Living with Super-Intelligent Robots Are we cave people?
Nobody is working harder to tear down the boundary between humans and computers than Ben Goertzel—the chief scientist behind Sophia, one of the world’s most advanced humanoid robots. Thanks to her artificial intelligence, Sophia can see and recognise people, understand and talk to humans and display more than 50 facial expressions. So Goertzel is ideally placed to assess how human-machine coexistence is influencing our everyday lives, and what its future will bring.
-
-
-
-
Mr. Goertzel, you haven’t seen the humanoid robot Sophia for several days. How badly do you miss her?
B
G
It’s true I am travelling and Sophia is back home in Hong Kong. But I am so intrigued by her that I keep working on improving her intelligence. So she never feels too far away.
-
-
-
-
What intrigues you about her?
B
G
There are at least three perspectives. From that of a roboticist, it is the question of how we get her to walk, talk and smile…
-
-
-
-
…Which are a robot’s first steps towards becoming a humanoid.
B
G
Yes, but even more complicated is the second perspective of artificial intelligence, or AI, that we are trying to put into Sophia.
-
-
-
-
And how are you doing that?
B
G
We are trying to combine several capabilities of narrow AI into something broader.
For scientists like Goertzel, the label of narrow AI is critical, as it describes the lowest of three AI levels. This first level involves algorithms that recognise patterns or learn and teach themselves, but only perform limited tasks, such as recognising faces or playing chess. The second level, which has not been reached, is Artificial General Intelligence (AGI), or human-level AI, where machines can plan, imagine and experience consciousness. This is Goertzel’s current goal for Sophia. The third level, Artificial Super Intelligence (ASI), is superhuman. Here machines reach the point of singularity, where they can improve themselves and thereby drive runaway technological growth into a completely uncertain future.
-
-
-
-
How narrow or broad are Sophia’s skills?
B
G
She combines several algorithms. One for image recognition, which detects a human face; another is a transcription algorithm that turns a conversation into text, which Sophia analyses to produce an appropriate response that she can then communicate through a speaker.
-
-
-
-
Some people are so keen to chat to Sophia that they pay $ 100,000 for that the privilege.
B
G
Yes, some companies book her for speeches or other occasions at that price. She has also been on TV shows, including NBC’s Tonight Show with Jimmy Fallon. That helps us to raise funds and to familiarise the public with humanoid robots.
-
-
-
-
What can Sophia do beyond entertainment?
B
G
A lot. We get hundreds of inquiries from people who want to buy armies of beautiful robots. Some of those requests come from car dealers, who want robots to help attract people into their garages and sell cars. Banks want them to greet customers and direct them to different counters. And a restaurant chain in China wants to buy robot waiters, which might initially just take orders.
-
-
-
-
But Sophia also does non-commercial jobs.
B
G
Yes, she has worked as a meditation assistant. Of course she is nowhere close to human intelligence yet, but we do trials on the meditation and give people questionnaires afterwards to see what impact it has on them.
-
-
-
-
And what is their reaction?
B
G
On the whole, it’s positive. Some even say to us, Wow, I could never enter into a deep trance state before,—but with the robot I got into it really deeply, because when Sophia looks into my eyes I feel seen by the robot, but I know she isn’t judging me.
-
-
-
-
So Sophia isn’t judgmental?
B
G
Not yet. But that brings us to the third perspective: human-robot coexistence. Eventually, we want an AI that figures out the crux of what humans want—what it means to be human.
-
-
-
-
How do you get there?
B
G
It’s tricky, since there is nothing like a codified list of the top 50 human values. Sophia would have to absorb these values by interacting with humans socially and emotionally.
-
-
-
-
So you need her to mingle more with humans?
B
G
A lot more. To do that, we are hoping to build one million simpler sisters of Sophia that could collect lots of data from their interactions. What Google did for search engines, we could do for robots. And we have a good chance of mass-manufacturing a simple Sophia by combining David Hanson’s technology for expressive robot faces, our access to South Chinese manufacturing close to our office in Hong Kong, and our SingularityNET and OpenCog AI software.
Of the numerous projects that Goertzel is involved in, these are the three most important. First, Hanson Robotics, founded by David Hanson, builds Sophia. Second, SingularityNET was founded by Goertzel as the world’s first public AI platform to connect international AI researchers with each other and with businesses that can request their service. Finally, OpenCog is an open-source software project, where scientists cooperate to create human-level intelligence.
-
-
-
-
So would Sophia’s sisters help you create the next data monopolist?
B
G
No, because through Singularity NET and OpenCog we want to work towards the opposite: to share our data and create a decentralised AI infrastructure. In that way, it won’t be a military agency or mega-companies that govern AI.
With SingularityNET, you successfully raised $ 36 million—but that is still tiny compared to the budget of the US Defense Agency, Darpa, or companies like Google. How can your decentralised model of data-sharing compete with such behemoths, which keep their data to themselves?
