Many of you may have seen this Andrew Ng video from December 2017 in his “State of AI” lecture at EmTech:
I watched it again this weekend and thought it well worth the time. It was good to be reminded about what it took to become an Internet company in order to benefit from the huge business opportunities that a new platform presented. That shift to a new way of doing business also re-defined what it meant to build products. The AI era is doing the same thing by changing the methods that we need to deliver AI products. If you have limited time, start at 20:00.
17:00 What does it mean to be an Internet company?
20:00 What does it mean to be an AI company?
23:30 AI Product Managers
— As PMs we like to start with drawing out the UX, but really, don’t start with wireframes. Instead, AI PMs first create/collect a data set that represents the problem space. Only then do they ask an engineer to iterate on the problem to deliver 90% accuracy for the problem to be solved.
— Every product needs to start with strategic data acquisition as the very first step. Ng tells the story about Blue River, who painstakingly collected a unique set of data — heads of lettuce — until they had the most unique data set for the problem. John Deere snapped them up for $300M as a result of having a solid and UNIQUE data set built up over time.
— There is an important cycle that starts with a Data set→ trained so that informs a Product → that Users interact with to get value and leave ineraction traces → that gets More Data → to continuously inform a Product → that keep Users coming back.
So, are you an AI Product Manager?
— Did you look at your problem space and collect the unique data source first? And only then ask Dev to strive/achieve the percent accuracy the market requires for the problem being solved?
— Do you know your AI performance/benchmark data COLD?
— Is your training data set UNIQUE and differentiated?
— Do you have every bit of legal data exhaust being created pumping back into your product?
— Are you constantly looking for additional sources of data that will continue to improve your product?
This framework is a great reminder about how to start AI products!