But from this comes an incredibly insightful comment from my dad:
Let’s begin with energy efficiency. One of the most remarkable facts about the human brain is that it requires less energy (12 watts) than a light bulb. In other words, that loom of a trillion synapses, exchanging ions and neurotransmitters, costs less to run than a little incandescence. Compare that to Deep Blue: when the machine was operating at full speed, it was a fire hazard, and required specialized heat-dissipating equipment to keep it cool. Meanwhile, Kasparov barely broke a sweat.
The same lesson applies to Watson. I couldn’t find reliable information on its off-site energy consumption, but suffice to say it required many tens of thousands of times as much energy as all the human brains on stage combined. While this might not seem like a big deal, evolution long ago realized that we live in a world of scarce resources. Evolution was right. As computers become omnipresent in our lives — I’ve got one dissipating heat in my pocket right now — we’re going to need to figure out how to make them more efficient. Fortunately, we’ve got an ideal prototype locked inside our skull.
To support a human brain you can't just factor in the calories consumed. You have to factor in the costs in energy to grow the food, etc.Here's what Dad is driving at: in order to sustain Watson for an hour, you might need 50 kilowatts of power, to pull a random number out of the air. To sustain a Ken and Brad's brains for that same hour, you'd only need 24 watts of power.
Watson's power comes from an electrical power plant, which can run at upwards of 33% efficiency. So (as an oversimplification) for every watt of power Watson requires, three need to be produced. This excludes line losses, etc.
Conversely, to feed Ken Jennings 1 watt, we must produce food. This food is grown in a field or as an animal, which is then harvested or slaughtered, processed, packaged, and shipped to a store. Fertilizer was almost certainly produced, and spread on the field, or hormones and high-quality feed was fed to the animal. This supply chain cost a marvelous amount of calories. An impossible number to calculate, but you get my point. Then consider that when a human consumes food, typically only about 20% of the glucose goes to the brain. In a way, a human body is only 20% efficient at powering its brain.
Just because Watson's energy consumption is explicit and the energy consumption of humans isn't doesn't mean that we're faster - or better.
So sure, Watson is a glowing, copper monstrosity that fills a huge server bank. But newer computers invariably shrink. Brains don't. Only a fool would believe that Watson's capabilities will always require massive server banks. Only a short-sighted person would think that only one person could use Watson at a time. While Watson was playing Jeopardy, much of its time was spent at idle. Could Watson have been playing 6 or 10 simultaneous games of Jeopardy without anyone noticing anything? Probably. Google's servers run millions of searches simultaneously...not doing that would be a waste of resources.
So imagine, 30 years from now, when Watson is now the size of a laptop, and 20 people can access it at a time, and it is half again more efficient than the human brain in terms of capability/joule.
What then?
_
0 comments:
Post a Comment