The Mainstreaming of Artificial Intelligence

Artificial Intelligence, Machine Learning, Data Science, Deep Learning, Neural networks……oooh, the buzzwords keep cometh.

This past year, 2017, was full of these buzzwords in the technology space. Every new startup have these buzzwords in their pitch now. Every company has job openings looking for people with these new skills.

To fill my curiosity, I started down the path of learning about machine learning. There was quite a bit of rehash of statistics, correlation, significance and so on. So far so good.

Quite a bit of uses cases can use machine learning. We see examples all around us on the most common apps we use. Enterprise software can also use machine learning in many ways. In CRM for example, you could use machine learning to not only predict which leads or opportunities will close, but you can create early warning systems. For example, in the case of renewals, machines could detect complex patterns and determine which customers will churn and if there is any early action that can be taken by the customer success teams. There is an entire industry around customer success brewing for the past couple of years now.

Things get a bit murkier when we start looking at Artificial Intelligence. There are higher level use cases where AI could be hugely beneficial. Driverless car is the most visible example of AI.

A person recently was killed by a driverless car. Clearly, the technology is not ready. (Human drivers can also kill when negligent, but in this case it seems a human could have avoided the fatality). Regardless, the bottom line is that machine based AI is not ready to match human intelligence.

Artificial Intelligence started to take hold in the 50s and 60s with some very early experiments. IBM had developed the Deep Blue by the end of 80s. At the rate at which technology has progressed in other areas, humans should have been ruled by machines by now.

Of course, nothing like that happened. AI never took off as expected. The likely reason is that scientist took a detour to understand intelligence and the human brain in the first place. Significant strides have been made to understand the brain using neuro-imagery and fMRI. Which then helped in understanding of diseases like Parkinson, Alzheimer, Epilepsy. But we are nowhere near understanding how the brain arrives at decisions and be “intelligent”.

To understand that further, I started delving into how the brain works. A recent bestseller by Jonah Lehrer “How we decide” goes into explaining our behaviors towards stimuli. (“I need that flashy BMW today”, or “I want this Kate Spade purse”).

Last week I started reading a book called “Incognito” by David Eagleman. Most of the book deals with the fact that 90% of our brain activity is unconscious and involuntary. Even the other 10% conscious brain is largely acting on behalf of the unconscious. The author also raises the point about why AI did not take off.

According to David, the brain’s unconscious “mind” does a lot of number crunching based on data available – past experiences, context etc. But the most significant finding according to the author was the idea of “Team of Rivals”.

This term was based on the famous book by Doris Kearns Goodwin on the decision making skills of Abraham Lincoln. Abe used to encourage opposing points of view on various matters of policy and then he would act on the final decision. Sometime even inviting members of the opposing party in his cabinet.

The book goes on to explain that the brain works in a similar fashion. Various parts of the brain will run algorithms to come to some conclusions. For example, you are planning your money matters. One part of the brain will focus on short term needs you have. Another part of the brain will focus on your long term needs such as retirement, health care. Yet another part will determine what your needs will be in case of an emergency. The brain will always have competing ideas about the problem and they push and pull in multiple directions. And finally there is a CEO in the brain that gets to decide the final outcome taking into account all the outputs from the various algorithms. Just as Abe would do.

(For a simpler demonstration, see what happens in your brain when someone presents a large piece of chocolate cake)

Most efforts on AI have been around finding the best algorithm to get the best results. But human intelligence does not function that way. So for us as product managers building software with intelligence, we need to allow our machines to think in various contexts and then build a “CEO” decision maker that can also intelligently parse through the outputs and arrive at a human like decision.

Perhaps that day is not that far. We can already see examples, though rudimentary, around us.

Also posted on my Blog.