This editorial was originally published on Jul 4, 2016. With Steve on holiday, this is an interesting look back almost a decade into the past at AI technology.
There’s a fascinating piece over at O’Reilly that looks at what we might consider Artificial Intelligence (AI) to be. The discussion looks at Deep Blue, Watson, and AlphaGo, all of which have defeated humans in game competitions where we might expect some intelligence is needed. We could argue that, but certainly, these computing machines have done more to display knowledge than the best humans at certain endeavors.
What is interesting is that each of these machines, while very competent in its area, is specialized. AlphaGo can’t play chess, nor can DeepBlue play Go. Each has been tuned to a specialized area, and also trained to excel in that area. This isn’t fundamentally different than humans that train and specialize themselves, though certainly we find humans have more capabilities in a general sense (for now) than machines.
As we look to grow intelligence, however, there is one thing that’s commonly needed in both artificial or machine intelligence and human intelligence: data. Whether a human is training themselves to solve a particular problem, compete in a game, or even excel in a sport, they need lots of data. We gather this with our senses as well as by examining what others have one, contemplating actions, trying out different actions, ideas, or concepts, and then adjusting to improve.
This is what researchers are also trying to do with gaming machines, with self-driving cars, and even with bots. That last item is interesting to me, as I haven’t paid much attention to bots. A long conversation with another SQL professional got me interested in, and intrigued by, the idea of software robots that might handle various complex tasks better than the FAQ method that so many applications and websites use. I wasn’t sure these would be useful, but I have found the Slackbot to be more helpful than the help or searches for some tasks.
There’s work to be done, and I know the Slackbot (and other machine intelligence software) needs to be trained better. This requires data. Lots of data, and possibly lots of hand holding from a human. For many areas, such as relatively low-level customer support or problem solving, I wonder if a bot could be trained to work better than the simple decision tree algorithms like those found in the Windows Troubleshooter.
There are various ways we might grow this software to help us, and make no mistake, we will need to grow it. Plenty of businesses are becoming excited about machine learning, the R language or Python, software bots, and more. In all the cases of implementing these systems, the one demand that will impact many of us is the need for lots of data. Data that’s organized, that is relevant, that we can use to separate out successes from failures, and evaluate our particular problem better. We will need to group data into knowledge and then feed it into software.
I think this is a bit different than how most of us have used data over the years. We’ve often collected, manipulated, aggregated, summarized, and spit data back out to (ultimately) some human that can make a decision. Most of us haven’t worked with sending data to a machine intelligence and somehow then helping it to understand how to respond on make a decision.
My suspicion is there will be lots of work for us in the next decade in helping machines to use data and understand it, maybe even to use them to help us gather, organize, clean, and manipulate data better ourselves. It’s an exciting time to be a data professional, and I’m sure some of you will work on a few very exciting projects in the future.
Steve Jones
Voice of the DBA 