Rise of AI Conference 2019 · Berlin

Decentralized AI

Ben Goertzel · SingularityNET

At Rise of AI 2019 in Berlin, Ben Goertzel argues that a decentralized, self-organizing 'society and economy of minds' is the best route to both artificial general intelligence and a democratic, participatory global AI. He frames general intelligence as a spectrum on which humans sit far below possible superintelligence, and describes AI development moving from today's narrow-AI revolution toward an AGI revolution and eventually a superintelligence revolution. Goertzel warns that AI development is currently concentrated in a dozen corporations and two large governments, whose systems are mostly used for spying, killing, selling and betting, and that this narrow set of goals risks shaping the superintelligence that eventually emerges, while also magnifying global wealth inequality. His response is SingularityNET, a decentralized, blockchain-based AI platform he founded in 2017, funded via a token sale, where independent AI agents, including his own OpenCog architecture, can interoperate and offer services. He describes deployments including the Sophia robot, a pizza-delivery optimization project with Domino's, longevity research, and a new enterprise spin-off, Singularity Studio.

Key takeaways

  • Ben Goertzel argues AI development today is dominated by about a dozen corporations and two large governments, mostly using AI for spying, selling, killing or betting.
  • SingularityNET, founded by Goertzel in 2017 and funded via a token sale, is a decentralized blockchain platform where independent AI agents interoperate and offer services.
  • Goertzel's preferred path to AGI is his own OpenCog architecture, a hypergraph knowledge store combining logic engines, evolutionary learning and neural networks.
  • SingularityNET and Hanson Robotics used OpenCog to run the Sophia robot as a meditation coach, which some users said deepened their meditative states.
  • The Rejuve network studies DNA from people aged 105 and older and lets contributors be compensated via cryptographic tokens if their data leads to a therapy.

Detailed notes

From narrow AI to AGI and superintelligence

Goertzel frames general intelligence as a spectrum: humans are more generally intelligent than dogs or AlphaGo, but are not maximally intelligent, illustrated by his own poor ability to navigate a 3,072-dimensional problem despite the two-dimensional grid cells in his hippocampus. He describes the field as being in the middle of a narrow-AI revolution, heading toward an AGI revolution and eventually a superintelligence revolution. He notes it is unclear whether AGI will come from a single breakthrough algorithm or, as in Marvin Minsky's Society of Mind vision, from many AI systems cooperating, and says he expects a mix of both.

Risks of centralized control

Beyond speculative existential risks, Goertzel highlights a nearer-term risk: AI concentrating wealth so that a small number of people who own automated factories and delivery systems prosper while everyone else is reduced to subsistence. He observes that AI development today is dominated by about a dozen corporations and two large governments, and that the AI deployed by them is mostly used for spying, killing or defending people, selling goods, or gambling on markets. Because these systems are trained to maximize shareholder value or state hegemony, he argues they will not represent the full range of human needs, and current deep neural networks, while effective at finding patterns in data, cannot generalize beyond their training data to invent new scientific theories, art forms or systems of government.

SingularityNET as a decentralized alternative

Goertzel's response is SingularityNET, a decentralized blockchain-based AI platform he founded in 2017 and funded through a token sale, with its beta launched in February 2019. He argues the same self-organizing, agent-based architecture needed for a Society of Minds approach to AGI is also what is needed to democratize AI: a network of interoperable AI agents with flexible APIs and shared memory, owned by everyone and no one. SingularityNET connects with the related Ocean Protocol network for decentralized big data, and Goertzel says over 50 projects now form a broader ecosystem combining AI and blockchain. His own preferred path to AGI remains the OpenCog architecture, a hypergraph knowledge store combining logic engines, evolutionary learning and neural networks, which can operate on the network alongside AI agents built by anyone else.

Applications: Sophia, longevity and enterprise AI

Goertzel describes several deployments: OpenCog and SingularityNET have been used to run the Sophia robot developed with Hanson Robotics, including as a meditation coach that some users found deepened their meditative states; a project with Domino's optimizes pizza delivery; a network called Rejuve analyzes DNA from people aged 105 and older to study longevity, using a cryptographic token so people who contribute biology data can be compensated if it leads to a therapy, unlike 23andMe, which he says sold sequencing data to GlaxoSmithKline without compensating users. He is also launching Singularity Studio, an enterprise spin-off working with companies including Ping An Technology, to help large firms without internal AI expertise integrate SingularityNET's decentralized AI into fintech, health-tech and IoT products.

