It all started with this blog post by John Battelle which mentioned Larry Page talking about AI at an AAAS conference.
Who wouldn’t be interested? Harish and I watched the short clip on ZDnet and we were both a bit taken aback by what Page was saying. (You can get a video of the complete speech on this page.)
If you look at your DNA its just about 600 MB compressed, which is smaller than any operating system. Your Linux, windows, any operating system. That includes booting up your brain, right … by definition. So your algorithms are probably not that complicated, its probably about the overall computation.
I spotted the obvious flaw there that genome size is directly related to complexity of the system. Now any biologist worth her salt will tell you that this is simply not true. Gone are the days when we believed that knowing the complete genetic code of an organism will tell us everything we need to know about that organism. Far from it, the questions raised by sequencing genomes are far more than ones it answers!
But what does this all have to do with artificial intelligence or AI? Well apparently, a lot of people are buying Larry Page’s argument! Now believe what you will about the complexity or simplicity of AI. I’m no expert in the field. But to use the supposed “simplicity” of the DNA “program” to prove your point about AI is plain wrong.
I spent some time explaining to Harish the biology behind my thinking and he converted his understanding into a blog post with a clever title. He also went around posting comments in the blogosphere talking about why Page’s logic was flawed and pointed back to his post. Except for a couple of people, most didn’t understand the point Harish was trying to make with all the biology in his post, so let me try it this one time.
Page’s argument as I understand it is:
- Human DNA is simple to understand.
- Human DNA programs for the human brain.
- The human brain makes human beings intelligent.
- Therefore, AI is simple
All cut and dried. What’s wrong? Well, a couple of things. I could debate about how simple or not simple human DNA is, but lets assume it is simple. We’d still be stuck at step 2. DNA does NOT “program” in any sense the human body or brain. Using the metaphor of a “program” is quite wrong and it is this precisely that which leads most people to make mistakes in assuming what DNA can or can’t do.
My point is that the sequence of DNA in a genome is an incomplete description of a living system. Therefore it is not a good way to estimate how easy or hard it would be to build an AI system.
I do not know enough about AIs, nor the current state of research in that field to be able to debate how close or far away we are from building one. But I do know enough biology to tell you that using DNA to argue that AI is around the corner is wrong.
