I don't really see this as stupidity. Aside from the prurient uses, think about long distance surgery. What if someone in orbit got struck by a microscopic meteor? This technology, in combination with those that already exist, would allow a trauma surgeon anywhere with a high-bandwidth connection to see the operating area in 3-D, control all available surgical tools, and feel the wound. The same would apply to those living in extremely rural areas (Western Canada or the Outback, for instance), or those that are so emergent that they wouldn't make it to a hospital (an accident on Everest as an example).
Sounds pretty useful to me.
Yeah, but I'm still thinking about those prurient uses.
(In control theory, it's called transport delay and it limits the achievable performance of feedback control systems.)
Not only do the users need frequent updates of the forces and positions (the article mentions 1000/second), but the system also needs to minimize the round trip time for a force to be propagated across the channel and for its resisting force to be propagated back.
I estimate that you can't do fine manual work with delays of any more than, say, a twentieth of a second. And even then, you'll have to learn to work through the system, mainly by slowing down somewhat.
The internet is not useful for this application because of the round trip delays. Not only are they too big, they're statistically uncertain.
Now, if you're going to do this job with brute force communication speed, you'll have to have a dedicated circuit, the most direct possible route (no satellite links!), and the lowest-overhead messaging protocol. So if you are using this approach and you want a surgeon in Sydney to operate on a patient in Topeka, I think you'd need a dedicated fiber optic line running directly through the center of the earth.
Another alternative, which I suspect they're working on, is to give each participant near-immediate locally generated feedback, based on a model of what's on the other end. The resulting force feedback would be modified by later-arriving data actually coming from the remote end. The model itself would be incrementally constructed in real time, being very rough at first and then getting better as the user continued to 'operate.'
But the construction of the model is quite problematic, especially when the object on the other end is connected to an essentally unmodelable nervous system.
IMHO, a problem nearly comparable to that of AI.
<)B^)