Arnold Kling  

The AI Cure

For Students... Easier to Break than to Fix...

Vinod Khosla writes,

I was asked about a year ago at a talk about energy what I was doing about the other large social problems, namely health care and education. Surprised, I flippantly responded that the best solution was to get rid of doctors and teachers and let your computers do the work, 24/7 and with consistent quality.

Most of the article is about health care. I think that the point is equally valid about education. A lot of education is feedback: the test gives the teacher feedback on what students are not learning individually and collectively. Right now, a computer could grade a multiple-choice test, but such tests offer only limited insight into how the student is thinking about what you are teaching. As computers learn to grade short-answer questions and essays, the real revolutionary possibilities will kick in.

Comments and Sharing

COMMENTS (7 to date)
Tom West writes:

A lot of education is feedback

I'd say, the *majority* of education is persuading students to make the effort the learn.

From my teaching experience, compared to that challenge, everything else (communicating the material and evaluating progress) is relatively trivial.

It's why I could never be a full-time teacher and am in awe of good teachers. After all, how many of us could convince people to do something they'd rather not, for several hours a day, five days a week, for *free*.

Seth writes:

A lot of everything is feedback.

Hasdrubal writes:

Look into Computerized Adaptive Testing that's used for technical certifications (Cisco, Microsoft, etc.) That software does a very good job of identifying your weaknesses through adapting follow up questions to your previous answers.

Grading short answers and essays can probably be done by a Watson style expert system at the same level as standardized test graders do. After all, they tend to use heuristics and grade based on whether or not the student addresses specific points in their answer. (It was easier to game short answer questions than multiple choice questions when I was in school: String the key words you went over in class into a coherent sentence and bam! you've got full credit.)

Insightful feedback and tailoring further material or instruction methods to the student based on short answers or essays might still be beyond computers today. But in my experience, it's not something that teachers do a lot of, either. Maybe they'll pick out one or two weaknesses to follow up on, but I've only very rarely had a teacher adjust their curriculum to address something they found in a test. And in those cases, they were things that a multiple choice test could have picked up anyway, nothing so subtle that you needed to parse essay questions to find.

Maybe the real gain from computers will come from a division of labor standpoint: The mechanical "remember this" or "learn how to do that" type of things will be taught by computers who have infinite patience and can easily adapt to an individual's pace. Then the more conceptual or creative tasks will be taught by humans. I.e. most lower level language, science and math courses would lend themselves well to computers (with maybe teachers available to answer questions that the machines just can't parse,) but philosophy, English composition and English lit really require human interaction. I'm pretty sure I would have done just as well learning calculus from an "interactive textbook" as I did attending lectures, but I can't imagine a computer giving me the feedback I needed to learn to build a decent paragraph.

Thomas Sewell writes:


It's a commercial product we've used with great success for a Charter school and with Home schooling.

Essentially, once someone has basic reading comprehension and basic arithmetic knowledge, can teach them the rest through early college math courses.

Math is obviously easier than other domains for this approach, but the program does look at the actual work performed, not just the answer (it's not multiple choice). It also has explanations and examples for each math principle being taught and follows a competency testing process where the student is required to periodically demonstrate a some previous knowledge to ensure they haven't forgotten anything. If they have, they are required to take a refresher section. It's also very flexible in terms of the order of learning, allowing students a menu of what they want to work on next.

So this is all closer than you might think....

Arnold Kling writes:

Aleks looks very interesting. Anyone else know anything about it?

Floccina writes:
Right now, a computer could grade a multiple-choice test, but such tests offer only limited insight into how the student is thinking about what you are teaching. As computers learn to grade short-answer questions and essays, the real revolutionary possibilities will kick in.

Interestingly facebook, texting and twitter are making people better communicators though practice. Spell checkers and hopefully someday soon, a good grammar checker will male some aspects of good writing easier.

Paavo Ojala writes:

I took statistics for behavioural science in It was far superior to the introductory statistics courses they offer in University of Helsinki. I hope they would have more advanced courses too.

It's not too special what they offered, but somehow paying for it and not allowing me to skip anything i didn't know just made it easier to learn the stuff.

But there's a lot easy improvements to make and i hope we would get a lot of competition. THose stanford courses of course offer free alternatives. They seemed to be rather good too.

Comments for this entry have been closed
Return to top