From learning the alphabet on sand in a Telugu-medium school to becoming one of the pioneers of artificial intelligence, Dr Raj Reddy, in an exclusive interview with TNIE’s Kalyan Tholeti, reflects on his journey, the early days of AI and outlines a future where every student could have an AI companion. Excerpts:
Q) You learnt the alphabet writing on sand and studied in Telugu medium till high school. Could you tell us when you were attracted to mathematics and how you managed the transition from Telugu medium to English when you went to Loyola College after high school?
A) I somehow naturally got attracted to mathematics, in particular geometry, theorems, and so on. When I solved a theorem or proved something, it gave me pleasure that I was able to solve it. I think that's what motivated me. But more importantly, I was naturally curious. Whatever, any book or paper I could get my hands on, I would read it even though I didn’t understand all of it. That was a natural part of my psychology, I guess. And so, by the time I graduated from Kalahasti High School, I was well-trained, I think, but in Telugu medium. Therefore, when I went to Loyola College, where it was all English medium, especially spoken by foreign lecturers like Italian, Irish, and Scottish priests, Jesuit priests, they would have their own accent. I didn't understand what they were saying. But the thing that saved me was they were teaching the syllabus in a textbook. And so, I could actually go read the textbook, even though it was in English. I knew enough English, but not fluent in the way you would be able to speak.
Q) Why I'm asking this question is that most students who come from Telugu medium, even now, are somehow afraid. It is not that they can't understand. The fear creeps in.
A) Yeah, I was not afraid. I was just frustrated. I didn't understand what the hell they were saying. And then I would have to do the homework or hand in an assignment. To figure out what to do with the assignment, I would go read the chapter. The thing that saved me was there would be an example in the text itself of how to kind of work out a solution. It gave me the idea that I need to figure out step-by-step the solution. That's what I did, mostly successfully. I never thought of it internally as a challenge. It was just, I have to do this. And so I just tried to do it.
Q) You studied civil engineering at Guindy Engineering College and from there went to Australia for your master’s. Do you remember any particular turning point?
A) The moment that I thought was a miracle was the fact I got into Guindy Engineering College! Because at that point, the Madras Presidency was split into Tamil Nadu and Andhra Pradesh. And if you came from AP, you could only apply to either Tamil Nadu or Andhra Pradesh Engineering College. I was kind of stymied. I didn't know what to do. So I was mentioning this to my father saying, since I don't have nativity in Tamil Nadu, I can't go to Guindy. He said, “No, we have lands there. I wonder if that is enough.” That was kind of unexpected. I did civil engineering because everybody said that's the only way you'll have a job. There were others doing telecom and so on, which might have been a better thing for me to do at that time. But civil engineering was just fine. I learned a lot. But I never got first class in civil engineering. Again, through an accident of history, it turned out I failed in one of the lab exams. Otherwise I had first class marks, but I failed in this one. That was not deserved, I think. There was a chief engineer or executive engineer or somebody from the services who was giving the hydraulics exam. He had a low opinion of all Reddys! He was a Reddy. He said, “All Reddys are playboys. They never study. They never do anything.” And he came there with a strong opinion about how he was going to teach them to be better engineers. So he failed me. But it didn't matter when I went to Australia and applied to the University of New South Wales. At that time, they never really looked at your marks. They looked at your experience and what you were doing. More importantly, my advisor was the head of the department, Stan Hall, who came back from a sabbatical from England where he was using computers. And as luck would have it, the university, at the same time, bought what is called an English-manufactured computer. English Electric Mark Deuce II. Deuce came from Ace. Ace was the original design that Turing did, where he kind of laid out the instructions and so on. This was the next generation version of it.
Q) But you took civil engineering again in New South Wales. How did you get into computers?
A) I worked with Stan Hall. The first day I saw him, he said, “I'm going to the computer centre. Come with me.” He was writing a program to do structural analysis. As he wrote the program, he explained what he was doing step by step, saying we need to take this structural design thing and multiply it by this. I understood most of what he was doing. After I did this for a week, I was good enough to be able to do the same thing or whatever he needed to do. So he said, “Why don't you go and do this work and then bring back the result tomorrow?” I became his programmer for the year.
