Episode 2: Integration of Artificial Intelligence in Education

00:00:00: [Music]

00:00:13: Welcome to the second episode of From Campus to Cyber, the official podcast of the German

00:00:19: University of Digital Science.

00:00:21: I'm your host, Professor Dr. Mike Friedrichs, co-founder of the German UDS.

00:00:26: In the second episode, we thrilled to have Professor Dr. Raul Rojas join us to explore

00:00:32: the integration of artificial intelligence in education and to unpack some of the most

00:00:37: significant trends emerging in this field.

00:00:41: Whether you are an educator, student or simply fascinated by the future of tech in education,

00:00:47: this episode is for you.

00:00:50: Hello Professor Dr. Raul Rojas.

00:00:53: You have had some interesting stages in your academic career.

00:00:56: Please give us a brief overview of your academic highlights.

00:00:59: I did a master's in science, also in mathematics.

00:01:03: And after that I came to get my PhD in Germany.

00:01:07: But since I was studying in mathematics for my bachelor's degree, I became involved with

00:01:12: artificial intelligence, especially algebra systems.

00:01:16: After we wrote the code for a system that could differentiate functions, could integrate,

00:01:23: and that was something very, very special at the time, that was the 70s.

00:01:28: Mr. Rojas, I have to ask something else, because I'm also a soccer fan.

00:01:34: And tell me about your robot soccer player project.

00:01:38: Well, then after I came to Germany to get my PhD, I was working naturally on two things

00:01:44: at the same time.

00:01:46: I was working on my thesis, but I was also working for what's now the Fraunhoferges

00:01:51: cell shaft.

00:01:52: I was working on a project for artificial intelligence computers.

00:01:56: And we were developing machines that could understand logic.

00:02:01: Then I got my PhD.

00:02:03: I did my so-called habilitation in Germany.

00:02:07: And I started teaching at Frey University here in Berlin.

00:02:11: And one of the things that people started thinking about in those years was that it's not enough

00:02:18: if a computer understands logic.

00:02:21: It also has to interact with the world.

00:02:24: So that means the computer has to be a robot in order to interact.

00:02:28: And computer soccer was a kind of laboratory, because you know what you want to do.

00:02:34: You want to have goals in the game.

00:02:38: But also the interaction is complex, because the different robots have to cooperate.

00:02:46: They have to see the ball.

00:02:47: They have to have a strategy for passing the ball.

00:02:51: And so it was an ideal project for developing artificial intelligence and robotics at the

00:02:55: same time.

00:02:56: Okay.

00:02:57: Let's move to the topic of artificial intelligence and education.

00:03:01: Professor Rohos, could you start by giving us a brief overview of the current state of

00:03:06: research in artificial intelligence?

00:03:10: It's difficult to say, because there have been many developments in the last years.

00:03:15: But let's say that the ideal is that you have the computer and the computer is like a good

00:03:22: teacher for you.

00:03:24: But that's something difficult to do, because if you think about, for example, a trainer

00:03:28: in ice skating.

00:03:30: So when the ice skater goes and makes a triple jump or quadruple jump, even, I cannot see

00:03:36: anything that's happening.

00:03:38: But the trainer can detect with his trained eye what's going on, what's wrong in the

00:03:43: jump, and then can correct the skater.

00:03:47: So that would be the ideal for education also, that you go and you start teaching or learning

00:03:52: mathematics.

00:03:53: Then you make a mistake.

00:03:55: And then the computer identifies.

00:03:57: There was a mistake, not just a mistake.

00:04:01: Where is the mistake?

00:04:02: Where does it come from?

00:04:04: Which principle of mathematics or which realm of mathematics you have not understood, and

00:04:09: then the computer goes and trains you in those parts that you have not yet understood so

00:04:14: that you can come back and complete the exercise in a correct manner.

00:04:19: So that would be the idea.

00:04:20: But it's very difficult, as we all know.

00:04:24: We now have systems like chat GPT and other large language model systems that are able

00:04:31: to complete mathematical proofs.

00:04:34: But then there is not yet a system that can do what I was describing, detect what's exactly

00:04:40: the problem in the student, and correct this problem.

00:04:43: And so there's still much to do in this field.

00:04:46: So if we just follow the discussion in the society these days, so you think it's more

00:04:52: than a hype, or isn't it?

00:04:54: It's much more than a hype.

00:04:56: And I'm really surprised.

00:04:57: So people like me who have been working in artificial intelligence for now almost like

00:05:01: 50 years, I'm really surprised by the pace of advance in the field.

00:05:07: So just at the end of the last century, people were talking about how to make computers understand

00:05:15: language and how to make computers understand logic.

00:05:20: Speech recognizers were really bad.

00:05:22: So if you had a speech recognizer for American English, a British subject who couldn't speak

00:05:29: anything that was understandable.

00:05:32: And so the advance that has happened in the last 20 years has been an advance at an exponential

00:05:38: rate.

00:05:39: And so it's really difficult to say what's coming on.

00:05:43: And there is a saying in computer science, artificial intelligence in general, that says

00:05:48: that any forecast that you make for the next 10 years is like science fiction in this field

00:05:56: because so much changes in 10 years.

00:05:59: So Professor Orohas, it's a great pleasure and of course an honor to work with you as

00:06:04: a colleague at the German University of Digital Science in the future.

00:06:08: What is inspiring you to create a master program in artificial intelligence?

00:06:16: Well I think that the German University of Digital Science is a big opportunity for bringing

00:06:23: German style education to the whole world.

00:06:26: So I think that there have been some attempts at developing online courses and online degrees.

00:06:34: But education in Germany has a certain flavor, has I think some advantages of education in

00:06:40: other parts of the world.

00:06:42: And so I think it's very exciting to try this new format, to try to develop courses

00:06:48: according to the German system of education, according to the way that we teach here, but

00:06:53: now for an international audience in Latin America, in Africa, in Asia.

00:06:58: And I hope that we'll also improve our way of working, our courses, that will give us

00:07:04: feedback to know what can we do better and then improve in that way.

00:07:10: Artificial intelligence especially is a hot field and I think that there will be a lot

00:07:14: of interest in these degrees when we start the next fall.

00:07:18: But I think that there is a challenge to describe such the content of a new master degree in

00:07:28: AI.

00:07:29: So can you give us a short overview of what are the key components in learning the outcomes

00:07:34: of this program?

00:07:36: Well, artificial intelligence has three main branches, one is symbolic AI where you work

00:07:44: with logic and language and everything is written in words.

00:07:49: The other is sub-symbolic AI where you work with images and then you have something like

00:07:54: one shot learning where you see an image and then you recognize the image the next time

00:07:59: you see it.

00:08:00: And there is also probabilistic thinking or probabilistic processing in artificial intelligence.

00:08:08: And we need the three types of artificial intelligence.

00:08:11: First symbolic AI because we want to do things like mathematics and then we want to prove

00:08:17: theorems and we want to have mathematical assistance.

00:08:21: We also want to recognize language and be able to argue in using language.

00:08:27: And then we need neural networks and sub-symbolic AI because in the case of neural networks

00:08:32: it has been shown and now we know that they are very powerful for recognizing images and

00:08:38: they can go very fast into the super human domain.

00:08:42: So there are no systems using neural networks that play better goal or chess than a human

00:08:49: that also solve problems even in mathematics faster than a normal human and better than

00:08:57: a normal human.

00:08:59: But on the other side we also need, we need to have safe systems.

00:09:04: We need to have something like probability in the sense that we can prove what we are

00:09:09: stating or we can give probabilities and say, for example, for a medical system we think

00:09:16: that this is this kind of disease with this probability and the doctor who is treating

00:09:23: the person can take a decision.

00:09:26: And so we need these three kinds of approaches.

00:09:27: We have that in the master.

00:09:30: We have the subdivision and we also have programming languages for doing learning in the cloud.

00:09:38: That means that you are big artificial intelligence systems now.

00:09:42: They are not just trained on the laptop or in a desktop.

00:09:46: They are trained using thousands of computers and we can access that cloud and we can teach

00:09:51: how to access that cloud.

00:09:53: So all that is contained in this master.

00:09:55: So if you just establish a new master in such a couple of words, what is different to other

00:10:04: AI study programs because I think you have a very good overview about all the offering

00:10:09: programs worldwide?

00:10:11: Well I think that one of the main advantages of learning with our program is that you get

00:10:17: a degree, a certified degree from a university in Germany and that's not something usual.

00:10:24: So if you look on the internet there are other alternative systems that I'm not going to

00:10:29: mention but you can take the courses and then at the end you have like a certificate but

00:10:35: you don't have really a degree.

00:10:37: Here you will have a degree.

00:10:39: The degree will not be as expensive as some degrees that are offered in the U.S. with

00:10:43: virtual education.

00:10:45: We will be more affordable for the students but the quality I think will be better, will

00:10:51: be higher because our courses are shaped in such a way that they build one after the other.

00:10:57: So there is a good sequencing of courses so that if you take the master at the end you

00:11:03: are an expert in these three branches that I mentioned and of course you then have to

00:11:08: specialize.

00:11:09: You have to say I'm going into speech recognition or I'm going into image recognition whatever

00:11:15: but you have an overview and you have a good foundation for specializing then.

00:11:20: Yeah that was a good description for young people I think just coming over to our university.

00:11:28: If you just talk to our industry partners and other universities some of the main topics

00:11:33: is talking about applications of AI.

00:11:37: What are some of the most exciting applications of AI during these days and which applications

00:11:45: have emerged recently?

00:11:48: Well as I told you forecasting over 10 years is like science fiction but we can say that

00:11:54: in the recent past the really amazing applications that we have seen are for example in biochemistry

00:12:01: predicting the folding of proteins just from the sequence of amino acids then of course

00:12:08: games computer games where now computers are unbeatable for humans.

00:12:15: There have been some also very interesting advances in formal systems in making computers

00:12:21: understand logic and proof theorems.

00:12:23: And also one of the applications that is coming now online are personal assistants.

00:12:32: But personal assistants which are much more powerful than Siri and those that we used

00:12:38: to have where you could ask questions and they would give you an answer after looking

00:12:42: for the answer on the internet.

00:12:44: Now they try to predict your daily work and they try to give you put your notice of things

00:12:53: that can happen during the day.

00:12:54: Like if you had a secretary or an assistant that is watching you over the shoulder and

00:13:00: telling you how to be better at your work, how to anticipate problems that you could

00:13:06: have.

00:13:07: So these kind of personal assistants I think they will become very popular and we have

00:13:12: to foresee what kind of assistants we also want to have and we need to have.

00:13:19: Maybe not everything but there are some interesting problems where we can be helped by the computer.

00:13:26: We talked about offering study program, master program of artificial intelligence.

00:13:32: In what ways is AI integrated into other fields of a typical university?

00:13:38: You mentioned some cases before but if you just have a look to our German University of

00:13:44: Digital Science.

00:13:45: If I look at other universities here in Germany and my experience is that other fields, people

00:13:53: in other fields have become very interested also in artificial intelligence.

00:13:58: But maybe beginning with the handling of data, what's called data science.

00:14:04: So people want to have their data analyzed, organized and they want to recognize patterns

00:14:09: in the data.

00:14:10: So through this label, data science, many methods of artificial intelligence have come

00:14:17: into biology, into psychology, into medicine, into many other fields.

00:14:23: So there is a kind of progression where you start, if you're a biologist, you start by

00:14:28: doing data science.

00:14:30: And then after that maybe you start developing your own recognizance.

00:14:34: For example, for leaves of a tree and then recognize the different types of leaves of

00:14:40: a tree and if a leaf is sick or is not sick.

00:14:44: And then after you do that maybe you recognize.

00:14:46: or you have an idea of how to do a personal assistant.

00:14:50: I was saying before for a biologist,

00:14:52: for helping the biologist to do its work.

00:14:55: And what I was mentioning before,

00:14:56: this folding program for proteins,

00:15:00: is like a magical assistant for biochemists

00:15:03: because they can now just dream a sequence,

00:15:06: put the sequence in the computer and see how it looks,

00:15:09: and see if it can be like a kind of anti-body

00:15:13: or antagonize for some kind of molecule

00:15:16: which is making people sick.

00:15:18: And so you enter into the field of design,

00:15:22: these designing things in other fields,

00:15:25: which until now were not using computers in many cases.

00:15:29: - So we discussed about the opportunities

00:15:31: in our university and also in industry.

00:15:35: If we just have a look at two other society,

00:15:39: so what is the greatest opportunity AI presents

00:15:43: to the society, what do you think?

00:15:45: - Well, I would like AI to help us with climate change.

00:15:50: Because climate change is the biggest challenge

00:15:53: that we have going on into the future.

00:15:56: I think that our challenge today

00:15:58: is not to increase productivity

00:16:01: because that's happening by itself.

00:16:03: Maybe it's even happening so fast.

00:16:06: Our challenge is not to produce more cars

00:16:08: or more refrigerators or more appliances.

00:16:15: That's not the challenge because that's just happening

00:16:17: at a certain rate.

00:16:19: Our challenge is how to reconstruct the world,

00:16:21: how to avoid that we get into chaos

00:16:25: in the next few decades.

00:16:28: And that's something that we have to solve.

00:16:30: And I hope that artificial intelligence can help us

00:16:33: to find solutions for many of these problems

00:16:36: by developing new kinds of energies,

00:16:38: by developing materials which are not so contaminating,

00:16:43: by developing solutions for heat,

00:16:47: for how to purify water.

00:16:49: So many of these technical challenges that we have today

00:16:52: are capturing CO2.

00:16:54: For example, that's something that people have dreamed of,

00:16:57: how to capture CO2 from the atmosphere.

00:16:59: So maybe there is a technical solution going that way.

00:17:03: And I would be very happy if artificial intelligence

00:17:06: could help us to solve this challenge.

00:17:08: - If we just have a look back to the university,

00:17:12: what do you think, is AI changing methods?

00:17:16: Is AI changing campus life?

00:17:19: What are the main transformations you see

00:17:22: at the university?

00:17:24: - Well, it's difficult to answer that question

00:17:28: because the main transformation in the last year was COVID.

00:17:32: So was the fact that everything had to be done online.

00:17:35: Then everybody went to the home office

00:17:38: and we were teaching from our homes to students

00:17:41: working in their homes.

00:17:43: And that made a big impact

00:17:47: because all these tools needed for interaction

00:17:52: from your home with other people

00:17:54: were developed at a very fast rate.

00:17:56: But so thinking now from the point of view of learning,

00:18:01: on the point of view of teaching,

00:18:04: what we need to do is develop these tools further

00:18:10: so that we just not only have this virtual learning,

00:18:14: but we also have a better virtual learning.

00:18:17: A virtual learning where, for example,

00:18:19: you don't notice that you are not in the classroom,

00:18:22: where you have more presence in the classroom.

00:18:25: I have been thinking about a lot about, for example,

00:18:27: extended reality, in which ways extended reality

00:18:31: could help us to make people feel at home

00:18:36: using these virtual reality tools

00:18:39: or virtual communication tools

00:18:40: and how that could improve learning

00:18:42: because there is something magical

00:18:45: about having the professor on the blackboard

00:18:49: because your attention is better

00:18:52: when the person is in the blackboard.

00:18:54: And so my big question is,

00:18:55: how can we reproduce that in virtual learning?

00:18:58: How can we make students feel like they are in the classroom

00:19:02: and maybe extended reality could be a way of learning?

00:19:05: - I'm very happy that we just worked together

00:19:07: in this field too

00:19:08: because we have another master program in virtual reality

00:19:11: and I think it's a favorite project of mine

00:19:15: talking about this meta-verse

00:19:18: and how just works university in a meta-verse.

00:19:21: So yeah, I hope we will have a very interesting

00:19:24: communication in this field.

00:19:27: On the other hand,

00:19:28: we just talked about the opportunities of AI.

00:19:31: There is the other side of the meta-verse.

00:19:34: It's risk and ethical considerations.

00:19:37: So with all these opportunities,

00:19:40: there's also come some risks.

00:19:42: So what do you think are the main and important risks

00:19:46: coming over?

00:19:47: - Well, the main risk for artificial intelligence

00:19:49: is number one is jobs.

00:19:52: So we have seen that in the last years

00:19:54: that productivity has been increasing very fast

00:19:57: and many jobs that were thought could not be automated,

00:20:02: actually automatable.

00:20:05: One example is translators.

00:20:07: So nobody thought that a translator from Germany

00:20:10: to English or from English into German

00:20:12: could be substituted by a machine.

00:20:14: But if you let your documents now run

00:20:17: through an automatic translator,

00:20:19: the quality is so good that you just have to make

00:20:22: a few adjustments and you have a good translation.

00:20:25: The people in Hollywood are striking

00:20:28: because the people who write the scripts for the movies,

00:20:33: they fear that a computer,

00:20:35: you just give a description of what you want

00:20:37: and then the computer writes the whole dialogue

00:20:40: between two persons or more persons.

00:20:42: And so there are many jobs that have become automatable.

00:20:46: And the question is, should we accept that and do nothing

00:20:50: or should we try to put some barriers

00:20:53: to the other some regulation?

00:20:55: And I think that we need regulation.

00:20:57: We need regulation in the case of substitution of jobs.

00:21:01: We need also regulation in the case of

00:21:06: that the computer is the only one taking the decision.

00:21:11: So there is always a human behind the decision

00:21:15: in some sensitive fields like medicine.

00:21:18: Or even in the case of a war,

00:21:22: you don't want an automatic plane

00:21:24: to be bombarding a country.

00:21:27: So you need people who are responsible for that

00:21:29: even if the war is justified.

00:21:32: But you need to have personal responsibility at the end.

00:21:36: And that's one of the big issues right now.

00:21:38: Who is responsible for what when something happens

00:21:41: or when something goes wrong?

00:21:43: For example, with the autonomous cars.

00:21:45: When the autonomous car goes off the road,

00:21:48: who is responsible?

00:21:49: The company who built the car,

00:21:51: the company who delivered the software,

00:21:53: the company who delivered the computer,

00:21:55: the company who delivered the sensors altogether,

00:21:58: who is responsible and who is going to have

00:22:01: to pay for the damages and accept even legal consequences

00:22:06: of this failure.

00:22:12: - How do you integrate this topic

00:22:14: in your AI master program?

00:22:17: - Well, there is a course especially about ethics

00:22:22: of artificial intelligence.

00:22:23: But I think that in all of these modules,

00:22:27: we should have a section.

00:22:28: We should have at least one of the courses every semester

00:22:33: or three semester where we explain the students

00:22:37: what are the consequences of applying

00:22:39: some kind of technology.

00:22:40: So I think this is very important also in Germany.

00:22:45: One of the first things that happened

00:22:48: after the Second World War was that the technical university

00:22:52: which was only a university for engineers,

00:22:56: got a humanities section so that engineers

00:23:01: should think about the society

00:23:05: and should think about the consequences

00:23:07: of what they were doing.

00:23:08: And the same thing is we have to do that

00:23:11: for in the field of AI.

00:23:13: - Are we as a university,

00:23:16: I have to work especially in some research projects.

00:23:21: So can you describe a typical research project?

00:23:23: For example, at the German University of Digital Science

00:23:26: about AI?

00:23:27: - I think most of the research projects

00:23:31: or the interesting research projects

00:23:34: are the ones where you answer the phone.

00:23:36: So you're sitting in your office

00:23:39: and then someone from Siemens or from Tesla

00:23:42: or Volkswagen calls you and tells you,

00:23:45: "We have this big problem.

00:23:46: "Our autonomous car is misbehaving

00:23:50: "or our machine is producing too many wrong parts.

00:23:55: "Can you help us?"

00:23:56: Can you come see what's,

00:23:59: diagnose the problem and build a fix for the problem?

00:24:02: And that's the way I started many of my projects

00:24:04: by answering the phone and getting to know people

00:24:07: in industry.

00:24:08: And the interesting thing is the impact that you have.

00:24:12: It's a little different where you just dream a project

00:24:16: and complete the project and the project goes

00:24:18: into the drawer and nobody looks again

00:24:21: or maybe you do two or three publications

00:24:24: which are cited but they don't have an immediate impact.

00:24:27: I like to have impact.

00:24:29: I like projects where there is an immediate implementation

00:24:33: and you can see the effect of the project

00:24:35: right after you finish.

00:24:38: - I know, Raul, that's impossible

00:24:42: that professors just look in the future

00:24:45: and give some impressions of what happens in the future.

00:24:47: But let's try to talk about the future of AI.

00:24:50: I'm looking forward.

00:24:51: What developments in AI?

00:24:52: Are you most excited about it?

00:24:55: - I'm very excited about the conjunction

00:25:00: of symbolic systems with sub-symbolic systems.

00:25:03: So I said at the beginning that there are these two

00:25:06: separate branches where you have logic and language

00:25:10: and then you have neural networks.

00:25:12: But now they are converging.

00:25:14: And now what we are seeing is that neural networks

00:25:17: are able to work with language or handle language.

00:25:21: They're also able to understand and complete logical proofs.

00:25:25: And one thing that people, where people are very active

00:25:28: nowadays is AI in mathematics.

00:25:32: So the mathematicians can have an assistant

00:25:34: and if you hit a wall in your proof

00:25:37: and you don't know how to continue,

00:25:38: maybe you can ask your assistant and say,

00:25:41: do you have any idea how it can prove this part

00:25:45: of the theorem that they have not been able to prove

00:25:47: or is there any other way of progressing in this direction?

00:25:52: And it actually become very proficient

00:25:55: in this kind of combination of logic with neural networks.

00:26:00: The main contradiction between these two different kinds

00:26:03: of thinking is that logic and symbolic systems

00:26:08: work step by step.

00:26:11: And neural networks work in one shot.

00:26:13: So the data goes through the network and gives you a result.

00:26:16: So how to combine these two things?

00:26:19: And this conjunction is what's producing

00:26:22: very powerful applications nowadays.

00:26:24: - And if you just have a look to the universities

00:26:28: of the future, so looking to the universities

00:26:30: in 30 or 40 years, what do you think

00:26:33: is the main transformation effect

00:26:35: if we just have this look in the future?

00:26:38: - I think that the universities of the future

00:26:40: will be a kind of mixture between learning on the cloud,

00:26:46: learning through the computer and with the computer,

00:26:49: the computer acting as a teacher,

00:26:52: but also the social aspect of going to a meeting place

00:26:56: and getting to know other people

00:26:58: and collaborating with other people.

00:27:00: So I don't think that any university network

00:27:02: can avoid this combination of alternatives

00:27:07: in within the next years.

00:27:11: So even universities which are very expensive right now

00:27:14: because when you send your kid to Princeton

00:27:17: or to MIT or whatever, you have to pay

00:27:21: like 60 or $70,000 a year.

00:27:24: And they want your kid to be there for four years

00:27:27: so that you pay the full tuition.

00:27:29: But I think that there will be a combination

00:27:32: where you go to there because these are beautiful places,

00:27:36: but you don't have to be there the four years.

00:27:38: So a part of your education will be this kind of

00:27:44: virtual learning with very good teachers

00:27:47: which are going to be artificial intelligence driven.

00:27:50: And if you have a very good teacher watching you

00:27:52: or the shoulder helping you to develop,

00:27:55: then you go to the university just to show

00:27:57: how good you are in your field,

00:27:59: how would you have become and met your colleagues there.

00:28:02: But I think all universities are going to be a mixture

00:28:06: and of course there will be purely online universities.

00:28:10: - So my colleague, professor, Dr. Raul Rojas,

00:28:15: it was a great pleasure to talk to you

00:28:17: and I see that the time is going on

00:28:19: and it would be great to talk to you some more time.

00:28:24: And I'm sure that we will have another podcast with you

00:28:28: after starting with the program and other things.

00:28:30: Did I forgot something what is for you very important

00:28:34: about AI?

00:28:36: - No, I think we have covered the different bases

00:28:40: and there is a lot to talk about AI nowadays.

00:28:44: But I think and I hope that many students will be interested

00:28:49: in our offer, in our mastering science and apply AI

00:28:53: and we will see the--

00:28:54: - I as a colleague of you,

00:28:56: I will just follow you in some cases of your models.

00:28:59: And I think for the students,

00:29:01: it would be a great pleasure working with you in this program.

00:29:04: So thank you very much that you take the time

00:29:06: and give us some insights in this field

00:29:09: of artificial intelligence.

00:29:12: Thank you very much.

00:29:12: - Thank you.

00:29:13: (upbeat music)

00:29:17: (upbeat music)

00:29:20: (upbeat music)

00:29:23: (upbeat music)

00:29:27: (upbeat music)

00:29:29: [BLANK_AUDIO]