Hacker News Reader

Comment by canjobear | Hacker News Reader

When we say the brain has poor computational accuracy, we’re usually talking about the conscious brain we’re aware of. But our low-level motor actions and perceptions, coordinated by the brain, require a lot of precise computation. These low-level brain computations are the thing to compare to AI, not our conscious thinking. Our conscious mind is more like low-precision software running on top of an enormously powerful computer.

View on HN

> But our low-level motor actions and perceptions, coordinated by the brain, require a lot of precise computation.

That accuracy is more likely achieved through fast, analog feedback loops than precise calculation.

Like a giant stack of op-amps. (https://www.computerhistory.org/revolution/analog-computers/...)

How come that in many sports people a good because they basically know everything by heart?

Prof. basball batters, for exampley, are completely baffled by prof. throwers of other sports.

We don't have analog hardware. Nerves are digital. The best we can do with them is pulse-frequency modulation.

I'm not convinced real neurons are just binary summators. But regardless of that, there's also chemical transmission involved, which is an analog thing.

Right, but the end state of all that is "neuron fires" or "neuron doesn't fire", or at best "neuron fires n times per second" - a binary state.

We have some behavior that depends on very exact timing of neuron spikes (e.g. determining direction of sounds by connecting signals from both ears), so that's kind of analog - though it does get reduced into a binary state, in the end either the "detector for offset of x microseconds towards left" fires or not.

It's unlikely we compute in any conventional sense, the hardware is reality, and is going to exploit every available effect that is energetically efficient.

Letting FPGA's "go" into the analog realm is an interesting window: https://news.ycombinator.com/item?id=21253267

Glial brain cells: https://news.ycombinator.com/item?id=22161192

> But our low-level motor actions and perceptions, coordinated by the brain, require a lot of precise computation.

But we don't actually do that precise computation though. try repeating any action exactly and you will see some inaccuracy.

no, you will see some variability

Accuracy is a function of variability and bias.

How do you even define accuracy for an analog process which solves for unknown formulations? Not saying you can't, but afaik there's no ground to call them accurate. It's certaintly chucking a lot of data in functional ways.

Reaction time and the physical operation of the body are things that seem obvious to me. Try to design a control system for a machine that could replicate a gymnast, for example.

Or a person walking. Boston’s Dynamics is just now making robots near to animals in their walking ability.

Sorry, but in my book responsiveness and functionality doesn't have a thing to do with accuracy (neither does parallelism if I may throw it in there).

Well, I guess I don't know what you mean by accuracy, then. I'm thinking of the accuracy required to hit the right note in a guitar solo, or the accuracy required to time a clean & jerk correctly, or the accuracy required to track a fast-moving object with the eye.

Those things can be surpassed by machines, of course, but if you wanted to design a machine to do all of them? With human efficiency?

> But our low-level motor actions and perceptions, coordinated by the brain, require a lot of precise computation.

Such as?

Identify a ball in a box of items, toss it up in the air, and then catch it. Congratulations, you've just exhibited more computational power than most computers are capable of, including some very precise physics simulation and inverse kinematics.

Reply on news.ycombinator.com