On this day, Professor Andrew Ng was at Imperial College London. He was not only there as a speaker for the (Data Science Institute, DSI) Distinguished Lecture programme but to also commemorate the 5th year anniversary of the existence of the DSI.

It was not a big suprise to see the full lecture theatre completely full of postgraduate students, vehement to hear from Prof. Andrew. Somewhat surprising though, was the fact that Prof. Andrew came in with a yellow notebook rather than a laptop as we were all expecting some sort of digital presentation. In fact, he chose to speak to the audience directly and use the white board to write a few important points. To me, this simple act, in itself, is an illustration of a true leader. He started off by speaking about his teaching courses at Stanford University. I personally think he is doing a fantastic job at revolutionalizing teaching methods. In particular, he refers to his own teaching system where students watch videos before coming to the class, and projects/ideas are discussed during teaching sessions.

On the same note, since he is also trying to automate grading system as well, teaching and research assistants spend more time discussing projects/ideas with students. Other important topics discussed include but not limited to:

  • ANI (Artificial Narrow Intelligence)

    • Up to now, there has been major development in supervised learning, where an algorithm is fed with thousands, possibly millions of examples before one tests on unseen examples. This form of Machine Learning is at the forefront of most technological application these days.

  • AGI (Artificial General Intelligence)

    • On the other hand, we are also far from achieving AGI - a state where an agent will be able to take decision on its own. Clearly, the human brain is a complicated system and although much work has been done in the Neuroscience research field, having a full understanding of how the brain works, remains a delicate topic of research.

  • Small Data

    • Another interesting aspect of the discussion was about small data. Machine Learning algorithms are data crunching in general. It is very hard to work with small data set. As an example, if we have access to millions of medical images, we can do a good job at predicting whether a patient is suffering from, say, cancer. What would we do if we had only 10 images?

  • Generalizability

    • Algorithms are in general very specific, for example, if an algorithm is designed to predict the market value of an object, it is very less likely to be used out of the box for an almost similar problem. Although, methods such as domain adaptation and transfer learning are currently being explored, Prof. Andrew believes that we are still far from building generalizable algorithm.

  • Jobs

    • Job security is a heated topic of discussion. While some people believe that people will be jobless as robots will take over our jobs, others believe that there will be a radical change in the job type itself. For example, Prof. Andrew says that there will be jobs where half of the work might be done by the robot but the constant presence of the human is still required. One funny example which Prof. Andrew gave was that we are still far from having a robot which will be able to give us a perfect hair cut!

  • Education

    • Moreover, education matters! If we want to move forward and embrace AI, we have to develop our skill sets in such a way that we will be able to survive a rapidly changing environment. Education allows us to harness the talents of brilliant people and a global online education system not only promotes learning but also enables people who cannot afford university fees to access world class learning materials.

  • Creativity

    • One question from the audience was: what is creativity? Prof. Andrew humbly said, 'I don't know the definition of creativity'. He further went on telling on that when his group publishes a paper, people say that this was a creative piece of work. However, he admitted that the work is actually quite dirty in the lab, especially when working with deep neural network where one has to build the 'right' architecture and do hyperparameter tuning well.