It's Man v. Machine on Jeopardy! this week as IBM super-robot Watson takes on former champions Ken Jennings and Brad Rutter. At The Atlantic, we're using Watson as an occasion to think about what smart robots mean for the American worker. This is Part Two of a three-part series on the exciting and sometimes scary capabilities of artificial intelligence. Read Part One -- Anything You Can Do, Robots Can Do Better.
The conventional wisdom used to be that becoming a knowledge worker represented the best path to a prosperous future. The advent of offshoring has increasingly called this proposition into question.
Today, offshoring is impacting knowledge workers -- that is, people with software jobs -- across the board. Someone with a software job could eventually be replaced by a computer similar to the one that currently sits on his or her desk.
Jobs in fields such as radiology, accounting, tax preparation, graphic design, and especially all types of information technology are already being shipped to India and to other countries. This trend will only grow, and where offshoring appears, automation is often likely to eventually follow.
IF ROBOTS CAN PLAY CHESS...
The automation of software jobs is tied closely to the field of artificial intelligence. To gain some insight into how artificial intelligence works in the real world, let's consider computer chess. In 1989, Garry Kasparov, the world chess champion faced off against a special computer called Deep Thought. Deep Thought was designed at Carnegie Mellon University and IBM. Kasparov easily defeated the machine in a two game match.
In 1996, Kasparov faced a new computer developed by IBM called Deep Blue. Again Kasparov defeated the computer. In 1997, IBM came back with an improved version of Deep Blue that finally defeated Kasparov in a six game match. This represented the first time that a machine had defeated the top human chess player.
Since then, computer chess has continued to progress. In 2006, the new world chess champion, Vladimir Kramnik, lost a match against a German software program called Deep Fritz. While IBM's Deep Blue was a completely custom computer about the size of a refrigerator, Deep Fritz is a program that runs on a computer using two standard Intel processors. It seems highly likely that, in the near future, a program like Deep Fritz, running on virtually any cheap laptop computer, will be able to defeat the best chess players in the world.
Could a computer formulate a strategy for an important legal case?
When we think of what it takes for a human being to be a world chess champion, most of us would probably agree that it takes a certain degree of creativity--at least within the confines of a highly defined set of rules. Yet, creativity is a trait that we are very reluctant to ascribe to a machine--even if that machine can beat a human at chess. This tendency to be somewhat underwhelmed by the accomplishments of machines, may have something to do with the fact that the working of the human brain remains a mystery.
Who can say what is going on in a human chess master's head when he or she plays a match? We simply don't know. And therefore it becomes to us something mysterious and especially creative. In the case of the computer, however, we know exactly what is happening. The computer is simply calculating through millions of different possible moves and then picking the best one. It is using a brute force algorithm. The computer's advantage arises not from the fact that it is genuinely smart, but because it is almost unimaginably fast. It's natural for us to give this brute force accomplishment a lower status than the creativity and precise thinking exhibited by an exceptional human being. But the question for us here is: will that protect us from brute force algorithms that can do our jobs?
... CAN THEY PRACTICE LAW?
If you agree that the game of chess requires creativity within a set of defined rules, then could not something similar be said about the field of law? Currently there are jobs in the United States for many thousands of lawyers who rarely, if ever, go into a courtroom. These attorneys are employed in the areas of legal research and contracts. They work at law firms and spend much of their time in the library or accessing legal databases through their computers. They research case law, and write briefs which summarize relevant court cases and legal strategies from the past. They review contracts and look for loopholes. They suggest possible strategies and legal arguments for new cases that come to their firms.
The first thing you might guess about these attorneys is that they are already subject to offshoring. And you would be correct: in India there are already teams of lawyers who specialize in researching case law not in India--but in the United States.
What about automation? Can a computer do the lawyer's job? One of the primary research areas in artificial intelligence has focused on creating "smart" algorithms that can quickly search, evaluate and summarize information. We see the fruition of this body of research every time we use Google or any other advanced Internet search engine. We can expect that such smart algorithms will increasingly be used in the field of legal research. The software may start out as a productivity tool to make the lawyer's job easier, and then eventually evolve into a full automation solution.
Obviously, it is easier to automate some parts of the lawyer's job than others. For example, finding and summarizing relevant case law would be a likely target for an initial effort. As I pointed out with the radiologist, automating even a portion of the lawyer's job will quickly result in fewer attorneys on the payroll. What about the more advanced or creative aspects of the lawyer's job? Could a computer formulate a strategy for an important legal case? For the time being, this may be a challenge, but as we saw in the case of chess, a brute force algorithm may ultimately prevail. If a computer can evaluate millions of possible chess moves, then why can it not also iterate through every known legal argument since the days when Cicero held forth in the Roman Forum? Would this be a "lesser" form of legal creativity? Perhaps it would. But would that matter to our lawyer's employer?
HUMANIZING THE ROBOT MIND
Although the practical applications of artificial intelligence have so far emphasized brute force solutions, it is by no means true that this is the only approach being taken in the field. A very important area of study revolves around the idea of neural nets, which are a special type of computer that is built upon a model of the human brain. Neural nets are currently being used in areas such as visual pattern recognition. In the future, we can probably expect some important advances in this area, especially as the engineers who design neural nets work more closely with scientists who are uncovering the secrets of how our brains work.
One thing that probably jumps out at you as we speak of lawyers and radiologists is that these people make a lot of money. The average radiologist in the United States makes over $300,000. In fact, we can reasonably say that software jobs (or knowledge worker jobs) are typically high paying jobs. This creates a very strong incentive for businesses to offshore and, when possible, automate these jobs. Another point we can make is that there is really no relationship between how much training is required for a human being, and how difficult it is to automate the job. To become a lawyer or a radiologist requires both college and graduate degrees, but this will not hold off automation. It is a relatively simple matter to program accumulated knowledge into an algorithm or enter it into a database.
For knowledge workers, there is really a double dose of bad news. Not only are their jobs potentially easier to automate than other job types because no investment in mechanical equipment is required; but also, the financial incentive for getting rid of the job is significantly higher. As a result, we can expect that, in the future, automation will fall heavily on knowledge workers and in particular on highly paid workers. In cases where technology is not yet sufficient to automate the job, offshoring is likely to be pursued as a interim solution.