Quotes

If that's what the baby AI is doing with all its time in its mind, what kind of a superintelligence is it actually going to grow into?

Ben Goertzel

AI magnifying global wealth inequality could lead to a lot of other problems.

Ben Goertzel

It should be all of us and everyone on the planet together.

Ben Goertzel

Questions answered in this talk

What is SingularityNET?

According to Ben Goertzel at Rise of AI 2019, SingularityNET is a decentralized, blockchain-based AI platform he founded in 2017 where independent AI agents, including his own OpenCog architecture, can interoperate, exchange data and offer services to anyone on the network.

What risks does Ben Goertzel see in AI being controlled by a small number of companies and governments?

According to Ben Goertzel at Rise of AI 2019, AI development is currently concentrated in about a dozen corporations and two large governments whose systems are mostly used for spying, selling, killing or betting, which risks magnifying global wealth inequality and shaping a future superintelligence around narrow commercial or state goals.

What real-world applications has SingularityNET's AI been used for?

According to Ben Goertzel at Rise of AI 2019, SingularityNET and OpenCog have powered the Sophia robot as a meditation coach with Hanson Robotics, optimized pizza delivery with Domino's, supported longevity research through the Rejuve network, and now power an enterprise spin-off called Singularity Studio for companies such as Ping An Technology.

Mentioned

artificial general intelligence, superintelligence, decentralized AI, Society of Mind, blockchain, token sale, life extension

Keep exploring

More from Ben Goertzel

More from Rise of AI Conference 2019

More on AGI & Future of Intelligence

Transcript

Auto-generated transcript, not edited. · 3,723 words

Show the full transcript

so this is exciting to be back here again for the rise of the rise of AI and we've got a much larger larger event this year with some so much more going on which of course is just symbolic of the actual rise of AI right i mean AI is pervading popular consciousness in every branch of Industry and government and so on it's quite a quite exciting and you know this is it's a bit unique positioning for this talk I'm giving now because the talk just before was really amazing articulate and eloquent in letting out a transhumanist vision for the future and we heard about about AI 500 years five hundred years from now and how a I will transform everything and and upgrade upgrade our minds and give us ability to reshape matter as we wish so hopefully a bunch of you were here for this previous start because it means I don't have to I don't have to explain all that I mean I think it it should be clear that if we can

get over a few small hurdles in the next few years and decades if we really get superhuman AI that is you know orders of magnitude more intelligent than current human beings and is reasonably benevolently oriented toward other sentences besides itself and you know this is this is gonna be amazing it's gonna give us nanotech and femto tech and cyborgs and and mind uploading and whole new ways of individual and collective experience that we can't can't imagine now and I really enjoyed the last talk summarizing this and of course everyone by now has also heard from Nick Bostrom Elon Elon Musk the Terminator and all the all the other dystopian possibilities and things that could go wrong if AI becomes advanced but has a malevolent or indifferent attitude toward other sentient beings or even even to even toward itself so I really love that we now have people talking about these far future possibilities you know in Deutsche Telekom and in front of a you know an audience of of largely very very practical people rather

than like a science fiction fan convention or something I've been working on the AI for I guess over over 30 years now I got my PhD in the late 1980s and I mean at that time let alone in the early 1970s when I first became enthused about these concepts from science fiction these ideas were not being discussed and in venues like this to audiences like this so what I'm gonna talk about now actually is a little more near-term and practical and short term focus than the last talk which is it's not it's not normal that I can say that usually I'm the furthest furthest out guy so but what I want to talk about here is some aspects of how do we work now so that the AI we are creating now sort of serves not only to deliver you know maximum benefit to people on earth now but so that the AI we're creating now serves as the appropriate you know seed or child form for super intelligence that that unfolds in the next