Q) Was that the first time you saw a computer?
A) When I went on the first day, that was the first time I saw what you would call an electronic digital computer today. But it was the size of a large room. It had a display. It was making music. I said, “How can it make music?” Apparently, somebody figured out the interference caused by signals was generating sound. So they transformed the noise into some signal which then they connected to this speaker and it came out sounding like music. So, all of that was quite surprising. It was in 1959–60. The computer was very interesting. At that time, the reason it was so big was it was all vacuum tube electronics. Semiconductors were just coming in. The first semiconductor computer that came out was from IBM. It was called IBM 1401 and another equivalent scientific computer called IBM 1620. By that time, I finished my degree and went looking for a job. IBM was happy to see somebody knew something about computers. They hired me.
Q) What was the reaction of your parents? Did they expect you would go abroad and do all this?
A) My parents were from a village. My father never went past school. I think he stopped in the ninth grade or something. My mother probably never went to school. But they were literate. They could read and write. In fact, I learned a lot from them. But they were kind of surprised even from earlier. When I was at the Guindy Engineering College, I joined the air wing of the NCC. And there, you have to learn to fly planes. I was flying planes in 1956 and 57. Once I was on a solo flying mission. I detoured from my flight path to go visit my village. And then go from there to Tirupati, did Pradakshinam around the temple. My parents were amazed, I'm sure, at that time. But when I went back to see them, they didn't say a thing. They behaved as though they expected it from me.
Q) So you caught the computer bug in Australia.
A) Yeah. In the University of New South Wales is when I got the bug. But I got more experience at IBM for three years.
Q) What made you choose to do a PhD at Stanford because most of us look for a job?
A) Yeah. Again, it's all the power of suggestion. My maternal uncle, my mother's brother, was my inspiration in school and college. He said, you know, you should get a PhD. Why? Because my cousin, who was about four years older than me, also went to Guindy and then to IIT Kharagpur. He ended up in Stanford. He was one of the early designers of silicon semiconductors. So he said, “You know, he's gone to Stanford. He's getting his PhD. Wouldn't it be wonderful if you also got to Stanford?” That's how I ended up at Stanford. But also, a more non-foolish answer is, I was reading papers about artificial intelligence when I was working at IBM. And there were three places that were doing research in AI. One of them was Stanford. The other was MIT. And the third was Carnegie Mellon. MIT had Marvin Minsky. Carnegie Mellon had Newell and Simon. And Stanford had John McCarthy. I had to pick one. At that time, there's no department of computer science you could apply to. They came later. I had to apply for a PhD in mathematics. But the committee of computer science faculty reviewed and admitted me to the computer science program. I ended up going there because John McCarthy was there.
Q) So you had a clear idea about what you were going to study.
A) I didn't have a clear idea. I said, wouldn't this be wonderful if I could work on this problem of artificial intelligence? I didn't know what I needed to do or how I was going to get there. But I was reasonably confident I could work on the problem. It never occurred to me that I can't do something. That was the crucial thing, I think. And whatever it is, even now, if you just ask me, can we work on X, I say, yeah, let's do that.
Q) Once you were at Stanford, how did you start?
A) I remember the first day. I came from IBM, where there is a strict dress code. You had to wear a black suit, blue or black, and black tie, and black shoes and socks. You had to look very professional. You cannot smoke or drink during office hours. So I said, okay, I can do that. And so when I came to Stanford, the first day I wore a suit and a tie and black shoes and so on. And I walked into the lounge. And there were all these kids, not kids, they were all older people. Some of them were already working at Stanford. This was in 1963. They were all wearing jeans and t-shirts and whatever. And I stuck out like a sore thumb. I felt embarrassed. And they looked at me strangely. But they didn't laugh at me or anything. They just said, come on in and talked and so on. If I had gone to some other place, not a university, educated people and so on, who knows what it might have been. They might have kind of made a fool of me or something. But they were very polite and friendly. I learned my lesson the next day. I wore a t-shirt and jeans. There was no problem.
Q) And then you got straight into work. I mean, the PhD, what did you begin with? What was the subject that you were in?
