Rogoff: Artificial Intelligence and Globalization

Ken Rogoff wonders if replacing people with intelligent machines, e.g. pocket economics professors complete with holographic images instead of university professors, will be a bigger factor than globalization and outsourcing in explaining changes in global job and wage patterns in coming decades:

Artificial Intelligence and Globalization, by Kenneth Rogoff, Project Syndicate: Today’s conventional wisdom is that the rise of India and China will be the single biggest factor driving global jobs and wages over the twenty-first century. High-wage workers in rich countries can expect to see their competitive advantage steadily eroded by competition from ... Asia, Latin America, and maybe even some day Africa. ... But I wonder whether ... another factor will influence our work lives even more: the exponential rise of applications of artificial intelligence.

My portal to the world of artificial intelligence is a narrow one: the ... game of chess. You may not care a whit about chess... But the stunning developments coming out of the chess world ... should still command your attention. Chess has long been the centerpiece of research in artificial intelligence. While in principle, chess is solvable, the game’s computational complexity is almost incomprehensible. It is only a slight exaggeration to say there are more possible moves in a chess game than atoms in a universe.

For most of the twentieth century, programmers were patently unsuccessful in designing chess computers that could compete with the best humans. ... The computers gradually improved, but they still seemed far inferior... Or so we thought. Then, in 1997, ... IBM’s “Deep Blue” computer stunned the world by defeating the world champion Garry Kasparov. Proud Kasparov, who was perhaps more stunned than anyone, was sure that the IBM team must have cheated... But the IBM team had not cheated. ... Since 1997, the computers have only gotten better, to the point where computer programmers no longer find beating humans a great challenge.

Only a game, you say? Perhaps, but let me tell you this: when I played professional chess 30 years ago ..., I felt I could tell a lot about someone’s personality by seeing a sampling of their games... Until a short while ago, I could certainly distinguish a computer from a human opponent. Now everything changed like lightning. The machines can now even be set to imitate famous human players – including their flaws – so well that only an expert eye (and sometimes only another computer!) can tell the difference.

More than half a century ago, the godfather of artificial intelligence, Alan Turing, argued that the brain’s function could all be reduced to mathematics and that, someday, a computer would rival human intelligence. He claimed that the ultimate proof of artificial intelligence would be met if a human interrogator were unable to figure out that he was conversing with a computer. The “Turing test” is the holy grail of artificial intelligence research. Well, for me, a chess game is a conversation of sorts. From my perspective, today’s off-the-shelf computer programs come awfully close to meeting Turing’s test. Over the course of a small number of games on the Internet, I could not easily tell the difference. ...

What’s next? I certainly don’t feel safe as an economics professor! I have no doubt that sometime later this century, one will be able to buy pocket professors – perhaps with holographic images – as easily as one can buy a pocket Kasparov chess computer today.

So let’s go back to India and China. Globalization proceeded at a rapid pace through much of the last century, and at a particularly accelerated rate during its last two decades. Yet the vast body of evidence suggests that technological changes were a much bigger driver in global wage patterns than trade. That is, technology, not trade, was the big story of the twentieth-century economy (of course, the two interact...) Are we so sure that it will be different in this century? Or will artificial intelligence replace the mantra of outsourcing and manufacturing migration? Chess players already know the answer.

Comments

[a]

The first time I was told that my job can be automated, I was not pleased, and I could not believe it. That was three years ago, and since then I start to believe that it can happen, since I have automated lots of my own work --- and I want it that way.

Like no one today wants the job of the machines, tomorrow, no one would want to do my job (once "research" becomes rountine data-mining that computer would do by itself).

What would I do? Good question! If all of the human need in material ways are being taken care of by computers and machines, wouldn't we live in a Utopia where we can do what we really want to --- be an artist? Tell a story creatively? Cook a wonderful meal and enjoy nature...

I would like to contribute in making my work doable by machines so that I can enjoy LIFE.

[spiny widgmo]

The problem is going to be the mad grab for power and wealth as the means of wealth production collapse into a few hands. I'm not sure our political and economic system can adapt quickly to both reward innovation and prevent envy and loss from causing problems.

[Pancho Villa]

What is the 'real' meaning of 'A.I.' in this context?

[cm]

Machines have to run on energy, and I figure anything that can approximate "interesting" subsets of human intelligence will have to expend far more energy than the human brain+body. OTOH the infrastructure to support electronic brains may be cheaper than that needed to support the "equivalent" human lives.

My take on the whole AI business is that when it comes to the complexity of "human affairs", I cannot see how a computer will attain the effectiveness and complexity of human brain function and human group dynamics (a vastly underappreciated aspect of human intelligence).

Automated phone systems are not AI, but I think how a phone menu tree relates to actual (qualified) support is pretty much what we are looking at here, in a metaphorical sense.