decades and centuries as we just heard about and in particular I'm going to talk about decentralization decentralized governance and control in large even global networks of a eyes and why that might be valuable or even critical to the next phase of AI development coming out in a in a positive way so I've spent a bunch of my career developing both formally and sort of in the popular sphere and in the research community the notion of a GI artificial general intelligence I've organized each year since 2006 a GI conference a mostly technical researchers the next one is in early August in Shenzhen in China right across the border from Hong Kong where I've lived for the last eight years so that's the AGI conference series we'd love to have you come to AGI 19 in Shenzhen we're looking at bringing together you know what the Chinese AI community is doing and what the the Western AI community is is doing and of course general intelligence is a spectrum humans are more generally intelligent than dogs or

than alphago for that matter humans are not the maximally generally intelligent creature possible and when we talk about super intelligence we're talking about a eyes with much much more general intelligence than than human beings I mean as an example if you give me a navigation problem in 3072 dimensional space I made you very poorly at it whereas navigating in two or three dimensions I'm okay this is an example of the lack of generality in the human brain I mean my hippocampus like everyone's has these two-dimensional grid cells we can make two-and-a-half these sketches we're not that good at making 3072 and a half these sketches as super intelligence with route access to its own brain and the ability to synthesize new brain modules that will could deal with thousands of dimensions or four dimensions right so you just like we're not the fastest possible runners are the highest possible jumpers we're not the smartest possible thinkers and we're now seeing amazing advances in AI but we're at the very early stage we're in the middle

of a narrow AI revolution in which AI is being rolled out one after another vertical market which and is solving one after another hard problem but we're about to see the AGI revolution which by the way should remind you all to read my book titled the AGI revolutions and then following that the super intelligence revolution which no one has written a book about yet we'll leave that for the AG eyes to write that one and how advanced AG I will emerge nobody knows I mean I've been trying to create it for a while many others have and maybe that one guy just writes the right algorithm and then the algorithm gets ten a hundred a thousand times human level intelligence it may be on the other hand the AGI comes about more from the cooperation of many different a eyes working together this was Marvin Minsky's vision of the Society of mind and I guess it'll probably be a little of each in my view I mean you'll have many a s cooperating together some

of course will contribute more profound abstract cognition than than others but it may be a collective sort of collective sort of intelligence and in terms of the risks of this sort of eventualities narrow AI seeding to AGR seeding to super intelligence of course there are risks that an AI decides to turn all the molecules in the solar system into new raw materials for its it's hard drive and in there there's science-fictional risks out there there's also near-term risks that we wound up with a situation where say ten thousand a hundred thousand a million people own all the robot factories that make luxury goods for them from from the mind to the delivery drone and then everyone else is reduced to subsistence farming and rummage around in the garbage from the robot factory right I mean so AI magnifying global wealth inequality could lead to a lot of other other problems as we worry about terrorism and so on clearly massive global wealth and the quality would not decrease the risks there and we can see

already things unfolding in a somewhat precarious and disturbing direction I mean right now there's a dozen companies and two large governments that are arguably leading practical deployment of AI on the planet although the invention of AI algorithms is still more from universities and and independent independent researchers so yeah we have to ask you know if you look at the AI we have on the planet right now which is largely concerned with spying on people killing or defending people in one region of the earth again against others selling people's stuff probably mostly stuff they don't need or beating other people on various gambling games like the stock market right I mean if this is if this is it like spying killing selling and betting right if that's if that's what the baby AI is doing with all its time in its mind what kind of a gir super intelligence is it is it actually gonna grow into and what I think you can even walk this point back into the underlying AI algorithm so I mean

I'm I'm a fan of hierarchical pattern recognition and deep neural networks I was teaching this and in the in the 1990s knows cognitive science professor but I think these sorts of algorithms are having so much success now face recognition game playing and so forth these are a small corner of the scope of known AI algorithms and they're strong in certain ways but very limited in in other ways I mean deep neural Nets as currently used are very good at recognizing a large collection of shallow patterns in a huge amount of data and finding sort of the interdependencies among these simple patterns this is very valuable if you have a lot of data and need to exploit the differential commercial advantage that gives you on the other hand it's really just learning patterns in the data that that you give it none of the neural nets that are being used now commercially at large scale are forming a really abstract representation of what they're seeing and this comes at the cost in terms of creativity and

