Back to the analog future for AI?

Interesting video about the strengths and weaknesses of analog vs digital computing in AI, the multiverse, etc.

At about the 17:42 mark, an Austin TX company’s analog chip vs the digital “chip”.
I can imagine Edge AI using a much smaller analog chip, rather than the huge digital thing.

We’re Building Computers Wrong (for artificial intelligence)
https://youtu.be/GVsUOuSjvcg

The description of the 17:42 chip, at 16:00 minutes, as analog while it is sort of a “variable resistor”… well, that “stretches” my imagination?
LOLOLOLOL

And, paraphrasing here, analog computers are less than 100% precise. Digital computers assumed dominance due to their speed and precision.

AI does NOT require perfect precision. If an image is judged as 95% likely to a chicken, or 98% to be a chicken… in the output, it’s still a chicken. (Ie Analog is the “good” vs the “perfect”.)

:rowing_man:???
ralph

1 Like

The description of the 17:42 chip, at 16:00 minutes, as analog while it is sort of a “variable resistor”… well, that “stretches” my imagination?
LOLOLOLOL

Digital is “on-off” while analog is a slider == “variable resistor”

And, paraphrasing here, analog computers are less than 100% precise. Digital computers assumed dominance due to their speed and precision.

Not strictly true. With digital you can get results only in increments (whatever the unit is) while with analog you can get all the values between the increments. The infinity of analog values is larger than the infinity of digital values. The difficulty is converting the analog value into a number so that you can do computations.

The precision they are talking about is not inherent in the system but in the variability of the hardware used to build the physical apparatus.

The reason for digital computing taking the lead is that they are universal problem solvers, you just create new software. Analog computers are hard wired to do one kind of task. When I first started programming many of the machines had plugboards like the one shown in the video. That’s not a “universal” solution by any means.

Analog computers should be perfect for training neural networks because they would do only ONE thing, matrix multiplication.

BTW, Tesla moved away from Nvidia GPUs to their own chip to build Dojo but even Dojo should be no match for an analog “matrix multiplier.”

The craziness of quantum is that it is both digital and analog at the same time (wave and particle).

Denny Schlesinger

5 Likes

<< The craziness of quantum is that it is both digital and analog at the same time (wave and particle). >>

Well, we are all both waves and particles at the same time.

The “craziness” of a quantum state is the superposition of multiple answers (states) at the same time.

Schrodinger’s cat is both alive and dead.

Alan

3 Likes

Denny. Give TMF some more tries,?
I’m hoping youll get this in email form, too.

Anyway. After just slogging along, its getting easier.
Ive NOT spent a lot of time AVTIVELY learning the new site. I’ve just “surfed”, and read the posts by others who are passing along information on how to navigate the site.
:alien:
ralph

1 Like

Denny. Give TMF some more tries,?
I’m hoping youll get this in email form, too.

Ralph:

The Fool community has been a welcome companion for over 20 years, entertaining and educational, and often just plain silly. It would be sad to lose it. Fortunately I stumbled on a way to avoid the “New, Improved, Interface Calamity.” I get emails of the posts that interest me. I read them in my mail client and occasionally I ‘Visit Topic’ to reply.

Win-win: Keep the community, avoid the website clutter.

The Captain

3 Likes