Hiroshi Ishiguro joins 'AI: More than Human' exhibition co-curator Maholo Uchida in conversation about robots and the journey towards understanding human intelligence. Ishiguro was an advisor for the Barbican exhibition, and is known for his pioneering work with robots. He is Professor of Department of Systems Innovation in the Graduate School of Engineering Science at Osaka University and Director of Hiroshi Ishiguro Laboratories at the Advanced Telecommunications Research Institute. His research interests include distributed sensor systems, interactive robotics, and android science.
Ishiguro with Geminoid HI-4 (2013) by Hiroshi Ishiguro Laboratory, Osaka University (2013/2013) by Hiroshi IshiguroBarbican Centre
Maholo Uchida: I would like to start at the beginning: why did you begin making humanoid robots?
Hiroshi Ishiguro: I began my career in computer science before moving into artificial intelligence and robotics. My motivation for making humanoid robots stems from an interest in understanding what makes us human, and what it means to be human. When building robots and studying how humans interact with them, I noticed how a robot’s appearance and movements significantly influences the outcomes of these interactions. Hereby began my research into human-like robots or androids.
MU: You’ve been making humanoid robots for the last twenty years. How have your projects developed during this time?
HI: I began my research into robotics thirty years ago and my initial goal was to create a vision-guided robot or a kind of automatic driving system. During this time, I came to understand how robots need to have two types of vision systems: panoramic vision for understanding location and surroundings, and active vision for understanding and recognising objects.
I am now interested in developing human-like appearances and behaviours in robots. I am particularly interested in eye movements and head movements, and how these are used to navigate environments. Therefore, whilst my initial motivation was focused on limited human functions, I have since extended my interests to study complex functions relating to intelligence and human-robot interactions.
"My initial motivation was focused on limited human functions, I have since extended my interests to study complex functions relating to intelligence and human-robot interactions."
Installation photo from the Barbican's AI: More than Human exhibition, featuring Hiroshi Ishiguro's Alter (2019/2019) by Barbican CentreBarbican Centre
MU: Much of your research focuses on the physical appearance of robots. Could you explain a bit about why appearance is important to consider when creating a robot?
HI: The significance and implications of a robot’s appearance is one of my main research topics and I am currently investigating how robots with differing appearances perform in a range of situations. I am creating many types of androids at the moment – female androids, male androids, elderly androids, young androids, and androids that have a neutral appearance or are mechanical-looking.
One of the significant findings from these studies has been around the importance of activating the human imagination. If our imaginations are activated when interacting with a robot, we become more accepting of it and therefore our interactions become more meaningful. My solution to this finding was the Telenoid. The Telenoid looks in many ways like a human but its age and gender are undefined, thereby encouraging people to imagine or fill in the gaps relating to its identity.
Furthermore, cuteness is very important in robotics and I think everything should be cute in some sense. If an android that has a humanlike appearance performs non-human-like behaviours, the android can suddenly become quite uncanny and therefore no longer cute.
MU: You have famously made a robot in the image of yourself, an Ishiguro humanoid. Can you tell me about your experience interacting with your humanoid?
HI: When I created my humanoid (named the Geminoid), everyone commented on how similar its appearance and behaviours were to my own, but I disagreed. This led me to contemplate on how I don’t really know my own appearance, voice, or movements. Of course, we are using my humanoid for other psychological experiments and scientific studies and we are discovering many interesting things, but that was the most important finding for me – that we don’t really recognise ourselves.
"If our imaginations are activated when interacting with a robot, we become more accepting of it and therefore our interactions become more meaningful."
ERIKA by Hiroshi Ishiguro Laboratory, ATR & Osaka University (2016/2016) by Hiroshi IshiguroBarbican Centre
MU: Moving beyond teleoperations, androids can now be programmed using AI. Using programming rather than teleoperation makes these humanoid robots appear more and more human-like. This transition can be seen in your robot Erika. Could you tell me a bit more about this project?
HI: Yes, I began the project with a teleoperated android, but we’re now focusing on autonomous androids with intentions and desires. This is a very important challenge because without intentions and desires the robot cannot have consciousness or emotions and is therefore unable to become more human-like. This is not a learning function, however; it is just a complicated computer program. Of course, we need to have a sophisticated computer program to simulate the intentions and desires of a human. And no doubt, the robot is going to become more and more human-like in the future.
MU: Do you plan to use machine learning technologies in your robots?