[cm]

a: I hope that wasn't "whining". :-)

[cm]

a: Don't worry. I don't know exactly what comprises your job, but I'm confident not all of it can be automated, and you will be tasked with performing only the parts that can't effectively, and will be the constant "bottleneck in the system" as whatever is automated will be done faster than yours.

I mean this quite seriously. BTW, I read a book called "Slack" in which the author poses the thesis that a large contributor to today's work stress is the cutting out of much of the "mundane" work that while boring, gives the brain, and in fact your whole system, a chance to recuperate (exercising judgement or otherwise expending "mental energy" at high intensity constantly is quite exhausting).

[calmo]

'Artificial Intelligence', such a rollicking good term, no?
[Check the label. No artifical ingredients for me please. I demand the genuine organic ingredients.]

Is the phone tree-menu automated systems the paradigm of AI? (Does Turing having something to say about this?) Are people still programing computers to play chess or is that a done deal? (Are these levels of AI comparable?)

I need that chess-smart computer to handle my service calls --the AI phone recipients. I would pay handsomely for this.

says...

I am no expert on AI, but I found some of the points displayed above not to hard to comment:

spiny widgmo : I agree with you. It is a social/political question, and it is up to the citizens of this country to decide what they want. But they need to understand the issue. So, again, education is important.

Pancho Villa: he said in the sense of chess playing, so I guess it is the old traditional way of number cruching, not the Google intelligence of using Baysian statistics.

cm, I enjoy debating with you a lot! Here is a quote from http://intensityindicators.pnl.gov/highlights.html

"Taking a long-term perspective, and using the simple Energy/Gross Domestic Product (E/GDP) ratio, the amount of energy needed to produce a dollar's worth of goods and services in the U.S. economy fell by more than half between 1949 and 2002." and if you read on below you will see they claim "because factors the affect intensity that are unrelated to the efficiency of energy use are included in the ratio. One example of such an 'other explanatory factor' is the shift of economic activity out of the industrial sector and manufacturing, that use large amounts of energy per unit of output, into service industries that use only very small amounts of energy ."

Energy would not be too big issues for AI. The volume of computation that a computer can do at low energy cost compared to Human Brain is simply not comparable.

[a]

cm, you are not whining.

True that it may take a while for AI to manage some of the routines and boring parts of "human affairs". And from my experience at work, once we learn and explore, the exotic inevitably becomes rountine and boring, and we quickly learn to automate it and move on to the next untraveled land.

calmo, there is already a machine that passed the Turing Test for the last three years in a row. Read the entry in Wikipedia. What is stunning is that there is a new way of being intelligent other than palying chess: it is more based on large observation database, and some statistical way to find solutions by observing the observations (like Baysian or Neural Network). In a sense, it is like how a baby learn to talk. Not much "Intelligence" in traditional way of "thinking and evolving", such as Chess playing, but works wonderfully by Google, for example.

So if you want to handel your service calls by computers, here is a suggestion: you collect the information of all your human servicing calls. And you train the computer with the data. At the end you might get a little monster that talks just like your TYPICAL servicer. And it is not even that hard to implemented.

Maybe there are cases you want to put your best servicer to handel the most complicated cases, but for the most part, your need might be satisfied quite well.

And I won't be surprised that such program would be realized in practice in a few years.

[a]

"I certainly don’t feel safe as an economics professor!"

If the pocket professor can also do research to answer sophiscated questions, or even better, to form the right question, ask it and solve it, ok.

I don't know if a chess player, at his best, knows how to ask the right question. I am not a chess player (used to play Chinese chess in 5th grade, but that is all).

[a]

cm, on "Slack":
I am trying to understand what you said about it, so I went to Amazon.

From my own work experience, "madane" rountine work bores me and sadens me. I like challenge of new problems. But the process of inventing new methods is hard to be monitored, efficiency is hard to be measured. Some times instead of programming, I was reading papers for half a day, or chatting with my colleagues to find out what is new in their work and how that affects the way I innovate.

That comes to the notion that, for example, IT is very hard to manage, because quite often the person on the task is maybe the only person who knows the demand of the work and hence can judge the efficiency of it.

All that said, I know people who actually enjoy routine reporting and if you push him/her to learn new things, he/she feels burns out. But if you let them learn some new things, they love it as well. I think the burn-out factor is more related to a person's capacity. Requesting people to do beyond their mental ability in knowledge worker space is to stress them out.

Of course, the manager, typically don't know the subject fields can also stress out workers by making unreasonable demand.

At the end, managing knowledge workers in the traditional way would be counter-productive.

[a]

Mark, I have not read much on your blog regarding open source. The Economist recently has a special report on Open Source, and I think Open Source would have a profound impact on the way we work, and particularly on the way the economy works.

