Information, Intelligence, Knowledge

Notes from the podcast Machine Learning, Recommender Systems, and Future of AI

I never felt comfortable with the term AI. I used to call it Augmented Intelligence (for my students) and preferred the term ML to AI. That was before I heard this talk. Now there is no going back.

Prof Jordan was amazing. It is the kind of interview you can listen to sitting glued to your seat for 2 hours!

I was jotting things down mostly paraphrasing from Prof Jordan’s replies.

  • What is happening today is not AI. It is an intellectual aspiration.
  • Thinking that you can have AI at the level of human intelligence is like ancient Greeks thinking about traveling to the moon some day. It is a similar challenge because we (humans) don’t even understand intelligence.
  • We have no clue how the brain does computation. We don’t even know what the signals in the brain are – it can be chemical, it can be electrical it can be ions or whatever. It is a problem (a fantastic one) for the next century.
  • We are completely dim about how thought emerges from the brain.
  • It will take us a couple of hundreds of years to get to human level intelligence.
  • What we have is Machine Learning – systems that learn from data and help us make decisions.
  • Right now we are good at imitating some aspects of human intelligence.
  • True conversational AI is not going to happen in our life time.
  • What goes for AI now is mostly Machine Learning
  • What we have now is pattern recognition and some level of prediction.
  • True prediction is not possible with just data sets. You need a lot of context and more.
  • What we have now is a little bit of reasoning¬† and it is not AI.
  • Prediction is difficult because the world is highly stochastic and we have massive uncertainty.
  • Gradients (as in Gradient Descent) are amazing Mathematical objects.
  • Statistics goes back to 250 years. It started as “Inverse Probability”.
  • The name statistics comes from “study of data for state”.
  • There are two types of statistics (decision theory) – Bayesian and Frequentist.

There were many interesting threads in this conversation:

  • Gradient Descent,
  • Optimization,
  • Recommendation systems,
  • Facebook,
  • Privacy, and more.

It is a podcast/video worth watching/listening to. You can find an outline here.

Michael I. Jordan: Machine Learning, Recommender Systems, and the Future of AI | MIT | Artificial Intelligence Podcast