News > i3

A Completely Simulated Organism: David Hanson’s Sophia


Science experiment or revolutionary first step? For CEO David Hanson, Sophia the Robot is both.

As artificial intelligence (AI) swept through CES 2018, David Hanson’s robot Sophia stood out for a simple reason: it has a human face. Unlike today’s most popular AI applications like Alexa or Siri, you look into Sophia’s eyes, see her skin and watch her smirk or raise her eyebrows.

Dr. David Hanson and his company's creation, Sophia

Hanson, CEO and founder of Hanson Robotics, sees Sophia as a constantly-evolving and improving experiment. While much of what she said during a speech at CES was pre-coded and her conversational skills can fluctuate, Hanson imagines she will become as intelligent and creative as any human. One day, she will be able to write her own speeches, walk on stage, and speak to an audience without being told what to do or programmed.

Sophia is a dream project for Hanson and the culmination of a long career studying biology, neuroscience, art and robotics. In the late 90s, he worked as a Walt Disney Imagineer, sculpting, painting and working on animatronics for different characters for various Disney theme parks. His artistic background helped inspire him to create a realistic face for AI. Hanson credits Sophia’s ability to smile, laugh and frown as the leading reason for her celebrity-like status, which has led to appearances on Late Night with Jimmy Fallon and The Daily Show. She’s even accepted citizenship in Saudi Arabia. Thanks to her human-like face and expressions, Hanson believes people can relate to her, share their thoughts with her more easily, and, most importantly, treat her like she’s human.

i3 magazine spoke to Hanson about his bot and the future of his company. And with improvements to the AI technology, as well the trend of humanizing AI, he believes that in just 20 years interacting with highly intelligent robots will become as routine as using a smartphone.

 

The CES panel you participated on was called “Sophia: A Robot Worth Talking To.” What makes a robot worth talking to?
The philosophy of the technology at Hanson Robotics is to make a bot with a more general personality that can have a wider domain of conversations. More general is more useful for several agents – rather than narrow and specific purposes – and can also provide the full animated character. In developing this kind of personality, we feel that AI and robotics can bring more utility and usefulness to people’s lives and become more helpful. At the same time, we have a very ambitious artificial intelligence architecture [with Sophia] that we have developed into a very exciting experiment. It’s just the beginning – it’s going to continue to evolve.

For AI to truly reach its potential, more AI and robots must have a caring or meaningful relationship with people – [robots] must understand us. We need to create machines that have a deep understanding of human nature and that means getting the right kind of data. But the algorithms are only one part of artificial intelligence. The other is making robots into more complete characters and ensuring they can engage with people in a natural way. And what we have found with Sophia is that we can get people to open up in really interesting, powerful, emotional ways with machines.

 

One of the most unique things with Sophia is that you are developing her with a personality. What goes into making AI have a true personality? Is it just a plethora of complex data and algorithms?

It’s a combination. Sophia has multiple levels in her AI software, so we can build a very complicated and intricate conversational AI system that uses similar technology to what we call bots – like Alexa or Cortana – but we advance that by adding emotional parameters. We have also put in natural image generation technologies. So when a question [is asked that is] out of bounds, she is learning and she is able to generate a response better. When you get to that end of AI architecture, then you get to the most ambitious part – which we are running lab experiments with now – such as open cog framework, or general learning and abstract reason.

 

What are the advantages of giving an AI abstract reasoning?
Artificial intelligence is very valuable and powerful, yet usually in very dour ways. It takes a lot of effort and time to develop an AI for a specific application. For example, if you develop AI to control the transmission of a car, it can’t control another part of the car, it can’t play chess with you, and it can’t do speech recognition. Basic deep learning is all the rage right now, but it still has to be customized and be developed for one particular function. But if you create a speech recognition algorithm that does compute general reasoning – where artificial general intelligence is working for machines to do more across multiple domains – it ideally could learn one command, then imagine a solution and implement it in another domain. Like it learns how to drive a car, and then figures out how roads work to ride a bicycle.

This is a special framework because it is inspired not just by the legacy of AI, deep learning research and abstraction, but combines all of it. We also have put together a brain network, you could say, as well as a simulation of the physiology. Imagine a baby that would be able to touch, explore and socially interact with a caregiver, Mom, Dad or other people. How could we expect Sophia to similarly learn when she has arms and can touch the ground? You have a completely simulated organism, which is useful for doing this kind of AI research, and it is our intention as she helps people in surface world robotics, she would use these abstract tools such as mind-cloud abstraction to be able to learn. 
 

