Machine learning may be most useful in tiny, embedded, offline processors

Originally published at:

1 Like

The same kind of logic applies to space probes – when it comes to adding sensors and robot arms and stuff, money is no object, but it’s all moot if the simplest decision (like “turn the camera to look at that thing”) requires consulting Earth over a multi-hour communications delay.

At the opposite scale, if/when it’s possible to inject someone with millions of medical mapping robots, if they all needed constant communication with a server then they’d be limited by raw bandwidth (as well as power). But if you could make each one smarter, and need fewer control messages, the saving would be multiplied by millions.

I think there’s, like, a general principle that as a system becomes bigger, the value of even a little bit of autonomy at the periphery grows non-linearly.


Erratum: I guess you probably mean antenna instead of battery?

1 Like

Just so long as they have safeguards against going Nazi.

Like the tiny processors in the autonomous flying killbots?

Distributed Society of Mind

Yup, thanks!

1 Like

I’d love to buy a cyber-hans I could slap on my vintage motorcycle engine, so I can stop worrying every time I think I hear some new sound. As it is, just mounting an oil pressure gauge helped a lot. Modern vehicles with full ODB systems could probably do much, much better (and maybe already do).

TIL what I am: I’m a Hans. No wonder my industry is dying and I now have a job doing this.

Hey did someone say “smart dust”?

This topic was automatically closed after 5 days. New replies are no longer allowed.