The ultimate vision of artificial intelligence are systems that can handle the wide range of cognitive tasks that humans can. The idea of a single, general intelligence is referred to as Artificial General Intelligence (AGI), which encopmasses the idea of a single, generally intelligent system that can act and think much like humans. However, we have not yet achieved this concept of the generally intelligent system and as such, current AI applications are only capable of narrow applications of AI such as recognition systems, hyperpersonaliztion tools and recommendation systems, and even autonomous vehicles. This raises the question: Is AGI really around the corner, or are we chasing an elusive goal that we may never realize?
Dr. Ben Goertzel CEO & Founder of the SingularityNET Foundation is particularly visible and vocal on his thoughts on Artificial Intelligence, AGI, and where research and industry are in regards to AGI. Speaking at the (Virtual) OpenCogCon event this week, Dr. Goertzel is one of the world’s foremost experts in Artificial General Intelligence. He has decades of expertise applying AI to practical problems in areas ranging from natural language processing and data mining to robotics, video gaming, national security, and bioinformatics.
ARE WE AT A TURNING POINT IN AGI?
Dr. Goertzel believes that we are now at a turning point in the history of AI. Over the next few years he believes the balance of activity in the AI research area is about to shift from highly specialized narrow AIs toward AGIs. Deep neural nets have achieved amazing things but that paradigm is going to run out of steam fairly soon, and rather than this causing another “AI winter” or a shift in focus to some other kind of narrow AI, he thinks it’s going to trigger the AGI revolution.
He states that “any other problem humanity faces – including extremely hard ones like curing death or mental illness, creating nanotechnology or femtotechnology assemblers, saving the Earth’s environment or traveling to the stars — can be solved effectively via first creating a benevolent AGI and then asking the AGI to solve that problem.”
Dr. Goertzel is founder of SingularityNET, a decentralized AI platform which lets multiple AI agents cooperate to solve problems in a participatory way without any central controller. “It’s an infrastructure for what AI pioneer Marvin Minsky called the ‘Society of Minds’ — an approach to AGI in which the general intelligence emerges from the interactions of multiple relatively simple AI agents,” says Dr. Goertzel. “It’s also a way of creating AI API marketplaces to provide practical AI services to businesses.” Dr. Goertzel created SingularityNET for one simple reason: “I intend to create AGI and when I roll out this AGI, I want it to be rolled out in a decentralized and democratically controlled way, rather than in a manner that allows it to be controlled by any one person or corporate or governmental entity.”
In addition to SingularityNET, Dr. Goertzel is working on OpenCog which he says is a novel architecture for AGI, based on a hypergraph knowledge store called the Atomspace. The system enables multiple AI algorithms, including neural nets, logic engines and evolutionary learning systems, to cooperate synergistically in learning and reasoning from and updating this knowledge graph. As Dr. Goertzel states, “It’s based on a sophisticated mathematical theory of general intelligence, which tells us how the general nature of general intelligence manifests itself in the specific case of human-like cognition. In the scope of AI approaches it would be considered a ‘hybrid cognitive architecture’ due to its integration of multiple AI algorithms based on different AI paradigms. However, it is different from other hybrid cognitive architectures in the depth of integration of the different algorithms and in its grounding in the mathematical theory of AGI.”
BARRIERS TO AGI
Despite the efforts of many researchers and companies, there are still many barriers to achieving the goals of AGI. According to Dr. Goertzel , “10 years ago the biggest issue was lack of funding for AI. Now the biggest issue is lack of funding for serious AGI approaches, as opposed to narrow AI systems that mine large numbers of simple patterns from big datasets, like most current deep neural net systems.”
Dr. Goertzel says the other major issue is that current computing infrastructure is not well tailored for AGI. However he believes this can be worked around in the short term via paying for large server farms, and in the medium term by building AGI-appropriate hardware like graph and hypergraph chips. He believes that his company has a workable software and cognitive architecture for AGI in its newest version of OpenCog (OpenCog Hyperon) and they have a way to teach and train it with their TrueAGI framework for connecting OpenCog Hyperon to robots and embedded devices.
At this stage, according to Dr. Goertzel, the algorithms and structures aren’t the bottleneck to achieving AGI goals, but rather it’s more fundamental issues of money and hardware. On the other hand, there are teams pursuing AGI with copious amounts of money and hardware, such as Microsoft’s OpenAI and Google Deep Mind, but Dr. Goertzel believes they are largely fast burning their resources pursuing intellectual dead ends.
WILL WE SEE AGI IN OUR LIFETIME?
One of the most often asked questions of AI researchers is will we see AGI realized within our lifetimes. Opinions on this vary widely with some saying we’re just a few years away, others believing that we are hundreds of years away, and yet others think we’ll never achieve AGI. Dr. Goertzel is of the opinion that we have between five and twenty years to achieve human-level AGI, with less than three years after that achieving super-human level AGI. In fact he believes his company “can fairly likely get to human-level AGI in 5-7 years if we can accumulate reasonable funding into our TrueAGI project, which is based on OpenCog Hyperon, the new version of the OpenCog AGI architecture we’re building now.”.
When asked about the more conservative or pessimistic views of other AI researchers, Dr. Goertzel states that, “they want to believe that human cognition is more special and elusive than it actually is.” To those who state that superhuman AGI is scary or potentially harmful to the human race, Dr. Goertzel states that “… these reactions are probably going to look very silly to people a few decades from now as they go about their lives which have been made tremendously easy and happy and fascinating compared to 2020 reality, via the wide rollout of advanced AGI systems to handle manufacturing, service, and all the other practical jobs that humans now spend their time doing.”
Of course AGI is not the only pursuit when it comes to AI. In addition to AGI, Dr. Goertzel is looking forward to the emergence of graph-processing and ideally hypergraph-processing chips. He thinks that putting graph transformations on the chip will do for both AGI and functional programming the same thing that GPUs did for deep neural nets. It has the potential to make complex scalable work far less costly and more tractable.
Dr. Goertzel thinks that the push from narrow AI toward AGI is likely going to be a very international effort. While no one really has the answer about if or when AGI will happen, there are certainly those like Dr. Goertzel working to make it happen in our lifetimes.
originally posted on forbes.com by Kathleen Walch
