A couple of years ago the directors of a women's clothing company asked me to help them develop better fashion recommendations for their clients. No one in their right mind would seek my personal advice in an area I know so little about—I am, after all, a male computer scientist—but they were not asking for my personal advice. They were asking for my machine-learning advice, and I obliged. Based purely on sales figures and client surveys, I was able to recommend to women whom I have never met fashion items I have never seen. My recommendations beat the performance of professional stylists. Mind you, I still know very little about women's fashion.

Machine learning is a branch of computer science that enables computers to learn from experience, and it is everywhere. It makes Web searches more relevant, blood tests more accurate and dating services more likely to find you a potential mate. At its simplest, machine-learning algorithms take an existing data set, comb through it for patterns, then use these patterns to generate predictions about the future. Yet advances in machine learning over the past decade have transformed the field. Indeed, machine-learning techniques are responsible for making computers “smarter” than humans at so many of the tasks we wish to pursue. Witness Watson, the IBM computer system that used machine learning to beat the best Jeopardy players in the world.

9 Comments

1. 1. zamalek 12:35 AM 6/21/12

   Quality article on a hot subject, and it cites the online machine learning course by the same author that I saw live this spring. The first lecture of the course expands on some of the points in the article:
   
   http://www.youtube.com/watch?v=VeKeFIepJBU
   
   Thanks Scientific American for making this important subject fun to read.

   Reply | Report Abuse | Link to this

2. 2. senhora 05:03 AM 6/21/12

   Which women's fashion company is this? I am surprised that a computer program can beat the stylists.

   Reply | Report Abuse | Link to this

3. 3. zamalek in reply to senhora 12:20 PM 6/21/12

   Computers are so powerful nowadays that all it takes to beat human beings is it to program them right. I guess stylists can overlook some stuff that computers will not overlook.
   

   Reply | Report Abuse | Link to this

4. 4. senhora in reply to zamalek 10:57 PM 6/21/12

   zamalek wrote: "Computers are so powerful nowadays that all it takes to beat human beings is to program them right."
   
   Fascinating and depressing at the same time. I looked into the online course you mention. It starts in 3 weeks and I may just take it to know what this is all about.

   Reply | Report Abuse | Link to this

5. 5. kiteman 10:41 AM 6/23/12

   Yes, it ran for three years. So, who won? Is this article just made up, to tell a story, or did I miss something? 'the entire process is automated' I guess that is what an algorithm is for. All the helpful suggestions seem to have been thrown in the bin. A confusing article.

   Reply | Report Abuse | Link to this

6. 6. Eco_steve 08:30 AM 6/24/12

   Zamalek : Try getting a computer to translate almost anything at all. Computers are hopeless compared to human translaters. As language is the basis of much thought, computers are clearly only good for one thing : Doing what humans force them to do via brute force programming.
   They may do what we tell them to do fast, but they are still just plain stupid.

   Reply | Report Abuse | Link to this

7. 7. zamalek in reply to Eco_steve 12:53 PM 6/24/12

   To Eco_steve:
   
   It is just a matter of time before computer translation impresses you. Speech recognition was pretty hopeless not so long ago, and it's now becoming a working technology. People were making huge bets against a computer program beating a chess champion, and it evetually happened.
   
   BTW, if you try google translate, you see that it is using machine learning to improve by asking people for better suggestions. With time, it will get there.
   
   You are right that computers are stupid in the sense of following blind orders, but if the orders are done right, the brute force of executing tons of orders in a flash wins the day. What I got from the article and the online lectures is that machine learning automates the *generation* of these orders, not just their execution.
   

   Reply | Report Abuse | Link to this

8. 8. HubertB 05:43 PM 6/26/12

   I will be glad to see better translations. The last time I tried a machine translation, it was so bad I went back to manual. For example it translated the English word pass into the Spanish paso regardless of context. When one car passes another it is pasa not paso. When a solder gets a pass to leave a fort, it is pase. A mountain pass is paso. The program got that one correct. That is only one example. After seeing such a horrible translation, I do not blame foreign language speakers for objecting to Google's attempt to translate English Wikipedia articles into something resembling their own language as written by an idiot.

   Reply | Report Abuse | Link to this

9. 9. zamalek in reply to HubertB 07:06 PM 6/26/12

   HubertB:
   
   Maybe machine learning has already done something. I cut-and-pasted "When one car passes another" from your comment into Google translate, and got "Cuando un vehículo pasa a otro"
   
   I don't speak Spanish so I can't certify that this is a correct translation, but it did get "pasa" not "paso" as you point out it should.
   

   Reply | Report Abuse | Link to this