Comment by photon_off | Hacker News Reader

Comment by photon_off | Hacker News Reader

Here's the freakishly quote: "...Garry Kasparov's brain is freakishly efficient, it's running on about 23 watts while Deep Blue doesn't know how to use energy that efficiently. So, one of the problems that biological computers have solved is how to take in energy and turn it almost into, magically, freakishly efficient computation."

The analogy should work both ways, though. I'm pretty sure Kasparov couldn't add 100 billion numbers together inside of a second, either. Nor is there any amortization of the cost of his currently 23-watt brain; how about the energy expended during years of development of his brain, whereas the computer could be produced with relatively low energy cost?

It should be of no surprise that humans are efficient at calculating things of survival importance, like emotion, decisions, etc. Over the span of billions of years, across tens of millions of square kilometers, under an algorithm where failure means permanent termination, our brains have evolved out of necessity to be that efficient. To me, it is far more freakish how efficient computers are at doing what they do. In under a second, billions of computations can occur in my several year old machine, all in some exact order, and without error. With that in mind, it's even more freakish that it often takes so much time to open Firefox.

The time it takes to open Firefox is dominated not by computation but by waiting for the correct chunks of data to rotate past your hard disk's read head. Even with an SSD I/O bandwidth and latency are probably still the bottlenecks. Even assuming disk I/O took zero time, it's likely that loading Firefox would still dominated by waiting, this time for cache misses.

Actually, Kasparov is doing somewhere around 200 trillion operations in his head every moment, not merely 100 billion. His brain isn't single-purpose, though, as Deep Blue is. The brain, as Montague points out, does all sorts of things consciously and subconsciously that Deep Blue has no chance of doing with only 100 billion operations/second. It's much more omni-purpose than Deep Blue's processors.

And, while it's interesting that Deep Blue can be produced with relatively low energy cost, it's interesting in the way a fly can be produced with low cost compared to a human. The fly is designed for a very specific type of life, and it does a nice job with it. But a fly is not nearly as adaptable as a human, and neither, ultimately, is Deep Blue. Our long setup period allows us to do some incredible tasks that we are still decades from understanding well.

The main difference between current computers and the human brain (life in general in fact), is that we are programming deterministic (well, sorta...) computers. Our brains and bodies are however using probabilistic algorithms everywhere.

Now deterministic algorithms are really efficient at say numerical things, but as we see time and time again the universe is a hostile place which hates programmers and engineers. Think of the checksumming and other overhead needed everywhere because cosmic rays and what-not can corrupt data on disks/going over wires. And things only get worse when our problems get more complex or our solutions have to be distributed. Soon you have more error handling code then normal code.

Probabilistic algorithms accept the fact that stuff will break and we might occasionally fail catastrophically but provided our catastrophic failure chance isn't so big the probabilistic approach is much more robust.

This isn't to say that a probabilistic approach is the only way things should be or that the human body does not require any error checking (in fact, our DNA has an error correcting code built in). But in a hostile and complex environment such as our universe a fully deterministic solution only gets you so far. There is a reason why a lot of AI/robotics solutions are going the route of machine learning, evolutionary computing and similar probabilistic solutions.

Indeed, a typical desktop computer can make a few calculations in the time it takes the light from the monitor to reach your eyeball.

Reply on news.ycombinator.com