The AntiBuzz: Yeah right.
Now and again I want to share what I study and think about on a daily basis – that is, artificial intelligence (AI) – and couldn’t think of a better way to put it than this week’s tagline.
The technophile gets off easy in this column. I have warned against technophobia, almost pathologically so, but we forget how easy it is to be overconfident in technology too. Much as the curmudgeon is guilty of too much distrust, the common gadget-lover is too optimistic about our computing prowess. It is as if all the world’s problems are in the bag, all we have to do is sit back and wait for computers to fix everything.
Computers don’t solve problems; we do. Over a year ago I waxed philosophic on AI. That article seems so long ago, yet the pop culture impression of AI continues to amuse and flummox me. I suppose as dentists you could relate, (is it safe?); what I research and work on has nothing to do with Cylons.
Instead, I want to address the common notion that “computer people” or “they” have it all figured out and know how to get machines to conquer anything from tax returns to robot soccer. I am behind that curtain; I am one of those “them” that does the work that makes your life a little easier. However as amazing as high tech solutions appear it is more amazing how little we know, how much is up for grabs, and how inadequate some of our knowledge and technology is. In short, computer scientists; both the researchers and the plucky entrepreneurs, are in the business of solving problems – any problem – and using computers to solve it.
Artificial intelligence is a broad umbrella, and carries the baggage of “replicating human intelligence” that either scares people, or makes them laugh. In reality, this umbrella covers any system designed to make intelligent decisions automatically. Often this entails methodologies that are general purpose and “let the machine figure it out” though there can be varying degrees of human coaching on the matter. Put another way, if computer science is the science of problem solving then the sub-field of artificial intelligence is concerned with problems where the path to the solution is not obvious, but might be sussed out by a machine capable of noticing patterns.
“How do we notice patterns?” is the open question here. If you want a practical example, think spam filters: Putting spam in one folder and non-spam in your inbox is an easy process by itself, but recognizing the spam is not.
Last week I attended a talk on stem-cell research and was confronted with a more noble problem. A cancer researcher is comparing the gene sequence of healthy stem cells with cancerous ones, hoping to find the defining traits of the disease. The problem is the same – learn to recognize the spam – gene sequencing is the poster child for shaming modern computing power.
Once their process was refined, the genes of a tissue sample could be mapped in about a month. Try to understand that this isn’t a work month. This isn’t like talking about how you spent a month working on a project, or spent a month dieting. We’re talking about a progress bar where the time remaining was “one month” and a computer, or group of computers, just sat there and computed for one continuous month, with no power failures or crashes or coffee breaks. (Try to imagine the required faith for this sort of thing: “No that computer is not frozen it’s just taking a really long time.”)
When this lab performed its first gene sequencing it took a year, and when the human genome was first sequenced it took two. The size of this data is enormous. As this researcher joked, “they couldn’t just attach it in an e-mail,” which highlights how much we take that for granted. You have the genetic code of a cancer stem cell on a computer and there isn’t enough RAM to open it with any standard machine, and even if you just wanted to take a peek with Notepad the process would be very slow.
Large data is sort of like the sun, in that just looking at it can be computationally painful, much less computing anything useful from it. In many ways, enormous data might as well be on old magnetic tapes, as a byte-by-byte scan is about the only practical way to do anything with it, (Fun fact: a lot of large data storage is still done with magnetic tapes).
Modern computing has obvious advantages, but less understood is how it is still rather inadequate. Our ability to accumulate data has exceeded our ability to do smart things with it, and doing smart things with information is precisely what I study; it’s not all chess and Terminators over here. My education in computer science thus far has impressed upon me that technique is just as important as anything else. We aren’t going to crack big data problems by just throwing more processors at it. Quickly summarizing large structures, or effectively approximating solutions are things that require real people with real intelligence to figure out. The machine is still just the tool.
Knowing that what I study could actually cure cancer, I’m feeling a little superheroic so I will wrap up this week with these melodramatic parting thoughts on AI: Where ever there is a difficult decision, AI will be there. Where ever a pattern or connection can be found, AI will be there. And where ever a strategy game can be be brute-forced into a smoldering crater, yeah, AI will be there too.