Ismael Faro is vice president of Quantum and AI in IBM’s Research division. After creating seven start-ups and arriving in Silicon Valley from his native Galicia, he moved to IBM. Another Spaniard, Darío Gil, now undersecretary of Science in the Trump administration, noticed him ten years ago to bring him to the company. There they created some of the first AI-based debate machines, the parents of ChatGPT.
Faro, 50, born in Pontevedra, traces his career here since he was unable to attend COU due to financial problems at home and ended up teaching his vocational teachers because he knew better. Those were years when YouTube tutorials were on newsstands: I bought a computer magazine and, if there was someone else who did it, I asked if they knew about programming. Thus he began to study assembly, a prehistoric programming language popular in the 1980s: “I am one of those lucky ones who, from an early age, already knows that he wants something, since I was 13 I was already dedicating myself to computers”, he says. Now he has created a scholarship program in Galicia so that interested young people like him have the greatest number of opportunities at their fingertips. A few days ago he passed through Madrid to give several conferences, one of which was at IE University, where this interview was conducted.
Ask. His father was watching him when they gave him his first computer.
Answer. My parents made a huge effort and bought me a PC, an 8088, equipped with two speeds, a turbo button and two floppy drives. I started breaking everything because I wanted to know what was inside. And my father told me: “The guarantee”. I said, “I don’t know what that means, but it’s important.” I will remember all my life.
Q. He taught his teachers.
R. In FP the teachers looked at me so weird. And I liked that in the last year they asked me if I taught students and teachers. It’s my passion. All the hobbies I have are computer related: computer music, computer animation. I have done many things in my life, but the computer has been my tool. I feel very comfortable with computers.
Q. They will have called you strange guy very lightly.
R. Yes, people don’t understand and say “this is a geek”. I look at it from the positive side. The only bad thing was when sometimes they used you as a little bunny that they showed as if to say, ‘Have you seen the magic it does?’ It was more for show than knowledge and I felt sorry for him. Not understanding it, they say it’s magic. And they want the show more than the learning. They don’t ask you ‘hey, how do you do it?’ They just want “the magic”. Then they don’t know what to do and ask me several times how to solve it.
Q. He says he has failed many times.
R. In 20 years I have done seven start-ups. What is success? For me it means learning and doing new things. For many people, success means making a lot of money and not going bankrupt. For me there is no such thing as failure, you close one door to open another. It’s something evolutionary. Every failure is like a master’s degree.
Q. He once said that he was determined to make machines think. How is it going?
R. Around 1993 or 1994 I was obsessed with the idea of making the computer talk. We need to understand what it means to make machines think. For me it is that the interaction distance between man and machine is shorter. Since I was 13 there is one thing that frustrates me a lot: I still use my fingers and look at a screen to program. While a user can sift and it has other tools that I as a programmer helped create, I am not able to interact with the technology in any other way. It’s all very sequential, I have to write in a linear way. The computer can do this in parallel. But when I gather information, whether I have to read it or listen to it, it will be in sequence. This is why I love artificial intelligence. Because I try to explore new systems of interaction with machines. I have sets set up with a camera on top that sees what I draw while I’m talking to the machine and the machine is projecting me. The interaction is different. It’s richer, there’s more context.
Q. The machine sees the drawings.
R. You can draw over what I draw. And you can talk to me. There are already two ways of interaction, not one. This flexibility is what I want to continue exploring.
Q. But is this thinking?
R. When I say think, I mean that machines are an expression of us. All digitally accessible information has been compressed to generate strong relationships between the most common concepts, while loading the less common ones. In this way you lose part of the information to condense it in one point and then extract it. When I say machines think it’s to be more proactive. Now they are usually passive. Current AI is very passive. If you log in to ChatGPT, what happens? What happens if you enter a ChatGPT and watch it?
Q. At first nothing. But then he suggests questions to maintain the interaction.
R. It’s already a step. The interesting thing is that with more context the interaction is smoother. And may it help you move forward. Now there are techniques in artificial intelligence like deep research where you search for something and the parallel machine opens up a bunch of browsers, reads them all in parallel, extracts the information, digests it and then structures it and presents it to you. And he asks you. Since you already have all the context, the conversation goes more smoothly. You can see a lot in the way young people interact with these chatbotsas if they were operating systems. They maintain context, they remember things, they are small steps that give the sensation of thinking, but not of thinking. When I talk about machines that think, it means that they help you collaborate, that they are like a companion.
Q. ChatGPT was released in November 2022, but years before, with the Debater project, IBM already had a debating machine: “The talk show robot”, we call it in EL PAÍS. When ChatGPT appeared, how did you experience it?
R. First of all it was obvious. In fact, we tried with a friend to do something similar with GPT-2. The big breakthrough was that they focused on interacting with humans, making it seem like a person was responding to you. The discussion was very focused on generating information with context, to discuss. It must also be understood that this leap also had a very high computational cost and involved the use of humans for training.
Q. He said that AI will be bigger than the Internet and that it will be just another tool. Isn’t it a bit the opposite?
R. I see the Internet as a concept inside a window. When you use the Internet, you use what you can see through that window. If you jumped in, you’d see more. But we have the limitation of that window and small eyes and fingers. The Internet is huge. You need tools. Although the Internet is there, you don’t have all of the Internet, you have to access parts of it. AI will be bigger than the internet because of the way people already interact. Eventually it will be used more because it reduces the complexity of use. If you search for information with a search engine, you need to open the links, creating your summary. Now ChatGPT does it. And the result will be similar. The time you haven’t invested in the second tool, you can use to do other things. It’s much bigger than using the Internet.
Q. And at the same time it’s just another tool.
R. It’s another way to access information. With the Internet you explore. In this you have a conversation with that information, but you don’t have to read all the information. All the tools we have innovated with will optimize our time.
Q. How is artificial intelligence combined with quantum computing at IBM?
R. If you’re trying to tackle a very complex problem, current computing can’t cover it: for example, simulating a caffeine molecule. What is done are approximations with artificial intelligence up to a certain point. From that moment on, the quantum can extend it. The work we do in the field of quantum computing is to improve the tools so that we can achieve this as soon as possible, but this is a task of both the hardware part and the software and algorithm.
Q. When will a quantum computer be operational?
R. We can argue a lot about what operational means: what use case does it solve? For example, today’s computers are already operational for those involved in quantum computing. For the rest I work to see it. There is no deadline, because when something breaks on the market, what was 10 years ends up becoming two. It is also a question of investments. At IBM we have been working on quantum computing for 50 years. Imagine all the investment of people and time to get here.
