Ben Goertzel has the iconoclast-genius look down pat: super skinny frame, a leather vest that might have been hip in the 1980s, heavy-metal locks of hair, and, on the day he sat down with DigFin, a fedora whose spots are somewhere between Holstein cow and Dalmatian dog.
He’s been involved with artificial intelligence for over 30 years. How did he get into A.I. in the first place?
“Probably from watching a lot of Star Trekin 1971,” he said.
A few questions from DigFinabout the history of A.I. made it clear how little your correspondent knows of the topic (despite, if he may, a pretty solid grasp of the original Star Trekseries). But Goertzel doesn’t mind: he lights up when talking about the different categories of the field.
From narrow A.I. to general intelligence
On this day, however, he’s meeting DigFinto talk about fintech. He has just launched a for-profit business called Singularity Studio that is a decentralized marketplace for A.I., and Ping An Insurance is the first big financial institution to announce it will be using the service.
Goertzel is CEO and chief scientist at a research foundation based in Hong Kong called SingularityNET. His business partner, Cassio Pennachin, the foundation’s chief A.I. officer, will serve as CEO of Singularity Studio.
Goertzel’s love is helping advance the field from today’s various “narrow A.I.s” that are good at specific, repetitive tasks, towards “artificial general intelligence”, or AGI, in which computers can learn, reason and imagine beyond the data they are trained on.
(Beyond that is “singularity”, the moment we achieve a super intelligence, in which computers create other computers using code beyond human capacity to design.)
He has been at this for a long time, in the areas of human-like robots at Hong Kong firm Hanson Robotics (built by engineers in Shenzhen), among other ventures.
One way to get there is to enable a multitude of “narrow A.I.” applications to interact, and to get enterprises to draw upon a field of apps beyond their immediate sector or in-house capabilities. SingularityNET has been creating such narrow A.I.s, and now Studio will serve as a means for all kinds of developers, users, and even the A.I.s themselves to interact.
“These A.I.s connect and talk to each other with many other different A.I.s,” he said. Some are highly specific, others are for general learning, but they are combining with each other. “They’re ingesting data from the internet, like it’s a primordial soup that will create new forms of A.I.”
Moreover, Goertzel wants to ensure AGI is available to more than just mega corporations or governments.
“This is not for the research lab,” he said of Studio. “It’s in the real economy, it’s in the internet, it’s impacting us. Many A.I.s have useful functions, like helping doctors, or predicting financial markets, and optimizing supply chains.”
A.I. for all?
But there are legitimate fears that A.I. could also wipe out lots of jobs, or that it’s being exploited by the U.S. and Chinese militaries. Meanwhile the huge tech companies in both countries are hiring or acquiring many A.I. startups, hogging the talent and the data. And these corporations, with government ties, focus most on developing A.I. for advertizing, surveillance and military functions.
“The top three uses are selling, spying and killing,” Goertzel said. “Do we really want a few mega corporations to architect the AGI that will guide our future?”
He decided blockchain technology could be used to support a way to support other forms of enterprise, help smaller or more niche developers sell their wares, and, because it’s decentralized, make A.I. available to far more types of users. Studio is a Dapp (decentralized application) store in which the A.I.s can also outsource to one another and share data.
“This increases the odds that AGI will be controlled by all users and data contributors, and not just a small group of powerbrokers,” he said.
From token to trading
Ping An, though, and the global banks he says are also experimenting on Studio are pretty mega-corporate themselves. For Goertzel, this is firstly a commercial enterprise; secondly, it’s also useful to give oxygen to a broader array of A.I.s, and thereby improve industries from finance to healthcare to agriculture.
This all results from a 2017 initial coin offering whose AGI token raised financing for SingularityNET.
Unlike most ICOs – unregulated fund-raisings for projects backed by a digital token – this one may actually turn out to create a useful business. As more users such as Ping An or banks buy A.I. products on the decentralized marketplace, which they access via the AGI token, the value of the token should increase in line with its utilization.
Benefiting financial institutions
But why should a financial institution care about A.I.s in unrelated fields? Why should developers think they can find new clients via a marketplace like this?
“A.I. is more diverse than people realize,” Goertzel said. “The focus in the media is on A.I. that makes money for big tech companies. This requires a lot of data, so this A.I. creates profits to companies that have the most data. It’s a real thing, but it’s not the only thing.”
For example, his son, a PhD student, is developing A.I. to prove math theorems. Goertzel gets excited about this sort of thing, but it’s not going to make the news. Similarly, he talks about genomics, using A.I. to identify sequences in genetic code that could extend human life; or agriculture, to diagnose plant disease.
But what’s that got to do with finance?
He sites an investment portfolio in emerging-market currencies and commodities. That hedge fund, if it incorporates A.I. designed for agriculture, can use it to spot relevant changes in, say, crop disease, that will have later impact on commodity prices.
Another example: he says some fintech customers are using Studio to hedge real-estate risk; or they are using views on agro-business to find assets to use as hedges against real-estate exposures.
But is this just a more complicated black box, one that develops trading strategies that work – until they don’t?
“No one can deal with regime changes,” he said. “We haven’t solved this, otherwise I’d be a quadrillionnaire. But we have a novel framework for regime changes, global risk assessment, and crash predictions.”
Given the deepening complexity and connectivity of global financial markets, Goertzel argues that the number of things impacting financial instruments is too great for a human to master. “If you get more A.I.s talking to each other, you get more inputs to products and risk assessment.” That extends to credit scores and insurance.
Ping An's use cases
It’s not just about diverse fields: it’s also about combining different styles of artificial intelligence. For example, Ping An has a strong in-house capability around neural nets.
Some of the A.I. programs that Goertzel has developed are based on cognitive science, while other people are better at A.I. in evolutionary computation, probabilistic methods, or fuzzy systems.
Ping An, therefore, can leverage Studio to access A.I.s that it can’t build internally. Its computer vision team can use these to augment their work on OCR (optical character recognition) to better identify Chinese characters in photographs; or to build facial-recognition models. The financial group is also interested in A.I. to get a better grasp of its data, while the biomedical team can use new A.I. algos for specific projects, including deploying them on Studio to interact with other A.I.s in the marketplace.
For now, users need the AGI token to pay for services on Studio, which is an Ethereum-based coin. Studio is working to integrate a fiat-to-crypto gateway so that enterprises can use it without needing to get into the token world. “We need payments for mass adoption,” Goertzel said.
Operating in China
Secondly, the firm has had to build a separate infrastructure for mainland China developers or users. Studio went live on its mainnet last month, but it can’t use the AGI token in China, where the authorities don’t want it to be used for people to swap out into dollars. Therefore Singularity Studio is working on a token that only converts in and out of renminbi.
Both sites will contain the same algorithms for sale. But from there, things could diverge, particularly given today’s U.S.-China tensions, particularly over data and technology.
Goertzel, an American operating a Hong Kong-based business with mainland Chinese customers, accepts these political tensions as just “part of life”, he said. There are no borders for core A.I. algorithms, which are all published in the same journals and are built using open-sourced code. It’s the data sets that are being walled off.
However, Goertzel says it’s a mistake to think of this is purely China versus America terms: “U.S. versus China is a misdirection.” It’s also a tussle between big tech companies hogging data at the expense of smaller ones, or the rest of the world trying to compete with American or Chinese tech.
Goertzel hopes by decentralizing access to at least the algorithms, A.I. won’t benefit only those two countries’ internet giants – even if one of them is his first publicly announced commercial client.