Last Friday, Karthik Muralidharan came to give a seminar at the University of Colorado Boulder on a very interesting experiment Karthik, Abhijeet Singh and Alejandro Ganimian had done in Delhi. Their paper "Disrupting Education? Experimental Evidence on Technology-Aided Instruction in India," has an appendix that surveys many experiments with using computers for education. In the talk, he summarized the message there as follows:

1. Hardware by itself contributes very little to learning. The slogan "one laptop per child" is a slogan that may work well politically, but "one laptop per child" by itself doesn't do much to advance education. 
2. Teaching software designed to complement regular classroom instruction contributes little to learning. The likely reason is that most kids in India are far below grade level in their understanding. So both the regular classroom instruction and computerized instruction at that same level go over their heads. 
3. Teaching software designed to find what level a child is at and instruct them at that level has large effects. This is the implication of their experiment, which randomly gave vouchers to some kids to attend an existing after-school center in Delhi that had this kind of computerized instruction. (Other kids, the control group, didn't get this instruction, but took some tests and answered surveys with the promise of getting a voucher later on.)

An important methodological point was that the progress made by the students in the computerized instruction wouldn't have shown up in a test that focused on their understanding of grade-level material. They made great progress on material that was in the curriculum for lower grades, but grade-level material was by and large still above their heads even after three months of the computerized instruction. It would have been truly miraculous if the computerized instruction could have jumped them up several grades worth of competence in only three months. It wasn't that good! But the students made a lot of progress from where they were.

I find this research very exciting. It both backs up what I claimed in my column "The Coming Transformation of Education: Degrees Won’t Matter Anymore, Skills Will," and points to some of the pitfalls that can lead to false starts on the road to computer-driven learning. 

The other lesson I draw from this research is that computer-driven learning has plenty of paths in which to progress. Even if all progress in computer-driven learning stopped in the US, it would progress in other parts of the world. Indeed, computer-driven learning is likely to look much better in comparison to substandard instruction by humans than top-notch instruction by humans. So there are many niches in which computer-driven learning can flourish even before the day it becomes competitive with top-notch instruction by humans. In the Delhi context, adapting to the level of each child would have been a virtually impossibility for the human teacher of a large class of kids spanning many grade-levels of competence. 

Besides having the capability to adapt to the level of each child, computer-driven learning has some other key advantages. First, the computer may show up more often than a human teacher will. Second, a computer program can push a child to constantly exercise herhis brain, in a way that sitting in a regular classroom may not. Even doing homework may typically involve a much greater fraction of time in dilly-dallying unproductive for learning than the fraction of time spent dilly-dallying when doing a computer learning program (assuming a low-skill human babysitter present in both cases to discourage flagrant shirking).

These days I am constantly amazed to realize how far we are inside the possibility frontier of what we can eventually accomplish with better social-science knowledge. Even if all progress in hardcore engineering, physical science and biological science (so that we had to make do with minor tweaks on current technology in those areas), progress in social science alone has the potential to generate massive improvements in human well-being. And the marginal contribution of social science is likely to be even greater if it leverages improvements in engineering, physical science and biological science. That is one reason "Economics Needs to Tackle All of the Big Questions in the Social Sciences." Karthik, Abhijeet and Alejandro are doing a praiseworthy job as part of that effort.