HI: Yes, of course, especially for voice recognition and visual recognition where deep learning is needed. If there is a pre-existing neural network for voice and facial recognition, we will use it. To clarify, a pre-existing neural network is an already trained neural network that is incorporated into the robot. Our purpose is not to simulate the developmental process of a person but to make a copy of that person. Therefore, we are using machine learning techniques for voice, pattern, and visual recognitions, but almost all researchers are using the result of a machine that has already learnt. Learning means the training of a neural network, but we are using a previously trained neural network. Therefore, we are not studying machine learning, we are just using the results.
MU: How do you envision the future for machine learning and robotics?
HI: At the moment I am not concerned with the developmental process. Everyone is expecting robots to learn on their own and become something more than human, but this process is difficult. Whilst in the future we may advance these developmental or learnable functions, I am not sure exactly how this will work. The human developmental process needs a growing hardware, a growing body, because without the growing body, the human cannot learn. Both the hardware and software therefore need to be synchronised.
Obviously, our body is limited in comparison to a computer network. A computer network can quickly gather huge amounts of data, but this process is not very human-like. If a robot was able to teach itself, it would take a long time to build up the necessary data through experiences in the world. Therefore, installing a real learning function in a robot is very difficult.
"The robot is going to become more and more human-like in the future."
Co(AI)xistence with Mirai Moriyama & Alter (developed by Ishiguro lab, Osaka University and Ikegami Lab, Tokyo University) (2017/2017) by Justine EmardBarbican Centre
MU: What kind of robot do you hope to create in the future?
HI: My goal is to install intentions and desires in a robot. In the future we are going to use more completed autonomous systems. These systems will be complicated, beyond the understanding of most people, so it is therefore essential the robots have their own desire to interact with humans. I want to develop machines that are programmed with this intention to support human beings.
MU: How do you think humanoid robots will be integrated in future societies? Do you think their integration will change the way we live?
HI: To help illustrate my point, I want to consider vending machines in Japan, which are a kind of robot. We have thousands of vending machines in Japan and everyone trusts them not to make mistakes. But imagine if they do begin to make mistakes, sometimes giving too much change, or sometimes too little. Or suppose the vending machines misunderstand the difference between Coco Cola and Pepsi Cola. In this circumstance people would likely give a human likeness to the vending machines and the interaction between vending machines and humans would begin to shift.
As we develop more autonomous robots with intentions and desires, we may witness changes in our society. At some point we are going to have societies where humans and machines co-exist in a symbiotic way and new social norms will need to be formed to maintain stability.
"At some point we are going to have societies where humans and machines co-exist in a symbiotic way and new social norms will need to be formed to maintain stability."
Installation photo from the Barbican's AI: More than Human exhibition, featuring Hiroshi Ishiguro's Alter (2019/2019) by Barbican CentreBarbican Centre
MU: Do you think humans and technology are more closely interlaced in Japan?
HI: Robotics is a very important industry in Japan. I have a hypothesis for why Japan is so successful in robotics and why it is comfortable accepting human-like robots or humanoids. I’ve named my hypothesis the Island Hypothesis. Japan is a small island with a homogenous society that has been ruled by the same royal family for almost 2000 years. This means we are like one big family and rather than forming a hierarchal society, we prefer to help each other find a role within it. Therefore, we aren’t concerned with distinguishing between robots and humans; we co-exist with one another. I believe this is a Japanese mindset, and this is what I am naming the Island Hypothesis.
In many ways, through medical advancements in prosthetics and transplants, we have seen in recent times how the body is no longer a key requirement for defining being human. Therefore, I believe that in the near future, in Japan particularly, the boundary between human and robot will disappear.
MU: You have previously described AI as ‘artificial humanity’. Is it therefore fair to conclude that by working with AI technology one is simultaneously working on humanity?
HI: My basic motivation is to study humanity. I am creating robots both to develop technology and to study humanity and I don’t think we don’t need to distinguish between these two things. The robot is a kind of mirror that reflects humanity and by creating intelligent robots we can open up new opportunities to contemplate what it means to be human.
"The robot is a kind of mirror that reflects humanity and by creating intelligent robots we can open up new opportunities to contemplate what it means to be human."
This interview was originally published in the 'AI: More than Human' exhibition catalogue.
AI: More Than Human is a major exhibition exploring creative and scientific developments in AI, demonstrating its potential to revolutionise our lives. The exhibition takes place at the Barbican Centre, London from 16 May—26 Aug 2019.
Part of Life Rewired, our 2019 season exploring what it means to be human when technology is changing everything.
