We might have just hit Peak 2017 Buzzword: a startup is about to launch an Initial Coin Offering (ICO) to fund a blockchain-based network of Artificial Intelligences (AI), called SingularityNET.
The driving force behind the project is Ben Goertzel, a Hong Kong-based AI researcher and Chief Scientist of Hanson Robotics, a company specialised in building humanoid robots — such as eerie talking head Sophia. Over the last few years, Goertzel has grown wary of the concentration of AI power in the hands of a few Silicon Valley giants.
“I don’t think that what’s happening—with a few companies essentially owning AI, hiring every AI researcher, and buying every AI startup— is best for humanity,” he says. “It means that eventually human-level AI will come from these big corporations.”
Making that undesirable, Goertzel’s reasoning goes, is the fact that those companies might train their AIs towards finding new ways of lining their makers’ pockets—for instance, by perfecting targeted advertising—rather than tap into their power for tackling serious issues like climate change. Government-funded AI research is also problematic, as the resulting AI would be in thrall to powerful lobbyists or self-serving bureaucrats.
To break the AI oligopoly, Goertzel is turning to the blockchain, the digital scaffolding underpinning cryptocurrencies like Bitcoin. Transactions on a blockchain are collectively processed by a multitude of computers (“nodes”), a structure ensuring that no single party controls the network. Recent developments have expanded the technology’s potential beyond exchanging digital currency: second-wave blockchains like Ethereum can run whole applications, able to interact with each other through self-enforcing sets of rules dubbed “smart contracts.” These features make the technology particularly attractive to Goertzel.
“SingularityNET’s idea is to create a distributed AI platform on the [Ethereum] blockchain, with each blockchain node backing up an AI algorithm,” Goertzel explains. AI researchers or developers would be able to make their AI products available to SingularityNET users, which would pay for services with network-specific crypto-tokens.
Initially, the plan is to have a system that provides visibility — and payment — to independent developers of AI programmes. “As a customer, you’ll be getting AI from anyone, be it a technology giant or a programmer in Ethiopia or Kazakhstan,” Goertzel says. At first, these programmes will probably be relatively mundane applications, such as translation services or image recognition software.
The wrinkle is that, courtesy of the smart contract mechanism, these AI agents would be capable of communicating with each other, and even working together when necessary. For instance, Goertzel says, a translation application coming across a picture while translating a file could automatically ask (and pay) a computer vision programme to caption the image. Over time, Goertzel hopes that these repeated synergies would go on to become something more complex.
“We want create a system that learns on its own how to cobble together modules to carry out different functions. You’ll see a sort of federation of AIs emerge from the spontaneous interaction among the nodes, without human guidance,” he explains. “It’ll have AI inside each and every nodes, and between them, and they’ll learn how to combine their skills.”
The expected endgame is that these swarms of smart nodes would get as intertwined as clusters of neurons, eventually evolving into human-level AI. Goertzel admits that it might take decades for that to happen, but he is positive that the primary purpose of the SingularityNET project is bringing about “beneficial Artificial General Intelligence” (that is: human-level AI).
Despite the hype which has characterised AI products — essentially, anything using some measure of machine learning — over the last couple of years, a hype partly fuelled by objective breakthroughs such as DeepMind AlphaGo’s achievements, the prospect of human-level AI spontaneously arising anytime soon is pretty remote. But if that were to happen, Goertzel says that what comes out of SingularityNET would be a strong AI that is beholden to no one and theoretically open to everybody to use; its decentralised infrastructure would also ensure that it would be very hard for ill-disposed parties to take it down. That resilience is good if the resulting super-intelligent entity is a juggernaut of artificial wisdom, humanity and self-restraint — much less so if what the network begets is the villainous robo-villain Elon Musk keeps warning us about.
But Goertzel is relatively untroubled by the Skynet scenario, which he thinks could be prevented by ensuring that SingularityNET’s nodes only host beneficial projects.
“One way to go about this would be to guarantee that a certain percentage of the network is devoted to tasks that are voted by the community as being of common benefit: charitable tasks, biomedical research, education,” he says. “If the first human-level AI grows up helping everyone, then it’s more likely that it’ll that ethos. It’s not a guarantee, but it’s certainly better than it would be if it were spawned from a killer robot or an advertisement engine.”
ICOs are becoming a popular fundraising model among startups, mainly because their unregulated nature is attracting a legion of amateur investors and speculators— resulting in what looks more and more like a coin-shaped bubble. That is partly justified by the ease and rapidity with which some ICOs have raised vertiginous amounts: take just exchange platform Bancor, which in June bagged the equivalent of $153 million in cryptocurrency in three hours. The acronym is reaching mainstream recognition thanks in part to the very public backing of celebrities such as boxer Floyd Mayweather and music producer DJ Khaled; on the other hand, some governments, including China and South Korea, have recently banned them altogether, while the US Securities and Exchange Commission has warned that some ICOs could be breaking securities law.
SingularityNET will sell 50 percent of its whole token trove, distributing the other half to its staff and to a foundation that will reinvest them in charitable AI projects. Goertzel is optimistic about the sale, which he thinks could be appealing even to technology heavyweights.
“I have been working with Cisco, Huawei, and Intel, and I think we can pull in a lot of major customers who want to buy a lot of tokens to do AI analysis for their own purposes,” he says. “In general, though, this ICO will allow us to start with a bang. We’ll be competing with Google and Facebook…so having a war chest would allow us to take on them more easily.”