A) Yeah. Basically, the great thing was, that was the time semiconductors made it possible to build smaller computers, not room-sized computers. The PDP-1 was four racks or three racks or something, each about six foot tall and three by three. There were three racks of these things. That was the whole computer. And so, for a background, in the US at that time, 1963, John Kennedy was the President. Because of Sputnik becoming the first satellite launched by the Russians, they established this whole agency called DARPA, Advanced Research Projects Agency. And their mission was to do high-risk, high-payoff things that are outside of any normal mission. They were given a huge amount of funding and said, do AI research. There were about three major universities, the only universities with well-known names. And the person that was running DARPA at that time is another scientist. His name is JCR Licklider. He was a professor at MIT. He was the one that talked about the internet. He called it the intergalactic internet or something like that, connecting computers that can do all kinds of things. I think he did not anticipate it becoming what it is today. He was a professor at MIT, joined DARPA and became the head of the Information Processing Office, and he had a fundamental impact. He hired very smart people. We had a stream of scientific leadership, where they continued to invest in people, find the best people and invest in them. 1963 was the first year they started funding AI. And I was one of the first graduate students in the program.
Q: So you started working on speech recognition?
A: No, I didn't. I was taking courses with John McCarthy and others. And then suddenly one day, we had this brand new computer. This one came with an analog-to-digital converter. That means when you speak, the changes in air pressure can be captured in a microphone and that microphone generates a signal which goes to the speaker normally. But in this case, it can go to an A-to-D converter, it goes into the computer. So if you digitise, what I did was, we took a second of computing, or a speech, where you can say, how were you yesterday? Or something like that. That's about a second. If you digitised it, what you got was 10,000 numbers. What do you do with these 10,000 numbers? What McCarthy said was, you figure it out, that's your problem. Having grown up in an Indian system, I was indirectly aware of Sanskrit grammar and all the other things. So if two vowels came together in your speech, you have to insert a consonant. I didn't know at that time, but since then, you know, I learned that there was a great linguist called Panini in Sanskrit, who specified all the rules for juncture.
Q: And you started developing speech recognition?
A: I could use a computer as a recorder. I was trying to recognise what I said. So I said, let me see if I can make it recognise the vowels, Indian vowels and consonants. And so that's how I started the problem of speech recognition, starting with vowels and consonants.
Q: So what was the Hearsay that I read about?.
A: The Hearsay was number two. The work I did at Stanford was recorded in a documentary. You will find it on YouTube. We demonstrated after three or four years of work, you could actually not only say the vowels and sound, actually you can say a whole sentence, like pick up the big block at the bottom right corner, you know, to a computer and a robot. I was in Stanford AI lab, right? I was working on speech, somebody else was working on computer vision, somebody else was working on robotics, and it would hear my command and go and execute the command. For that, it needed to see what is in the bottom right corner, how to find it, and then instruct a robot arm to go pick it up. And we did all of that, I think it was 1968. So it took me five years to get from knowing nothing about speech to get to a stage where we could actually do voice commands. And the result of all of that is in a documentary called Hear! Here!, a pun on what we call homophones. By that time, I had got my PhD and was hired as an assistant professor at Stanford in the same department. I think in the long run, speech will have a major impact.
Q: In phones now, Siri says this, and we interact with it. I think the origins of it go back to your research.
A: Right. Absolutely.
Q: So what is Hearsay?
A: We started with words and spoken words, which are disconnected, like pick up a big block where you pause between each word a little bit, to continuous speech, but small vocabulary, like 1,000 words or less, to go to larger vocabulary system. Then all of those systems were trained in one individual voice. Since the data was collected from me, when it listens, if I'm speaking, it recognizes. If you spoke, it didn't recognise it. So there's a whole set of problems that we had to solve. By 1999, we were demonstrating speaker-independent connected speech recognition. We were actually dictating at that time. There was a company started by one of my students, Jim Baker, called Dragon Systems, which was the first large value company to hit the market.
Q: From 1969 onwards, you have been with Carnegie Mellon University. During all these years, you have been involved in research in AI.
