Andrew Ng, a co-founder of Coursera and the chief scientist at Baidu Research, recently gave a talk at Deep Learning School. At one point in his presentation (10:16:42), he described the current process of building desktop and mobile apps in the industry: product managers and designers iterate on ideas and interfaces before implementation. However, with rising popularity of machine learning and its complexity, it is difficult to grasp what is achievable with those algorithms. Consequently, it is hard to write product requirements or specification, so Andrew often receives the question: “What can artificial intelligence or deep learning do?”
He offers two rules of thumb that, while not perfect or complete, are helpful in thinking what is possible with machine learning.
- Anything that a person can do in less than one second.
- Predicting an outcome of a sequence of events.