Futurist Kevin Surace On AI, IoT And Cloud: Solution Providers Have To Embrace The ‘Big Disruption’
Joseph F. Kovar
‘I do not have a crystal ball. … All I can do is help us take a little journey together three years, five years, 10 years into the future, maybe 20, and say, with these technologies coming along, what does the future look like for us? What does it look like for our businesses,’ says futurist and entrepreneur Kevin Surace at the NexGen 2021+ conference.
Building An AI And Neural Net
Building and using a neural net, one of the keys to AI, requires a few steps, Surace said.
The first is to collect millions, many hundreds of billions, of pieces of data. Unfortunately, he said, few companies have clean data. “It’s a mess,” he said. ”They change databases, they change tables, everything’s different. And now there’s an entire industry of doing nothing but going in and cleaning the data. Why? Because the data is nuts. And that’s a real problem. So you have to start with an excellent database, an excellent set of data in order to apply any ground rules to learn something useful from that data.”
The second step is to sort that data, which has likely been input by different people using different models.
Finally, one has to find something useful from that data, Surace said.
What users are looking for in essence are correlations, or looking at past changes to predict others, he said. But that is difficult. An easy example of a correlation would be, when people smoke, they have a higher incidence of cancer. However, he said, there are a lot of spurious correlations that require humans to sort through and say whether they make sense.
He cited as an example a project in which IBM ran natural language processing on doctors‘ notes to look for correlations on cancer but found it didn’t work. For instance, a doctor might have written, “Mary claims she is not smoking.” But half of the doctors mean to say ”Mary is no longer smoking,” while the others have meant to say, ”Mary is smoking, but she tells me she is not smoking.”
“The note was the same, yet the meaning was different,” he said. ”And if the doctor‘s gone, we have no way of knowing what that doctor meant. But for many doctors, that was their hint that ’Mary says, ‘wink, wink,’ no longer smoking,’ meaning ’Mary’s smoking.’ So this is poor data. And if it’s 180 degrees out of phase, I can no longer get reasonable correlations.”
In the end, this is a very hard problem to solve because of the biases of the humans plus the data, Surace said.