IBM and AI: Researcher talks current and future projects

AI will replace human workers. Robots will take our jobs. These headlines have been ubiquitous. There's a real fear in today's society that automated workers will replace human workers, and it's a fear that's well-founded.

Yet, if you ask IBM and AI researcher Costas Bekas, he'll mention that there's another view out there, too, one that sees AI not as a replacement for humans, but instead as a tool for humans to enhance their own skills and help them make new discoveries in their fields. It's an idea that AI is more of an augmenter rather than a replacer.

In this Q&A, Bekas, manager of the Foundations of Cognitive Computing group at IBM Research -- Zurich, speaks about some of the narrow AI projects IBM is now working on outside of its publicized Watson and Project Debater undertakings.

Editor's note: This interview has been edited for length and clarity.

As an IBM and AI researcher, can you discuss narrow AI and some of the projects IBM is working on?

Costas Bekas: Narrow AI means that we use artificial intelligence for specific enough domains that can now, these days, make a difference. Whereas we see broad AI is still a few decades away, narrow AI is already upon us.

Whereas we see broad AI is still a few decades away, narrow AI is already upon us.

For instance, we just released a system, IBM RXN, that is able to predict the output of complex chemical reactions. With it, you can say 'Here are two or three substances, molecules, and I'm going to let them react and see the output.' And it's accurate.

Now, without AI, this is complex, meaning that today, a human would require 25 to 30 years of experience to predict this accurately 80, 85, 90% [of the time]. This system is freely available for all chemists out there, and now there isn't a minute that goes by without two or three people testing it."

There's another IBM and AI project, the IBM Corpus Conversion Service, that looks to be able to automatically read and sort PDF documents. Can you talk a little bit about that?

Costas Bekas

Bekas: Well, we're working massively on systems to ingest all of the world's medical information and literature. There is a humongous amount of documents that hold medical information. Especially in the business part of a hospital, we've found that there is a humongous amount of highly fractured information. Easily sorting that is extremely important for healthcare and, really, for all businesses.

We have been working on massively scalable technologies for the ingestion of highly complex documents, like PDFs. This month, we showed our Corpus Conversion System, which can automatically ingest hundreds of thousands of pages of PDF information.

This notion of making discoverable useful documents is one of the main things we do at IBM. We apply that in the medical domain, we apply that in the domain of finance, insurance, we apply that in the domain of making chemicals and materials.

In the media, there is often talk about how AI is threatening human jobs. As someone who works with IBM and AI, what do you make of these concerns?

Bekas: Let me go back to IBM RXN and give you some examples of comments we saw on Reddit about it. Here's the thing -- people were saying it looks like parts of chemistry could be really automated with this type of technology. But, wow, how many other avenues are we able to explore that we couldn't before because our brains and our time were focused on these other things? So they immediately saw this as a game-changer because it opens up other avenues for the human mind.

This is exactly what we've been seeing now -- you solve the problem, you scale the problem and this, of course, automates some avenues, but it opens up many others. Advice that we give is to invest in strong education, invest in a clear-thinking mind and creativity, and have in mind that AI will replace repetitive parts of jobs that are not very fulfilling.

Think of AI like a calculator. Go to any computer science or mathematics department at a university. Do they have courses that teach you how to compute standard things like the sign or the cosine of a function or the exponential of a number? No. They are taken for granted so much [because] they are embedded in our computers.

It's a little like cars. In the past, the average car owner knew a little bit about cars, maybe enough to correct or fix the car if it was broken. Nobody does that now because now so many parts are automated.

We strive for security, we strive for low consumption, we strive for eliminating accidents, so this kind of automation has opened up the door to the really important things, which is how to keep [someone] from getting killed after an accident or how to avoid getting into an accident altogether."