A: Yeah. So basically, what is AI? Different people give it different definitions. In particular, some people say it's going to replace human beings. That was never the case. We never talked about that. We said, can a computer do things that human beings do? Why did we do that? We thought, even as early as 1965, the role of computers and AI is to enhance the mental capabilities of the human being. Whatever you do, if you want to cross the ocean, you build a boat or a submarine. If you want to fly in the air, you build an airplane. Think of AI as a tool that does the same thing for your mental capabilities. Our brain will be able to invent things that are much more powerful than we can do by ourselves.
Q: Some people think soon we may have AGI, and then you may get super intelligence, and all kinds of frightening things..
A: Yeah, of course, we will have AGI. We will have super intelligence. Because I think computers will get even faster. But what does that mean, AGI? There are two definitions. These are my definitions. Most people kind of say AGI is human level intelligence, but that's not something you can measure. One definition of a general intelligence is someone that knows a lot of things about a lot of things. Like Aristotle was supposed to be one of them, polymaths. He knew a lot of stuff. Leonardo da Vinci, Ben Franklin, Jefferson. There were a whole bunch of people we call polymaths who knew a lot of things. Some of them were also autodidacts, meaning they can learn stuff by themselves. They don't need anybody else. The most brilliant example is our own mathematician Ramanujan. He learned all of mathematics by himself. And somebody said it took other serious mathematicians 300 years. And Ramanujan in one lifetime of less than 30 years invented all of that, right? You would call it super intelligence, right? So we have seen examples of polymaths and autodidacts. And we are now getting to a stage where computers may be able to do that. So what are they? AGI, if you agree with this definition, is an intelligence that knows a lot of things about a lot of things. But you need quantitative measures of that. The quantitative measure of AGI is, I don't know if you know, about advanced placement tests for high school students in the US. If you pass it, then when you go to college, you don't have to take that course again. There are 38 such tests that people take. And most common ones are all in computer science, math, physics, chemistry, and certainly mathematics and so on. Then there's English literature. There are other subjects that not many people take, like archaeology or something. And so what they did was they gave all these 38 tests, actual tests, to all the three foundation models that are there today. One from OpenAI, one from Gemini, and one from Claude. So those three did extremely well. They got A plus grade, 95%, on science, math, physics, and chemistry, and so on. And they got a B in English, and literature, and history, and things like that. They got a C in archaeology, and anthropology, and things like that. That's not surprising to me, because they got a C in those subjects, because there's not enough data to learn from. Whereas in mathematics, we have hundreds of years of accumulated knowledge which is all available on the web. So today people are predicting AGI and superintelligence. I think it'll come to pass. But it won't happen tomorrow. It may take 20 years. It may take 50 or 100 years. Maybe not 100 years. Something less than that. When we have it, we will be able to make inventions and discoveries much faster. People say, if there are superintelligences, they'll take over the world. That's a stupid expectation. They won't take it nor will they kill all of us. That, again, doesn't make sense. Why? Because, there are other species on the planet. We don't go around killing everything else around us. So many of these things, what they call existential threats, I think are complete nonsense. But there are things, like if I'm hiring somebody because I need a mathematician or something, in the future I may not need to hire one. Or I may hire one when I need 10 people. And the other nine will be AI mathematicians. And so, the expectation is that human beings will carry in their pocket a polymath autodidact where you can learn everything and do everything because you can learn from a tutor. If you have Einstein in your pocket, every time you have a problem, you can just say, what do I do? And they'll tell you, and you do it. I can't do that now, right? If I go to school, I'm one of 50 in the classroom, and I'm not going to be able to ask questions. That is the biggest challenge we have in the community today. We need to change the entire education system. If you are a child, you can ask, if my AI companion will help me and will do everything that I need to do, why should I go to school? Why should I go to grade 1, 2, 3, 4? I can give you an answer, but before we do that, that is a serious issue. Namely, what is the role of education? What is the role of specific syllabuses we have? I think most of the existing syllabuses are completely useless in the future. We have to throw them out and start afresh with a new set of ideas. Think of me as me and my AI companion.
Q: Does that apply even for a developing country like India?