B
G
My hope is for a development similar to that of the music industry. Warner, Sony and Tower Records also had a lot more money than the folks who created Napster. But peer-to-peer music sharing became the popular way to distribute music, and it overpowered these hegemonies. Similarly, Microsoft had a lot more money than Linus Torvalds, who developed Linux. But Linux now operates the majority of phones and servers on the internet.
-
-
-
-
So you’re hoping for a similar decentralised revolution with your AI projects?
B
G
Exactly. You need something powerful enough, which one person uses and gets all his friends to use. Then you have reinforcing cycles and network effects that can overcome large institutions. Meanwhile, big companies help facilitate the process. Linux, for instance, started out at the margins. But then IBM and Google put money into Linux’s development. So you create network effects that disrupt certain industries, and then other big companies adapt to profit from it and facilitate it.
-
-
-
-
Sceptics like Tesla’s CEO, Elon Musk, fear such a development and warn that the progress of AI is the biggest risk that humans currently face.
B
G
As a scientist, I have to admit that we have no basis for predicting positive development of AI. We’re like cavemen sitting in the cave, having just invented language and trying to foresee whether this language will be good or bad. The cavemen will not foresee Shakespeare, the invention of Facebook or differential calculus. Probabilities are based on extrapolating from prior similar situations. But with AI, we are trying to create something entirely new.
-
-
-
-
Does that not scare you?
B
G
No, because non-scientifically, I have a good feeling about it. I see AI as the next step in the human ascent through the creation of tools, industry and computers. Each of these steps scared people that they would lose their way of life. And they did. But then new generations couldn’t imagine going back to the previous way of life. I think that will happen again, just at a faster pace so that it will already affect my immediate family.
Goertzel’s family is an extended group with diverse international influences. He was born in Brazil but grew up in the United States, where he went to university at the age of 15 and met his first wife, Gwen Yorgey. Together they have three children, Zarathustra Amadeus, Zebulon Ulysses and Scheherazade Okilani Natasya. After remarrying twice, Goertzel recently had another baby boy with his wife Ruiting Lian. They live in Hong Kong.
-
-
-
-
Your first grand-daughter was born a few months ago. When she is 51 years old, your current age, what will her life look like?
B
G
By then, she will probably have a chip plugged into her brain to connect her to a superhuman AI, as naturally as my children now use their smart phones. The chip can probably send telepathic text messages and look things up on Google with the power of thought …
-
-
-
-
… Sounds creepy!
B
G
Well, the definition of creepy is constantly changing. Take my grandfather. He never wanted to get money from an ATM. For him, money coming out of a wall was creepy. He only ever went to get it personally from a guy in the bank.
-
-
-
-
But in contrast to an ATM, robots can be threatening if they are smarter than humans, but lack values such as compassion.
B
G
To be honest, even if we introduce human values into the superhuman AI, it might not be enough. Just consider where human values have led us: we kill almost all the inferior species and keep a few around in the zoo.
-
-
-
-
Could that be the fate of humans?
B
G
That is the pessimistic scenario for human-machine coexistence. Say an ASI becomes 1,000 times smarter than us. It might decide that species such as humans are only an inefficient use of energy. So the ASI will keep some of us alive in the People Zoo. The rest of the mass energy that we consume would be transmogrified into some phantom technology RAM for the Super AI.
-
-
-
-
And how do we avoid a future in the People Zoo?
B
G
Future AIs need more than human-level compassion. They need super-benevolence. And that is possible, as you can recognise if you look at some humans who are super-benevolent. So on the benevolence scale, the Super-AI has to be more on the, say, Mahatma Gandhi side then on the Donald Trump side.
-
-
-
-
And how do you turn an intelligent robot into Gandhi?
B
G
We need to raise it in an environment that teaches it about helping others and being compassionate. This is part of the agenda of using Sophia as meditation assistance. There will be AIs that are bank clerks and greeters at Walmart, but we need AIs that interact with social workers, doctors and nurses; then they will absorb the benevolent side of human values.
-
-
-
-
So it’s critical how the first AGIs grow up?
B
G
Yes, a lot of the difference between the Trump AGI and the Gandhi AGI comes down to what experiences you put the AGI through when it forms its mental structure. The question is: Do you want the focus of its early stages to be killing, spying and brainwashing—or helping people? The next decade or two may be decisive.
-
-
-
-
To avoid AIs developing into tyrants?
B
G
I think their character will be neither loving or tyrannical. Rather, a lot of different components will integrate; some will be developed by crazy decentralised projects like ours, others by government agencies or big companies.
-
-
-
-
Similar to the development of the smartphone?
B
G
Exactly. For smartphones, some components like GPS were centrally developed by the US and European government agencies while the scalable manufacturing of the phone was done by the big tech business. However, the majority of phones use the open-source operating system, Linux. So for AI, we likewise don’t need to wipe out the influence of governments and big companies. We just need to get enough decentralised, democratic and benevolent elements into the mix.
-
-
-
-
If that works, how will humans and machines coexist in the optimistic scenario?
