TechRepublic’s Karen Roby talks with an AI expert who believes we need to rethink our approach and focus more on cost benefit tradeoffs and resourcing.
TechRepublic’s Karen Roby talks with an AI expert who believes we need to rethink our approach and focus more on cost benefit tradeoffs and resourcing. The following is an edited transcript of the interview.
Karen Roby: I’m Karen Roby for TechRepublic. We’re talking with Arijit Sengupta, he’s an AI expert with over 20 years of education and experience working in artificial intelligence and even wrote a book called AI is a Waste Of Money. So Arijit, you obviously think we need to re-evaluate our approach to AI… explain!
Arijit Sengupta: The answer is to go back to the fundamentals. See, when we always did business, we considered three things. We have done predictions before AI of course. We also considered cost-benefit tradeoffs, and we considered resourcing. So, business was always a matter of balancing various things, right? What has happened now is that we have gotten so obsessed with the prediction part that we have completely forgotten about the other two parts, cost-benefit tradeoffs, and resourcing.
What Aible is doing is saying, “Well, why don’t we ask a human being what they care about, what matters? What happens if I get a prediction wrong one way? What happens if I get the prediction wrong the other way?” So, “I told you something will happen and it didn’t happen. I told you something wouldn’t happen and it did happen, right? What are the costs? What’s the benefit? Now let’s create an AI that’s optimizing to deliver to that.” Don’t force human beings to learn AI. It is easier to teach AI to speak human than for humans to learn AI. So, we have flipped that equation. We are using AI to understand your reality and then create a second AI to optimize for that reality.
Karen Roby: So how do you think we should reshape our thoughts on this?
Arijit Sengupta: The first thing is don’t think of it as magic. The sooner we stop thinking of AI as, “I just want AI to add something, and a magical answer comes out,” the sooner we’ll stop getting hurt by using a drone. The second part of it always starts with impact. “How will AI actually deliver a business result?” Business people try to maximize benefits, minimize costs, minimize risks. If you make sure you’re starting from what you’re trying to achieve, instead of starting from, “Where’s my data? What can I point the AI at? And what does the AI predict?” You’re going to be in a better place because you’re starting from where you want to end up, which is business impact.
SEE: Artificial intelligence: A business leader’s guide (free PDF) (TechRepublic)
The third part of this is actually very simple. When you think of AI as magic and it’s all about getting the right data and pressing a button, what you do is you go for really big complex projects like the entire company’s going to invest several months to get an AI project done.
AI is going to be more like the browser. All of us, every single one of us, will have the ability to contribute to our own AIs. So you need to focus on many, many small projects. “How do I go find $10,000 of value here? $50,000 of value there? How do I give the tools to many, many people?” So there’s a sewing company, a more than 150-year-old sewing company called Merrow Sewing (Machine) Company. Their CEO is a guy called Charlie Merrow.
So, he was trying to use AI and he applied it to his sales optimization. And he found $3 million of value in two hours. Now $3 million of value in two hours, and by the way, this was written up by Gartner and others, so this is not us telling the story. But what is important there is when you’re a business exec, without a data scientist, you can do that AI yourself and you can do business impact yourself, you feel empowered.
That’s different from saying, “I had a bunch of smart people come in and do this for me and I have no clue how this worked.” The other thing we did is at UC Berkeley, during the AI Summit, we took a bunch of high school kids, history majors and MBAs, and put them up against expert data scientists. And after two hours, we found the high school kids beat every data scientist.
Once we give the data scientists five days, only four out of 11 beat the high school kids. This is where the AI is going. It is more about having that human curiosity where you’re trying to understand the problem, understand the cost-benefit tradeoffs, and how do I explain that to the AI, as opposed to, do I know what a neural net is and do I know what a gradient boosting model is? Because that can be commoditized. Curiosity, human ability to try to understand the situation is much harder to commoditize.
Karen Roby: So Arijit, where do you see that AI will be having the biggest impact, say, five years from now?
Arijit Sengupta: AI is like the Internet or the PC. It will have an impact everywhere. In fact, I think it’s the most transformational technology we have ever seen. But what I think people need to understand is no expert on Earth knows what will be the right AI five years from now. The best AI technology is evolving much faster than the PC chips did. So remember Moore’s law, where the PC chips were evolving so fast that if you wrote a software with one generation of chips, the new generation came in and it just got much better automatically?
Everybody who built chips-specific software died out. Because the next generation was so much better than if your software could just move from chip to chip, it did wonderfully. AI is evolving faster than Moore’s law, but we are still stuck in the mindset of, “Which chip am I going to write a custom program for?” You have to build that organizational knowledge of translating the problems into something an AI can solve and not obsess over the underlying technologies. It doesn’t matter. Because in five years, the winning technologies will be completely different from the winning technologies today.