Machine Learning Will Solve These World Issues

What important world issues do you envision machine learning solving in the next few years? originally appeared on Quora - the knowledge sharing network where compelling questions are answered by people with unique insights.

Answer by John Langford, Machine Learning Researcher and Vowpal Wabbit author, on Quora.

In the next few years, I expect machine learning to solve no important world issues.

I do expect it to solve a growing number of problems at companies which in aggregate make various little bits of the world work better while developing the tools needed to actually solve some world issues. People will be able to get the news they want easier, the human/computer interface will work better, and more routine tasks will become automated.

Further out (10-30 years), we can expect machine learning to contribute to:

1. Automated driving. How does this solve a world problem? Well, when transport becomes a utility rather than a possession there will be fewer cars, they will contribute less pollution, and there will be far fewer wrecks. As a bonus streets won't be clogged with parked cars all the time.
2. Mass personalized education. In The Diamond Age there is a book that many, many people use to educate themselves. The details will differ, but I see no reason why this can't be effectively enabled using machine learning techniques. A future world where everyone can learn at their own pace is enabled with machine learning, and machine learning is needed, particularly in early years of education where interaction is key.
3. Rationalization and personalization of healthcare. The United States has setup a perverse system where the most expensive solution to a problem wins which is why healthcare costs twice as much as in other countries. I expect the revolution to really start with personalized healthcare via personal devices (Fitbit, Microsoft Band, etc...) both because these areas are lower risk and because they aren't as entrenched around old ways of doing things.

In general, machine learning is a good plausible tool in situations where:

1. The right answer is not obvious.
2. The setting is digitized.
3. The same decisions must be made many, many times.

It's not all roses though. The elimination of many relatively routine jobs needs to be handled well by society to maintain a civil society. The potential is there---society as a whole is richer. But that is not at all the same as saying that it will work out well.

• Machine Learning: How can one become a good machine learning engineer?
• Research: Whose or what lab's research in machine learning do you find the most interesting?
• Deep Learning: Will Deep Learning replace all other kinds of machine learning?