B
G
Humans will have two non-exclusive options. One is to take your mind and upload it into the super AI mind matrix and become one with the superhuman AI mind…
-
-
-
-
…and the other is to keep living the old way, like your grandfather did?
B
G
Yes. Maybe some of these people will have a brain implant that lets you use Google queries, but it doesn’t give you such rich real-time interaction to take you beyond the human state of consciousness.
-
-
-
-
So would life for those people be similar to today?
B
G
Just without many annoying parts. People will have a molecular assembler that guides chemical reactions and builds what they want. So they can tell it, Hey assembler, make me a hamburger, or a chair, or a sex robot, with these specifications!
-
-
-
-
And people would finally be free to pursue their creative hobbies?
B
G
No one knows how they would spend their time. Maybe 95 percent of them would spend all day in a virtual reality of porn robots.
-
-
-
-
How far away are we from such digital debauchery?
B
G
Technologically, we have already passed several milestones on the way to AGI. I don’t think we need a new theory of the brain, a new type of mathematics, any radically new algorithms or new kind of computer hardware, like quantum computers—although those would be cool and would help.
-
-
-
-
So what is missing?
B
G
So far the AI field has focused on narrow tools for one specific function…
-
-
-
-
…like a computer that can beat the world champion in the board game, Go.
B
G
Yes, though that’s an interesting case to show our progress in AI. Early in 2017 the AlphaGo program beat the world champion, relying on a dataset of more than 100,000 previous Go games. Half a year later its successor, AlphaGo Zero, was programmed only with the rules of the game, then honed its skills by playing against itself—and eventually became even stronger than the original AlphaGo.
-
-
-
-
But its skills remain focused on a single board game.
B
G
True, so I think in the next phase we will connect various AI algorithms and data structures into systems of systems. There, the different AI systems will interact in a coordinated way.
-
-
-
-
Why do most AI researchers still focus on narrow tasks?
B
G
Because many of them work on certain AI applications in lucrative fields like finance or advertising. Fewer work on more complex AGI. And even fewer try to apply AI to the problems of developing countries…
-
-
-
-
…which is something that you do.
B
G
Yes, thanks to a chance meeting with the Ethiopian AI developer Getnet Assefa. When I visited Getnet in Addis Ababa, I realised how many talented computer engineers there end up with dull jobs like maintaining someone’s Windows system, because there are just not enough high-tech jobs in Ethiopia.
The challenge of bringing IT jobs to Ethiopia is indeed formidable. With more than 100 million inhabitants, Ethiopia is the second largest country in Africa. However, not even half of those aged 15 and over can read or write, falling short of the average literacy rate for SubSaharan Africa (64 percent). Only 16 percent of the population has internet access, often through mobile phones, and there are only 580,000 fixed broadband subscriptions—fewer than one per 100 inhabitants. However, 42-year-old Abiy Ahmed became Prime Minister in April and has since pursued vigorous reforms. He has shown personal interest not only in liberalising the economy and negotiating an end to the conflict with Ethiopia’s neighbour, Eritrea, but also in IT issues. Abiy holds a bachelor’s degree in computer science and—as Goertzel enthusiastically reports after meeting him—actually knows how to write computer code himself.
-
-
-
-
What are you doing to try and improve employment prospects in Ethiopia’s IT sector?
B
G
Five years ago, Getnet and I launched an AI outsourcing company in Addis Ababa. We employ software engineers, many of them for SingularityNET and Hanson Robotics. In the meantime, they have also opened our eyes to options where AI can help with local problems.
-
-
-
-
In what way?
B
G
We are developing diagnosis tools for crop diseases. So a farmer can take a picture of a plant and use some basic tools to measure soil conditions, then get the result whether his plants are affected by a certain disease.
-
-
-
-
Is this really helping farmers?
B
G
Not yet, because we realised the farmers can tell what disease it is anyway. The tricky part is in the early stages; they can’t tell whether a disease is going to get serious or not, and whether they need to get pesticides. But over the last few months we have progressed a lot in that prediction.
-
-
-
-
And will your approach be deployed?
B
G
Yes, but probably in China first, because the farming sector there is more organised. In Ethiopia it will take years to get it deployed in all the subsistence farms. So if I hear of people starving there, I sometimes feel like I’m in a race against time.
-
-
-
-
As in the case of your own life?
B
G
In a way, yes. Because if we advance fast enough in AI and molecular technology, it might only be a few decades before we’re able to resuscitate dead people who have been frozen. So, I have signed up for biopreservation myself in case I die. But it’s even more urgent with my parents…
-
-
-
-
…since your mum has just celebrated her 75th birthday party.
B
G
With her, there is definitely a race against time in quite a personal way. So far we are losing that race, as loved ones keep dying. But living is better than dying as far as I currently understand, so it would be great if an AGI soon contributed to progress so that our loved ones can live forever.
-
-
-
-
Ben Goertzel, thank you for the interview.