A: Especially for developing countries. Let's say we are all illiterate. But I have an AI companion speaking to me in my ear. And, hey, we are going to do X, Y, or Z, whatever you want to do. And what you need is that you should be able to speak. You don't have to know how to read. You don't have to know how to write. You don't have to know any mathematics. Your companion knows all of it. You don't need to do any of that. But you are the boss. You are kind of setting the pace. You say, what are we going to do today? And it will come back to you. As for the plan, if you are five years old, you should learn how to read, write, and do arithmetic. But I don't want to do that. Then your AI assistant will come back and say, but if you don't know how to read or write, or if you don't know, when somebody comes and tells you something, you won't be able to tell the truth from the falsehood. So that's where we are. So we have a whole lot of stuff we have to do. Even if you don't have the knowledge to read and write, you can function like a PhD. But you need to have enough education so that you can think and distinguish right from wrong. Deep fakes are going to be all around. And so the question is, when you see one, is it real? So the slogan we have created about education in the future is “we need to teach them learning to learn.” What does that mean?
The issue in the future is that a human being should know how to learn anything they need to learn. That's what we call learning to learn. And you can learn anything you need to learn by yourself as an autodidact, because you don't need other people telling you. You don't need a teacher. And all you have to do is to say, oh, I have to write R. I don't know what R is. How could I write it? Your assistant says, look at the screen and it writes it. So now you do it. Now you have to either do it or refuse to do it. It's kind of stupid. I don't want to do it. So it's a back and forth thing. And education becomes an exercise in motivation of the student. I won't learn what I need to learn unless I'm motivated to learn. So we have to change education right from kindergarten, all the way to PhD.
In order for me to make intelligent decisions, I need to understand that knowledge. And here is a guide that's saying, I'll teach you that. It'll only take you 10 minutes or half an hour. That is what you normally would have taken in a conventional classroom a whole year or a month or something. You can learn in one hour. That's what we call just-in-time learning. You learn every subject when you need to learn it. And if you don't need to learn it, then it's okay.
Q) You were awarded the Turing Award. You won the Legion of Honor, Padma Bhushan. I mean, that list is so long. How did you feel and did you ever imagine that you would go to that level?
A: Never. The first major award I got was Legion of Honor from President Mitterrand in 1984. That was because I was working in France as a chief scientist for a research center which tried to use computer science and AI in helping the human race. And then, you know, I was elected to the National Academy and got the Turing Award at some point. I don't necessarily covet these things. I don't assume that I deserved it. More recently, some wanted to give me another award. I said, I'm too old. I don't need any more awards. Here is a young person, give it to him because he certainly deserves it. And receiving this award will help him in his career.
Q: Recently, India hosted the AI Summit, Global AI Summit. What is your opinion on it?
A: I'm very happy we did it. It raised awareness of AI in the country. And the government is also investing a lot of money, which I think is important to be at the forefront of this technology. And unlike others, we have a huge benefit to be had from it. So, for example, people talk about AGI, right? We talked about it. Yes. For most people, why do you need AGI? In our case, I call it multilingual AGI. The definition is, can you read any book, watch any movie, talk to anyone in any language, anywhere in the world? So if we can, the technology already exists. And if we can actually perfect it and make it easy for people to use, just like we're using Zoom today, that would be fantastic. And that would benefit India a lot more than having some general intelligence taking, you know, anthropology tests in college board exams.
Q: How do you see the future panning out?
A: I think many of the things that are being predicted will happen. What I want to see is if the country can provide a standard AI companion for every student from ages 5 to age 21, that would have a tremendous impact on every one of us. What we need to teach the children, school-going children, is that they should also ask questions rather than just being told what they should learn. And the way you start it is to say, here is the syllabus. Do you know what this is? If you know, answer these questions. If you don't know, listen to this video and then we'll ask. So, it's a dialogue-based learning, conversational learning, rather than sitting in a classroom with a lecture. That's what I am looking forward to in the future.
Q: What is your advice to students?
A: My advice to students is get your parents to buy you a smartphone. Download Google Gemini app. If you don't know how to do it, you should learn that minimum thing, or when you buy it, it'll come already downloaded. Then all you have to do is ask it, saying, how do I download another app. It'll tell you. You can start with things that you need and then you say, now I have this, what do you want me to do? It'll kind of say, if you're learning this stuff, here are some subjects you should learn. But since you don't know how to read and write, I will read them to you. You know, tell me which one sounds interesting. That's how I got into AI.