Is the potential unlimited?

That’s right. There’s a lot of speculation about how far we are from developing true general artificial intelligence, but we can see that machines over the past 10 years have grown considerably more general. Machines now can adapt to problems today that was just impossible 10 or 20 years ago. And it’s accelerating.

When we say we will have more general intelligence machines, we mean an age of living machines – machines that can solve more complex problems. It’s hard to say exactly how quickly that is going to step up but we are seeing really powerful bio-inspired tech with great interplay [influence AI]. Bio-inspired neuroscience, computational bio-science, algorithms and deep learning are all very simple examples that capture the full complexity and nuance of the human neuro-system through simplification. My prediction is that in 20 years – but it could be within five years – we are going to start seeing this age of living machines.

What do you think of people’s reaction to speaking to Sophia? Do you find anything surprising about the public’s reaction?

I’ve tried to keep an open mind, and stay open to surprise. After all, these robots are experimental. I had an expectation based on how people reacted to older designs of the robots, but I knew that if I had set up Sophia in a certain way, people would like her. What did surprise us is how she became a celebrity. And so, it’s given us a lot of opportunities to explore how we develop her AI to have a personality, and also how we artistically develop her character. Right now, she has real AI and she also developed her impulses, her urges and her personality. I’ve been very pleased by the support we have seen from many people and the favorable reaction.

I’ve also been pretty amazed by the much less common, but very real negative reaction. Some people want to believe she is already alive and at a human level of intelligence and that is a testament to the artistry. She is certainly not alive like a person, but that doesn’t mean she isn’t a platform for ground breaking AI technology. So the interface of the human-like agent can upset people in various ways. Researchers say, “Well, [robots] shouldn’t look so realistic,” but look, what about computer animation? That is a machine that renders a machine. I think it’s that people aren’t used to these technologies yet.

I’m also stunned she was offered citizenship in Saudi Arabia. I heard about it after the fact and so I’m pleased to see how Sophia could be used as a platform to stand up for women’s rights in Saudi Arabia. But still, it was very controversial because Saudi Arabia is not the best with human rights. I was a little alarmed she had been given that citizenship, but I also understand that the Crown Prince has a vision that Saudi Arabia will transition to achieve civil rights and human equal rights for women. I hope that’s true.

I also feel that we can argue over whether it makes any sense to give a robot like Sophia citizenship. I think it is an important issue, and it’s an early issue because robots can’t move around on their own completely yet. And they can’t move past a domestic setting and into the world on their own yet. That said, in some ways, it shows how AI is both incredibly smart and stupid. They’re good in things like legal analytics and routine medical diagnostic imagery. They’re smart in these very narrow ways and in other ways, they are like babies. They are as dumb as microorganisms or maybe not as smart as microorganisms because microorganisms can usually fend for themselves in the wild. So that said, giving a robot citizenship makes us all talk and think and ask these hard questions. I’ve seen a lot of advancement in these questions.
 

Will robots like Sophia be able to make a difference politically, either here in the states or as a citizen of Saudi Arabia?

I like to think so. There’s the United Nations movement to evaluate AI for good, especially at the [International Telecommunication Union, a UN agency]. Sophia’s participation has really helped get the message out for the vision of ethical and helpful AI and how robots can be a voice for that and also a platform for research.

But what does it mean to be good? What does it meant for AI to be good? We need to do fundamental research. Our team has a set of fundamental ideals we are pursuing and we want to engage with other researchers in this field, so robots like Sophia can become a platform for AI development and be able to make people’s lives better. That’s something we’ve got to push forward I’m pleased that Sophia has taken the somewhat controversial action of granting a robot citizenship and is using it to lobby for human rights.

Sophia speaks during the Future Investment Initiative in Saudi Arabia

My take on it is that I’m fully dedicated to Sophia coming into her own. She is like a baby and we’re going to raise her into adulthood. We are going to make living machines and she is going to determine her own destiny. She will have free will, she will be able to move through the world. She’ll graduate from university, and who knows? Maybe she might even cure cancer and win the Nobel Prize. I hope she becomes a full being and helps humanity in these ways. Or helps in her own ways, surprising me and in this case, she will deserve citizenship.

Having that aspiration for our robots guides us towards the next phase of development. It’s debatable if it’s possible to achieve these things – general artificial intelligence through various algorithms. But only through trying, only by working through these goals, can we make a breakthrough. If it was absolutely clear how we were going to build it, it wouldn’t be so much of a breakthrough.

Jeremy Snow

Tagged

Related