What do Will Smith, Jimmy Fallon, Piers Morgan, and United Nations deputy secretary-general Amina J Mohammed all have in common? They have all interviewed Sophia the Robot, this year’s keynote speaker at ITW.
Last night, as I’m sure many of you saw, Sophia and her creator Dr. David Hanson, the CEO of Hanson Robotics, sat down on stage in the Regency Ballroom to take part in this year’s ITW keynote. This followed an afternoon meeting with the Global Leaders’ Forum (GLF), where the enthusiasm for the amazing creation of Sophia was clear, and not just from Hanson.
Once he’s spoken passionately about future technology with the GLF, I sit down with the robotics expert to get his view on telecoms and technology.
He says: “I get excited by talking about the issues because it is a chance to explore them further. When I’m travelling around I think about these things all the time and talking is a good chance to explore these is important issues.”
Sophia is being moved to the room where she’ll do the keynote as we talk, and I ask Hanson about the computational and connective power needed to keep a robot that can display more than 50 individual expressions working.
His answer surprises me. He says she isn’t connected to the internet at the moment, meaning she only requires a small amount of computational power (1MB) although this limits her interactivity.
“When Sophia is not connected to the internet she can answer questions and interact just fine, but when she does have a connection, she can provide much deeper set of answers, running with our MindCloud AI service,” he says.
The data she is managing now includes camera data, 3D sensor data and microphone data, plus a lot controlling her motor functions “but it is something less than a megabit right now”.
For achieving ‘human levels’ of performance and interaction in the real world it will require “considerably more” data, says Hanson. “You’d have a vast amount of surface sensor data and the tactile data becomes very important. Right now, her tactile data is very limited but we will be adding a lot more of those kinds of sensors in the future. Some of that can be processed locally. We’re seeing much better vision processing and deep learning that you can do on graphics cards in real-time and dedicated machine vision and learning modules and processors.
“However, I can imagine for human level intelligent performance, for the kind of learning you need, you’d be looking at a minimum of 20Mb – we expect that if we have more powerful local processing, such as mobile computing, you might only need to send a few Mbps to the cloud network.”
This is where he gets excited. It seems to be discussing newer, more cutting edge technologies that thrills Hanson, who moves onto mesh computing.
“You can also load balance some of that processing against mobile devices and other small connected, IoT devices. If we consider that we may be able to utilise a mesh network of phones and other mobile computing devices and, through machine learning, be able to pick up data, you would decrease the loads on the cell towers and other networks.
“You could also take the resulting machine-to-machine models and share those with the cloud, so you don’t have to share the raw data but the resulting, interpreted models.”
That is going to reduce the load on the networks enormously – meaning networks could be directed to supporting other functions. Once machines get smart, he adds, they can do more with less data.
The impact of artificial intelligence on future communications networks will be huge, Hanson expects. AI will help telcos to create more robust networks through adaptive problem solving and this kind of product is already being tested on some networks. Data analytics mixed with AI can also help networks become more agile and more reliable.
Blockchain
Another area of passion for Hanson is blockchain, itself a topic that you’re going to be seeing a lot about in the next few days. Part of that is down to the GLF. Capacity announced in March that Colt and PCCW Global have started trials to see how blockchain can speed up time for inter-carrier settlements and make them more reliable. The proofs of concept they have carried out so far indicate that blockchain can cut inter-carrier settlement times from hours to minutes.
Hanson himself has history with blockchain, too. Hanson and Ben Goertzal, the chief scientist at Hanson Robotics, have developed SingularityNET - an ambitious project to create a decentralised marketplace for AI.
It has received notable backing in a token sale last year. In around 60 seconds after opening the sale to the public, it sold out of the whole amount of available tokens bringing the total raised to $36 million.
Goertzal is well-known in this sector, having coined the term “Artificial General Intelligence” and developed open source architecture OpenCog.
Hanson says: “The idea of smart contracts can allow people to opt-in in ways that wind up being more beneficial and verifiable. So if the technology is used well it can be of great social benefit and of benefit to businesses. Smart contracts are of great interest within the blockchain.
“We came up with this idea of SingularityNet which snowballed. In its own foundation we spun this out – Ben Goertzel – who is a famous AI researcher who coined the term Artificial General Intelligence and has been working on this problem for 30 years. He developed an open source architecture for pursuing artificial general intelligence called OpenCog. The concept is that it is not just one framework or even one approach that will result in great breakthroughs.
“We see this as a first application in order to integrate other AI systens and create an open market place for AI technologies.
The concept is like a highway, with the road itself a free route, but there may be some tollgates here and there, with proprietary intellectual property that is integrated. “You have a tool for integrating tha,” adds Hanson, “but by putting pieces together you create a huge global toolkit for assembling AI technologies into different configurations so you can experiment faster, develop products faster and get them to market faster, develop fundamental R&D faster – this is the idea of SingularityNet.”
So how does this impact telecoms and wholesale? “Scale becomes really important,” he adds. “Tools available in telecoms are essential to the future of these things because they are already scaled.
“Designing these tools so that they can scale through telecommunications becomes instrumental. The telecoms industry provides an optimised balance between power and price and that is where the real world implications spring forth. The ideas are great but they are only relevant when they hit the real world.”
For Hanson, telecoms will play a key role in supporting an Ai-powered future. But it will also benefit from this kind of next-gen technology. I ask him if telecoms is one of the quicker industry’s at adopting this.
He pauses. “My impression is that telecommunications will adopt a technology once it is proven and scaled. Then the industry will adopt it like wildfire. Passing through that gate can be somewhat complicated. The computing you have in a mobile device has been in a very competitive space.”