What is your take on that? Love to read about it on your blog, and read other people's informative comments.

[Pancho Villa]

Well, I was rather thinking of 'A.I.' as some kind of in this case, metaphorically seen, 'artificially created intelligence' for example in China, mainly due to the 'technologies transferts'

says...

a: "the amount of energy needed to produce a dollar's worth of goods and services in the U.S. economy fell by more than half between 1949 and 2002"

Here the mirror image of that question: by how much did the purchasing power of said one dollar fall? In other words, if the dollar buys about half in 2002 than in 1949, we break even.

[cm]

Related point: If you count only US energy expenditure, and offshore energy-intensive stuff, then, well, have we one BIG energy-efficiency achievement here.

Previous comment also mine.

[cm]

a: How long have we heard that automation will create more leisure. The "paperless" office, etc.

The actual track record implies that use of resources and demands for speed ("efficiency") are always pushed to the max. Or in organizations that are managed by bean counters who view things through their balance sheets, it's pushed to the max subject to fiscal constraints, which means that you are asked to deliver the max of measurable product per unit of time, regardless whether you are paid hourly or salary. And churning of paper is managed in terms of dollars per day. If anything becomes cheaper (on the balance sheet), more can be churned. We are using more paper, and work at quicker pace (and except for select industries, arguably also harder), than ever.

[James Sedgwick]

Chess is not a hard problem for computers. The reason is it's easy to look ahead at many branches, and it's fairly easy to define if a branch is good or bad. A game like Go is a much more interesting problem for AI, since it does not lend itself to such bruteforce solutions. The worlds best Go problems are not close to any competent amateur player, and people have been working hard on the problem for many years.

[Richard]

As a guy who makes a living as a programmer, I have some issues with this article.

(1) The number of chess moves doesn't exceed known atoms in the universe. To pick a game which has these characteristics, use Go.

(2) Most chess strategies used by machines use brute force methods, i.e., from the point at which the game currently stands, work at all possible first moves, second move responses, third move counter-responses and so on. The method is computationally intensive, but given a machine with lots of MIPS and a lot of computational time you can match the moves of grandmasters.

(3) More generally, writing code that alters itself, or alters its own expectations, is very hard work. Most expert systems rely upon a statistical analysis and scoring method that is validated against expert judgements; that system is more consistent (and less expensive) than hiring a team of experts.

(4) Go (mentioned above) is a game where even a player of modest ability can beat the computer. The number of possible moves is so vast that brute force methods don't work well for figuring out good Go moves. Under those constraints, an artificial intelligence system really does need to somehow operate in a neural network fashion, learning from its own mistakes.

(5) Given the continuing increases in computer power and the decreases in expense, I won't be surprised to see more expert systems set up. For lots of techinical problems where multiple factor need to be weighed and the information is incomplete, expert systems could prove to be a useful adjunct. But I won't be trading in my physician for a machine any time soon.

[calmo]

What is this thing called Intelligence, love? Shall we let those Chess champions decide? The Go Champs? The Cray XT3? The designers of the XT3?
We need more of it to decide if it's the genuine article and not that other stuff, yes? Am I sufficiently programmed to be able to decide what is intelligent? Already you might think that I have conceded too much by allowing the possibility that machines might be more than instruments of intelligence ("what is intelligent" rather than "who is intelligent"). Maybe not.
Maybe you think Artificial Intelligence is confined to machines. (more or less elaborate programming) Maybe not.
Before the Age of Artificial Intelligence we knew what this thing was called love.

[lvs]

I definitely believe that it will be possible to have intelligent machines that will pass the Turing test. I believe the knowledge is out there on the internet - and what is required is to build a computer that can learn from that knowledge. But a computer is without senses. It is like putting a person in a box where he cannot use any of his senses except reading lots and lots of books. I have written more in my blog:

http://indradhanush-laal.blogspot.com/

[Saul Wall]

It is likely that once the understanding of how intelligent systems operate is attained and are constructed, A.I.s will need to be educated. In addition to being provided all the books, music, art and movies available, which would take quite some time for an A.I. to learn (as opposed to simply store) they would also learn about humans by pretending to be humans on the Internet. By interacting in chatrooms and blog comment sections, they could test hypotheses and make new observations. Maybe someone you have discussed things with online is really an A.I. in a lab somewhere; learning how to maneuver robots on campus during the day and increasing its language and social skills at night.

End transmission.

[Mike Riley]

I have to agree with Richard, this article doesn't go beyond the scope of finite systems, and human intelligence does.

[Julio]

I happen to know a lot about this particular issue and have no doubt that most jobs being currently done by people will be done better (soon) by machines.

You may of course disagree, but grant me that for a moment.

This being an economics blog, then, here's a question:

Is an economy where machines do most of the work compatible with private ownership of the means of production?