Tag Archives: ML

Learning Machines and the Future of Academics

Institutions will try to preserve the problem to which they are the solution.

– Shirky Principle

Learning, How does it work?

There has been progress and evolution, but the roots of our academic institutions are essentially medieval. For all the progress that has been made, for a variety of technical and social reasons, the whole system is largely hierarchical and based on lineage. Expertise was always a scarce resource and the time and investment to transfer expertise required physical proximity. While we have passed the stage where participation is solely based on exposure to Latin and Greek as a filter to participation, on several levels there is still a strong bias that filters on context and circumstance. Subtle and sublimated as that bias might be, these filters may be least obvious to those who benefit most and have the power to change anything. Consequently, we have not yet fully leveraged available human potential. The present is not evenly distributed.

I did the advanced track of the Artificial Intelligence and Machine Learning classes from Stanford in the last 10 weeks and wanted to share a few thoughts. Technology and efforts like these have the potential to change everything about how people learn.

Information wants to be free. The marginal cost of broadcasting the highest quality lectures from the best teachers on the planet is trending to zero. That is changing everything. Stanford is changing it. MIT is changing it. Khan academy is changing it. Know It, Busuu, and probably a long list of education start ups I don’t even know about are going to be changing it. There is a good chance that this transition disrupts the university system as we now know it. In every sense of the word disrupt.

The two Stanford classes had a slight overlap in topic, but they were qualitatively very different. There are plenty of reviews about the classes already. What I’m interested in is slightly meta.

How do people learn? What is the incentive? What is a measure of progress? And what can they do with the things they learn?

In particular, what is the most effective path to someone being productive in a deeply technical skill?

What Possibility…

Now, back to the Stanford classes. The contrast between the two approaches provoked some thoughts.

Sebastian Thrun started out by stating the purpose of the AI class is 1) to teach you the basics of artificial intelligence and 2) to excite you. They definitely delivered on that purpose. Sebastian and Peter Norvig split time covering an introduction to AI. The format was video lectures with embedded questions at the end of most videos. The format was the same for the lectures, the homework, the mid-term and the final. Watch the video, answer the questions. Done.

The ML class used a different format. This system was also video lectures. Andrew Ng’s presentation in the video medium felt natural and flowing. This class didn’t cover as many topics but almost every topic came with a programming assignment. Questions in the lectures were not graded, but there were weekly review questions and the programming assignment. You were allowed to resubmit the review or the assignment multiple times with no penalty, so you were graded, but getting 100% was really a measure of persistence. (Andrew seemed excited to be teaching people. The thank you he gave in the concluding lecture was so heartfelt, I wanted to give him a hug. Andrew made me feel like it was a true honor for him to teach this. The honor was all mine.)

At the end of AI, you had learned some things from watching videos and got graded for submitting a bunch of forms, at the end of ML, you had learned some things from watching videos and had the opportunity to have working code to train neural networks, support vector machines, k-means clusters, collaborative filtering, etc. On the one hand you have people tweeting their scores on the other you have people BUILDING SELF DRIVING CARS!

By three methods we may learn wisdom: first, by reflection, which is noblest; second, by imitation, which is easiest; and third, by experience, which is the most bitter.

– Confucius

Take The Next Step

Which brings me to the point I really want to make. What is an education? What are academics? The pursuit of knowledge and understanding? These things people are doing and building to help people learn are amazing and inspiring, but that’s only one part of the equation… the dissemination of knowledge, understanding and skill. What about creation?

Scientific journals which at one point served as a filter of quality and point of aggregation, now act as a barrier to access. If the internet does anything, it disintermediates. This current system of publishing slows and prevents the access to information. The ‘publish or perish’ tenure and research grant funding process also creates disincentives to open collaboration. I imagine a future where collaboration in research is open and transparent. Experiments aren’t done in secret and partially explained in publications, but all the methods and results are shared and updated in real time. Like a Github for science. If I can’t replicate results, I open an issue. If I find an interesting pattern or insight, I open a pull request. Everyone can see everything, streams of open data. This has to scare the living hell out of some people. There is a lot of time, money and personal identity tied up in the current system, but its essential inefficiencies are not beneficial or necessary.

(Aside: Resistance to this is not unlike what we are witnessing with the entertainment/media/copyright lobby that resulted in the SOPA legislation, where entrenched institutions attempt to prolong the last gasp of disrupted models of creating and capturing value. That resistance won’t fix outmoded approaches to servicing markets that no longer exist, it can only stunt the growth of emerging models. Piracy is a distraction. People always made copies and traded media, just the medium has changed. People have also never had a problem trading for something they value. People love to buy stuff they love. Compete in the market. Embrace the opportunities.)

Finally, there would be a benefit to more permeability between academics and industry. There are literally billions, maybe even trillions of dollars worth of technology shelved in universities. Industry loses on the opportunity to greater utilize research and expertise while academics often lose touch with the reality of practice in the wild. We all lose on the prospect of more abundant prosperity. In most cases there is a risk and implied disincentives to transition between the two disjoint worlds, which in some sense don’t even respect each other’s reality. If the system facilitated a properly incentivized flow of people and information in both directions, I can’t help but believe both would be better off.

The open questions now are how quickly the transitions happen and to what extent to those personally attached to the status quo resist. Same story, different stage.

tl;dr We live in amazing times. You can either understand how to build self driving cars or you can’t. You will either help others do it, or you won’t. Get ready for the next level or better, help make it happen. Special thanks to Stanford, Andrew Ng, Sebastian Thrun and Peter Norvig for their contributions to the future.

Advertisements

%d bloggers like this: