Frederic Kaplan

18.09.2008 Interview de CADI / Special Issue Knowledge transfers(s)

"Our future will be shaped by teams of engineers and designers who will have learned how to work hand in hand"

Read the special issue

French version

“Our future will be shaped by teams of engineers and designers who will have learned how to work hand in hand.”

Interview with Frédéric Kaplan, Researcher in artificial intelligence.

CADI: Frédéric Kaplan, could you give us more details about your research activity and the new territories you are currently exploring?

F.K.: In parallel with my thesis, I was taken on as a researcher by the SONY Computer Science Laboratory, where my duties consisted in studying how a large number of machines could gradually manage to come up with a new common language by interacting with each other and thus giving birth to a culture of their own. Then I gradually began to focus on the relationship between developmental psychology and artificial intelligence, and that is how I started – thanks to the Aibo – to study the algorithms enabling machines to benefit from long-term learning.
More recently I left SONY in October and moved on to the École Polytechnique Fédérale de Lausanne (EPFL) where I have been overseeing a team focusing on interactive furniture within a large laboratory run by Pierre Dillenbourg. My duties involve re-exploring some of the techniques elaborated over the past few years revolving around objects that, though they are robots, do not look like robots at all: tables, lamps that can sense elements of their surroundings and interact with them in many different ways.

CADI: As part of this interactive furniture initiative, do you work in collaboration with designers?

F.K.: I had already worked with designers during my tenure as a researcher at the SONY Computer Science Lab. Today I work in tight collaboration with several designers and students from design institutions. As part of my research in robotics, I have worked together with the École Cantonale d’Art de Lausanne (ECAL) on the construction of a robot’s playroom. Our goal was to come up with new learning opportunities for our inquisitive robot: a robot’s ability to learn is consistent with its morphology. Most of the objects thought up by designers are adapted to human morphology. With our project we took up the challenge of building a set of objects and infrastructures adapted to a robot’s morphology and learning abilities. This project called upon both scientific and design-oriented research. This first interaction with designers really appealed to me and spurred me to repeat the experience. The kind of imagination I draw on when building machines or inventing algorithms and the type of research led by designers – be it interaction or industrial designers – are truly complementary. I learn a lot when working with them and, at the moment, I am working in collaboration with several design schools including the École Cantonale d’Art de Lausanne and the École de design Nantes Atlantique. One of the students from the latter – whose final thesis work I had been asked to supervise – is currently doing an internship within our team here in Lausanne.

CADI: Yes, he is currently working under your supervision…

F.K.: First I was his advisor in the framework of his project exploring the links between robotics and pedagogy. He is now focusing on the interaction with a lamp prototype we have been developing here at the EPFL. I am always delighted with this type of interaction that always proves to be extremely enriching and stimulating in my view.

CADI: Could you give us more details about the workings of this lamp?

This lamp is a miniature projector with a camera set on its side. Thanks to algorithms we have elaborated, this camera is equipped with digital recognition and can also recognize other objects, if need be. Not only can this lamp produce light, it also displays interactive images. For instance, you can project a button and then press it, move it around and use it in many ways thanks to digital recognition. In the end, this device is not so much a lamp as a computer of another kind in the shape of a lamp. This shape fosters the development of a large number of new applications. Our point is not to recreate a personal computer but rather to think up a congenial interpersonal computer that could easily be used by several people at the same time since users would not be facing a screen but a projected surface. Of course, we could have a finger emulate a mouse but this would not lead us anywhere. There are so many things we can do with our fingers that we would not be able to do with a mouse. Using your fingers is like controlling ten mice at once. How can we take advantage of these new possibilities? This is one of the issues Clément Gault is currently researching.

CADI: What are the main issues and hot topics in the field of robotics and artificial intelligence today?

F.K. : Today we are faced with several main issues in artificial intelligence: the first one is the need to build machines that are able to come to a more accurate understanding of their surroundings and to have an actual “presence”. They need to learn how to analyze the actions of people around them so as to act accordingly. Indeed, we long to see more and more bridges between “their” world and ours emerge. The second challenge consists in making a profit from each machine’s unique and personal story with a view to anticipating its behavior. A moving robot gradually gets to know your environment more in depth and becomes able to develop new behaviors consistent with its discoveries. In the same vein, we also wonder how objects can follow unique “life paths” along which they keep evolving after being purchased. Today, the industrial process is mostly concerned with designing products made to remain the same even after being purchased. One of the great challenges this type of artificial intelligence is taking up is to try and produce changeable objects that gain more and more worth as they interact with their users. Machines can engage in this type of life path in a passive fashion - simply soaking in their environment – or in a more active manner – exploring their environment in a more specific way. This “life after purchase” is going to have more and more impact upon our interaction with objects.

CADI: We know that the advent of intelligent robots can frighten a more or less recalcitrant part of the population. How do you react to this state of things? What is your standing point with regards to the notion of ethics in robotics?

F.K.: I discuss this issue for almost 100 pages in my book Les machines apprivoisées (Vuibert, 2005), so I might have trouble summing it up in a few words. But I can give it a shot… First one must make the difference between learning and non-learning robots. In the end, non-learning robots are standard machines that can handle a wide variety of tasks. In general, their behavior is clearly defined. It is therefore quite easy to identify who is responsible for what in the event of a non-ethical use. However, the situation is more complex and paradoxical as far as learning robots are concerned. Indeed, one cannot entrust learning and ever-developing robots with serious tasks because you must keep in mind that they need to be able to make mistakes in order to improve their skills (indeed there is no learning without making mistakes). Few are the “serious” fields of activity where the use of potentially erring machines would be tolerated. For instance, you cannot make use of a door that would gradually learn how to open, nor of a missile that would gradually learn to reach its target. Even more so that if the device does not work properly, one won’t be able to point the finger at whoever could be held responsible for the defective programming. Therefore the trend leans towards truly open systems that could be applied to fields such as leisure or entertainment in which defects and mistakes do not entail too significant consequences. We could also design devices that would challenge and provoke people without necessarily being useful objects. Since we cannot predict these objects’ future actions, the only way to ensure they will be harmless is to make sure they cannot wound anybody in any way. That is why we dote them with anti-finger trap shapes, we make them light enough not to cause too much damage when they fall. In fact, that kind of ethical rules are not overly exciting, they are nothing but the norms to be complied with by companies manufacturing toys and other similar items. Any type of musings about “laws” to protect ourselves from the actions of robots primarily stem from literature. Isaac Asimov did depict such laws but in his writings they were just an efficient signature literary feature enabling him to produce a whole series of novels. Most of Asimov’s novels deal with how those laws are broken because interpreted in such and such a way. However, I doubt that we will go as far as creating robots able to come up with that type of interpretation and to understand what those laws actually mean. We are talking about literature here, not about the ethical issues at work in the field of technology.

CADI: You mentioned the need to limit the damage potentially caused by robots through more adapted shapes. This is where design steps into your field of activity, right?

F.K.: Yes. Design and its validation. From that perspective, a robot is not too remote from a standard product. When manufacturing an object children are going to play with, you must make sure they will not be able to get injured in any way when interacting with it, just as with any other type of toy. Except that when manufacturing a mobile object, you must anticipate any situation and configuration in which the item is likely to find itself so as to make sure that, no matter what happens, it will not injure anyone.

CADI: Besides the fact that it contributes to people’s safety, how does design prove interesting to people working in the field of robotics and focusing upon the creation of futuristic items?

F.K. : Though it is an absolutely crucial feature of modern robotics, I forgot to mention that artificial intelligence has greatly evolved over the past twenty years and has now become really different from its standard, primary form. As a matter of fact, researchers have become aware of the extreme importance of the body. A robot learns through its body and its body determines what it can learn. The notion of disembodied machines is now totally outdated and we no longer see robots as mere computers trying to grasp the world through the eye of a camera. This cannot be so. To manage to recognize objects, one must be able to interact with them, one must have a body and be able to feel. The body of a robot defines how intelligent it will possibly be. This being so, design cannot be overlooked and must be taken into account because it is crucial in shaping this fundamental element – the body – which will in turn determine what the robot will be able to learn. For instance, imagine you are a ball-shaped robot: if you move about rolling, there are a number of things of our world that you won’t get. Along the same line, if you are a six-legged robot your vision of the world will be shaped by this peculiar morphology. Our human intelligence is based upon our body. Therefore, by inventing new bodies, you invent new forms of intelligence. In that sense, design lies at the heart of the type of artificial intelligence that will emerge in a near future.

CADI: Now let’s focus on Clément Gault – whose project (Pepi, the teaching robot) you supervised. Could you tell us what exactly appealed to you in his initiative?

The main issue addressed in Clément’s project is that only when one understands how they are operated do robots become interesting. If the artificial intelligence algorithms you have created seem to be operated by some kind of magical power, then they won’t be of much interest. Technology becomes way more interesting when you understand how it works. To see this happen, we must give people access to technology. We must implement what it takes to give people access to it. Many Internet users have a vague picture of how the search engine Google works. As they do not understand the workings of the algorithm, they elaborate a mental picture enabling them to use it. The great challenge we are faced with today is to manage to explain to people the basics of artificial intelligence algorithms – how a robot can be programmed, how a robot can learn – because owning a robot is much more fruitful when you understand how it works. Clément aimed at exploring ways to teach to a robot, by working out a new type of programming embodied in the physical world. He has ventured on several quite interesting avenues of reflection. But a lot still remains to be invented in this field. As far as I am concerned, I dream of pedagogical devices that would make pupils discover artificial intelligence as early as sixth grade, and even earlier because some basic aspects of artificial intelligence can be picked up quite easily. If we dote these devices with the right amount of abstraction or design, we could manage to pass this knowledge on and thus to allow people to grasp how these machines work. This way, people could reach a more acute understanding of the workings of these machines and use this experience to cast new light upon human and animal intelligence.

CADI: And this could reconcile people suspicious of artificial intelligence and of the machines doted with this technology…

F.K. : Exactly. In my view there is more to it than just making things accessible to a general public. Researchers eager to share their thoughts and discoveries can write books dealing with robots and their workings – as I did by trying to put the issue as clearly as possible so as to make it accessible to a broad readership. However, learning through reading is not as efficient as learning through experience, by observing how the robot learns, by acting upon its learning process and noticing the consequences. This enables one to get a much more accurate picture of the situation. I have been focusing upon this issue for many years now. When I worked at SONY, I invited a bunch of ten year olds to come to the laboratory and asked them to spend a whole week in the company of a robot. On a daily basis, I asked them to describe how they perceived the robot’s behavior, how they understood its workings and so on… Keeping track of this evolution during a week proved absolutely gripping. In my opinion, this issue is central to the notion of robotics, and we should implement the means necessary to build the tools and pedagogical elements giving people the opportunity to tame those technologies. One of the main issues of the twenty-first century will be the re-appropriation of artificial intelligence technologies by each and everyone of us. And designers will assuredly play a crucial role in this taming process.

CADI: As an onward-looking defender of new technologies who anticipates the world of tomorrow by thinking up robots, could you tell us how you envision the future?

F.K. : Philippe Bultez Adams, Maxence Layet and I have just published a book called Futur 2.0. Comprendre les vingt prochaines années (Fypeditions) which collects texts written by researchers, sociologists and artists wondering about the many aspects of this multifaceted future. As far as my field of expertise is concerned, I am positive that the post-purchase life of objects and media is soon going to invade our modes of consumption and our usages. The history of objects before and after their purchase is going to become more and more obvious, and to become one of the key factors of their alteration. I started to research this very topic by investigating in the field of furniture, buildings and also books. If objects take on a new historical dimension, we might be compelled to redefine the very notion of “object” as we know it.

CADI: Could you explain to us more concretely how a book can actually manage to remember its own history?

F.K.: Well, for instance, if you have a means to keep track of readers and contents, when using an e-book (in which you can download and exchange all types of content) or when using regular books equipped with an identifier coupled with a user identifier, then you can trace back the history of the book – and yours too at the same time. A new dimension then comes to life. I conducted studies on books borrowed from libraries: I observed the notes readers had left on them. Today you have the opportunity to turn a book into a historical artifact by scribbling notes on its pages. Tomorrow, the information will be much more dense: for instance, we will have the possibility to know when most readers quit reading a book before reaching page thirty! Today we know if books are being purchased. Tomorrow we will be able to know if they are actually being read.
This can already be observed in the field of music, for instance. Thanks to the de-materialization of music, each musical media can “know” if it has been listened to and be informed of when and how it has been listened to, and so on. Indeed, the artificial intelligence techniques we just mentioned can be applied to media – a MP3 file in this case – to products and not necessarily exclusively to robots. We could simply use them for historical purposes – it can indeed prove interesting to keep track of this activity and maybe to share its content – but we can also use them to make medias more self-sufficient. Researchers at the Viktoria Institute in Sweden have already ventured on this path by exploring how songs never listened to could slip away from your MP3 player to sneak into that of your neighbor where it would hopefully find a context suiting it better.
My vision of the future also revolves around another axis: the possibility to give objects a more active dimension, to work with new kinds of materials. As I said many robots of tomorrow will probably not look like robots at all. Many objects will be robotized. This fundamental evolution will undoubtedly have a strong impact upon design. At once, designers are going to have to deal with elements of interactive matter, robotic elements. A space for creation will then open up, within which the temporal behavior of objects will act as a key element.

CADI: Speaking of which, what role will designers play with regard to these objects that are able to feel and act, those elements of interactive material?

F.K.: Collaborations between engineers and designers can be conceived in two different ways. In the first case, designers are told: “Thanks to this new technology we’ve just come up with, we now dispose of small-sized robotic elements – for instance. Please make it yours. What can we do with it? In what context? At home? In greater surroundings?” The point of this method is to start from a neutral technology and turn it into new usages; it proves quite efficient because technology-oriented professionals, though eager to see their findings applied, usually lack the necessary imagination to grasp the full potential of their discoveries. Thus the suggestions made by designers are iteratively woven into the work of the engineers; this way, prototypes can be brought to life quite quickly. This is what I call a “bottom up” collaboration: from technology to its embodiment through products.
The other method advocates to start from precise contexts and issues derived from modern-day society and to trickle down towards technology. The point is, for instance, to ask designers to ponder over the library of the future – taking into account the upcoming dematerialization of books, the rise of multimedia content, of interactive furniture and of robotized objects. The feasibility of the ideas submitted by designers is then discussed with engineers, thus giving birth to a virtuous circle promoting creativity. In both cases, the success of a fruitful collaboration lies in the existence of technological or contextual requirements likely to steer and stimulate the creativity of engineers and designers. You also need to have a quick-paced work rhythm, to spend time together and to feed upon enthusiasm. Though not always trouble-free, this type of collaboration is the key to twenty-first century innovation. I am convinced that our future will be shaped by teams of engineers and designers who will have learned how to work hand in hand.

CADI-Speci ...