and the ability the ability to to generalize broadly so I mean you're not you're not gonna see a eyes based on current versions of deep neural networks like inventing a new scientific theory or a new art form or a new system of government there's something yeah I mean you're gonna see them churning out new things according to the patterns that they were trained on and you know this is not bad if these are tools being used for specific purposes on the other hand if you have a world that's run by these pattern recognition algorithms that slavishly adhere to the day that they were trained on then this is this is a bit interesting and when you consider that the goals of these narrow systems are being trained by supervisors and reinforcement learning are the goal of maximize shareholder value of this dozen corporations or maximize you know hegemony and stability of these two large governments I mean if you have a eyes that are narrowly trained to pursue the goals of these organizations according to

all the relevant data on the planet this these a eyes aren't gonna represent the full scope of you know passions dreams inspirations ambitions of humans on the planet and they're not gonna serve the full scope of needs of humans on the planet because many needs of humans on the planet aren't that important to these two large governments are these dozen corporations so this this leads one to think that perhaps a more broadly democratic participatory self-organizing approach would work better for the evolution of AI on the planet as to an approach where two governments and and the dozen corporations are controlling everything and then this this was really the inspiration for creating singularity net which is a decentralized blockchain based platform for AI which which by the way is being connected with the ocean protocol network that Trent McCulloch II described in this stage and an hour or so ago so ocean trance Network is focused more on blockchain for big data and we're using blockchain in a similar way for AiAi processing and obviously the

can and do connect together in a very natural way really the what inspired me to put the last couple years of my life behind singularity net is observing that you know what you need to create a GI according to a society of minds type framework and what you need to democratize AI from a more general you know social political futurist framework it's the same thing you want a network of AI agents that's you know self-regulating where the society of a eyes controls governs and organizes its its own self and where you have flexible system of evolving api's for different I ask you use to communicate to each other flexible mechanisms for a us to share memory with each other I mean what you want for Society of Minds AI is the same self-organizing agent system that you want for the purpose of democratizing AI but you know that this thing that Ocean protocol singular Gina and others are doing basically it's trying to build a whole alternate parallel ecosystem for AI based on unblocked chain

and decentralized control technologies rather than based on large centralized databases which needs a lot of things in these platform developers ai developers customers so building this whole ecosystem is is hard and we're at the early stages of it on the other hand I mean building Google Facebook and all these things was hard and they you know they did did away with her or drastically reduced in impact the large corporations that came before them so we we started singularity net project in 2017 and funded it via a token zone and ICO and we launched the beta version of the platform late February of this year put some example AI agents on it and we're now doing developer workshops and working with various various large large enterprises to actually actually use the platform most recently a project with the Domino's Pizza to use our decentralized blockchain based AI to optimize pizza delivery so this this this I think can trigger recursive self-improvement leading to the singularity because programmers like to eat pizza so if if the AI

is optimized to deliver pizza faster then the programmers can can program faster and then they can program better algorithms to speed up pizza delivery so you have you have X exponential improvement right but so we're we're integrating a lot of different things into this singularity net framework my own preferred approach to AI remains the open cog architecture that I talked I talked about here last year which has a weighted labeled hypergraph knowledge store and you have logic engine evolutionary learning neural networks all working together so if if OpenCog is the golden path to AGI then singularity net will come out as a way for many open cog agents to you know rise toward AGI pulling together contributions from any other AI agents that are good at other specialized kinds of learning could be computer vision time series prediction whatever it is on the other hand if open cog is not the best approach you know some new AI algorithms come up with by some 16 year old coder in his basement in Kazakhstan or somewhere

they can put it on the singularity network it can provide services to whoever wants it using this decentralized AI marketplace and then that and that can become a dominant feature of the society of minds I mean that's the beauty of having a decentralized network owned by everyone and no one anyone could put an AI into the network it can then talk to the other AIS in the network exchange data and make API calls and it can provide AI services to anyone who puts a request request into the network so we're we're using this for a number of things now I'm gonna know in the last couple of minutes just go pretty quickly through some of the current and some of the in the works applications of this network so we we've used OpenCog and some single unit functions to control the Sofia robot there are many control systems behind sofia but this is one that we've used some time in my work with with Hanson robotics we used OpenCog and singular unit to how Sophia

