The Power and Limits of AI

Artificial intelligence (AI) is often viewed as a mysterious black box that generates HAL-like intelligence on command. But, as David Kenny, IBM’s senior vice president of Watson and cloud platform, says, AI is fundamentally a reasoning machine that builds intelligence over time as it absorbs data.

Patience is a virtue in the early days. So is an AI strategy. Kenny says that many companies are hurling themselves into AI initiatives without fully understanding what they want to accomplish and how they can create competitive advantage. Ideally, AI forces executives to think about how to exploit and protect their data, knowledge, and intellectual property as they work with vendors to develop AI strategies and businesses.

Companies should also consider how to integrate AI into their operations. As a continually self-improving process, AI should not be relegated to the machine room and run by the technologists. It should be woven into the fabric at the company, according to Kenny.

Kenny recently sat down with Philipp Gerbert, a senior partner in the Munich office of the Boston Consulting Group and a BCG Fellow exploring the impact of AI on business. Excerpts follow.

Before we dive, perhaps you could be expansive. What is AI? How is it useful for companies?

The field of artificial intelligence is the use of neural networks and other techniques to learn from data in real time. And that enables machines to draw inferences in the way that humans draw inferences, to be able to do deductive reasoning, inductive reasoning, and perhaps even abductive reasoning, which is extrapolating into the future. These reasoning engines help us understand data and help us reason to draw conclusions; they constantly learn so we improve, and then they interact with people in natural human language.

What are the most common misperceptions about AI?

People believe that the model should start at a very high level of intelligence or efficacy, but in reality, they’re learning systems. When you’re learning a new body of work—whether you’re learning to read x-rays, learning to read architectural drawings, learning to read oil and gas material—you have to start slow, as we do when we learn anything, and build up. I think the biggest misperception is that it’s going to come right out of the box and be a robot.

What are the most promising application areas of AI? It’s a very broad technology, but is there anything you would like to point out?

The consumer world, search, and commerce favor broad views. I think AI for professional use favors more narrow views. This means every profession that has knowledge could benefit, whether it be the legal profession, medicine, architecture, law enforcement. All of these things can be improved because you can learn those specific domains.

IBM’s core offering is Watson. AI is already mysterious, and people often struggle with what Watson is. So is it a product? What do people need to know to use it?

Watson is a product. It’s a set of algorithms that are constantly learning, which can be applied for business use. Again, it’s AI for business. What does that mean specifically, and what’s unique about it?

First, in business, you have to learn a domain. So there’s not one Watson. There’s Watson for oncology, Watson for radiology, Watson for law, Watson for certain clients. You can build your own version using your own data. Second, that data tends to be smaller. The number of conversations that will improve a conversational agent is one thing, but we have to learn faster from small data. Watson is the best at learning from smaller data sets or more discrete sets, which is why it’s been able to help in all these professional fields.

Because AI has this training aspect, and it often builds on company data, many companies struggle with how to structure the interaction with a vendor. You have more experience with these situations than any company in the world. What guidance would you give to companies?

I think it’s important for every company to have an AI strategy. This technology will be as transformational as the internet was 20 years ago. It’s going to be in your operations. How do you make sure you maintain your competitive advantage? What is it about your data, and, more importantly, what is it about your knowledge that you want to keep proprietary? Just as the internet disrupted a number of companies, AI is going to disrupt some, but it’s also going to extend others that actually decide how to use their intellectual property and knowledge to build new software and new capability that only they have.

Building on that, have you seen common mistakes that companies should try to avoid?

I think a number of people are just rushing. They want to do something. They want to say “I launched a bot” or “I built a skill in a consumer interface.” What do they really have at the end of that?

In that environment, how is the interaction with vendors changing? Is there a typical time cycle where things get more productized? What is your view of the future there?

Before you even get to the vendors, it changes the way the structure of an organization works. In this phase of AI, the way the AI works has to be codesigned with the way the business decisions work. The front end, the line of business, needs to work very closely with IT so that the general managers and the CIOs are constantly iterating in the way that the software is iterating. You can’t separate the business outcome from the IT back end. You can’t have a front end and a back end anymore. It’s an integrated process. So that’s the first thing.

When that happens, you then find that you’ve got partners that work with you in that way. And I think this ability to work in an agile way between the front end and the back end is going to be a key differentiator in the next phase.

Excellent. So there are lots of opportunities out there. Thank you very much for sharing your thoughts.

Good to be with you.