AI that even my flawed human brain can grok – just!

Artificial IntelligenceI have undertaken and completed a few MOOC’s now. Most successfully through Coursera, as the format works for me, the institutions, lectureres and materials are really high quality, and the real-time pacing (and pressure) has really motivated me. The subject that has had me stumped is AI – Artificial Intelligence. I started different courses, on Coursera and Udacity, and always bombed out a few weeks in, when the maths (rather than the computing) left me in the dust. My maths is OK (so I thought) but I just do not have the foundation needed for some of the concepts of Statistics, probability, Linear Algebra, and just some of the notation. The pace of the courses just does not allow the time needed for my poor, limited human intelligence to catch up. Some of these courses are switching from real-time to permanently open, self-paced, so I should have another go.

The point is I found an AI course I could complete, and I learned a huge amount. University of Edinburgh’s Artificial Intelligence Planning course is pitched at some very practical planning and sequencing problems that machines have been able to solve through the application of (yes) Maths, programming algorithm’s and some creative approaches to reduce complexity (and time). The style of the course worked for me, the concepts were delivered in an open way that allowed students to use just any programming language they liked to solve the challenges and assignements. I used Processing as I am familiar with it’s fast-start Java base, but I also had access to all the Java libraries for maths and data structures. I completed it but also passed with distinction, which I am naturally chuffed about.

So what have I learned? I have learned that artificial intelligence is really about harnessing what computers can do (process lots of stuff methodically very quickly) to compensate for what they are less equipped to do (intuit, guess, experience) to arrive at solutions to optimisation and sequencing problems that Human’s may not have the time, or familiarity for things like experience, intuition to be of any use. At Human speed the AI approach (searching a problem space) seems inefficient, but it can be proven to be very accurate and at computer speeds, time is typically not an issue.

Yes, my maths was an obstacle still, but the real breakthrough for me was understanding that (in this form of AI at least) the art is respecifying the problem for the machine, rather than simulating the human approach.

The lectures were very good. The challenges and assignments were tough, unscaffolded, but fair tests of understanding. For University of Edinburgh this was one of their largest cohorts, although it was probably the most intimate MOOC I have taken part in so far – around 10,000 registered and active at the end of the course. The organisers made heavy use of social forums to let sudents participate and support each other. The most novel of these was the use of Second Life to host virtual turorials, where typically only a few dozen were present. At these numbers, it did work, but they were unstructured, and with every minute of study time precious to me, I think they could have been higher value. If nothing else, it gave a chance to ask questions directly to the professors, which is an opportunity most MOOCs have not matched in my experience.

My thanks to University of Edinburgh and the instructors, Dr. Gerhard Wickler and Prof. Austin Tate, for finally giving me a door into AI that I could fit through.

No comments yet.

Leave a Reply

Powered by WordPress. Designed by Woo Themes