be a meditation coach to people which was was was quite fun and had a profound positive impact on some people who really felt you know meditating with the robot as a guide helped them get into deeper states of consciousness and they were able to do even with human meditation teachers or or on their own where we're working on a virtual assistant which isn't yet ready for primetime yet but we I mean the aim here is one that actually understands the context underlying its its interactions rather than just being a souped-up chat bot which is which is what most of them are now a bunch of work on life extension biology I mean just in case AI works AI goes comes more slowly I'm I'm 52 we can use the we can use AI to make in longevity pill to let me survive till that so the AGI comes about and would should have some interesting announcements on that soon we've been looking at DNA data from people aged 105 or over and our AI tools

with OpenCog embedded in singular unit platform connecting with other bio AI tools we are discovering some unique patterns about why some people can live live that long I mean how the how they use that for therapy is another question but we're we're also looking at using token aa mcc's and the sort of cryptographic token underlying singularity net to make an ecosystem not only of a eyes but of people uploading biology data pertinent to their health health and longevity so that people who upload their biology data into the network if that data is used to make a discovery that then is a valuable therapy they can get some compensation from from some monetary compensation using the cryptographic token for these it was made of their data unlike an say 23andme know if you have them sequence your DNA they sent you a report of it they also kept it in their database and sold it all the GlaxoSmithKline right that's not terrible I mean blacks I was using it to discover new therapies but wouldn't it

be nice if you had some agency over the use of your data some some compensation some ability to participate in early clinical trials so we're we're a network called rejuve connecting patients clinics a eyes and and bio labs toward AI for Life Extension we're launching a spin-off called singularity studio which is AI for the enterprise so we're approached by a lot of big companies I mean Domino's was one but there's a lot of other household name be working with ping on technology large insurance company in China and a bunch of other household names I can't disclose yet but we're working with a lot of big companies who want to use the singularity net AI but they don't necessarily have the internal expertise to so we're building vertical market specific products initially for FinTech them for health tech and IOT and doing some consulting to help big companies integrate these these products into their infrastructure so that's more you know how do you bridge the world of corporate IT with this crazy decentralized blockchain based AI

network by a combination of products and services and so if yet if if anyone works with a large corporation that's interested in exploring intersection of AI and blockchain you know now before we mind upload and become super humans super human super intelligences yeah you can email me or contact our biz dev people and we're looking at this singularity studio enterprise AR company you know it's a way of helping drive utilization of the decentralized network because any any platform which is ultimately what singularity net is it needs it needs supply and demand right the supply is AI developers putting code in there to accompany our OpenCog and other code and it needs demand which is users which could be you know entrepreneurs or hackers or it could be large companies who on their own or with the help of singularity studio use this to give you know additional AI firepower to what they're doing I mention ocean protocol so we actually have a new industry organization with 50 and growing different projects to bringing AI and

blockchain together and they're all connecting to each other so it's a network of network singular the ocean a bunch of others all using blockchain for AI and cooperating together to try to form a new AI ecosystem which is decentralized democratic owned by owned by by everyone and no one and right now we're running singularity net stuff on server farms around the world but we're also working on a decentralized approach to computing as well well you can use mesh networking through our partnership with right mesh and then bits and pieces of the global AI brains thought process it can be decentralized you got some AI processing running on your phone just as some of the data feeding the AI mind is it's coming off your phone right send it so this is a society and economy of minds extending through embedded devices now when when brain computer interface and goes further we can put the put the device device in here and then then we have some bio computing going going on as well so I

think to briefly sum up since I've over time I mean we are at a very critical juncture here the futurist visions we heard in the last talk the Utopia of you know super human AGI bringing us amazing frontiers of experience and technology this can all happen a lot of less interesting and exciting things can also happen and you know we are the ones who are going to guide this and part of this has to do with what the eyes we have now are used for and who controls what they do it should be all of us and everyone on the planet together and we do have the technologies to make decentralized and democratic a AI happen thank you

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.