of someone doing stuff with their hands all day, walking around, talking, etc.
Problem is that is too indirect. Imagine trying to do that intricate work on sight alone. Like trying to do stuff with numb hands. Without tactile feedback, we are useless in a lot of these tasks. But the tactile feedback is not instrumented. To the extent that machine learning works, it needs to have all the inputs and outputs instrumented, and feeling is left out.
We can see something done and have an explanation and internalize it and then 'get a feel for it' later. But AI doesn't currently feature this ability to really translate that sort of instruction to an entirely different set of sensation.