New Project to Develop OpenCog-Based Intelligent Video Game Characters Starts Up at Hong Kong Polytechnic University….
Many video games now feature AI characters, but the intelligence involved is always limited, focusing mainly on pathfinding and simple rule-based decision-making. There have been a few notable exceptions, like the cute little evolving beasts in Creatures, and the teachable monster in Black & White, but nothing verging too close to general intelligence.
But even so, there is a great potential for synergy between deep AI research and video games – and this has been noted in the research community for some time, for instance in a classic 2001 paper by John Laird and Michael van Lent, titled “Human-level AI’s Killer Application: Interactive Computer Games.” As observed in that paper and by many others over the years, game worlds provide an appealing way to give AI systems an environment requiring perception, action, language, emotion and socialization, but without the engineering difficulties associated with robotics. But, in spite of the emergence of a conference series devoted to the synergy of AI and gaming (the AIIDE series), there hasn’t been a huge amount of work really taking the promise outlined in the Laird and van Lent paper seriously.
As many readers may know, the OpenCog project has touched on the AI/gaming intersection somewhat already – it has been used to create an OpenCog-controlled virtual dog that can learn new tricks via imitative and reinforcement learning. This application originated via a collaboration between Novamente LLC and The Electric Sheep Company to make AI virtual dogs in Second Life, which unfortunately was never commercially launched due to a downturn in the virtual worlds market.
So it’s with great excitement that we now announce the start of a new project at the intersection of AI and video gaming, hosted at Hong Kong Polytechnic University and involving OpenCog at the core.
The M-Lab at Hong Kong Polytechnic University was recently awarded a grant by the Hong Kong government’s Innovation and Technology Fund (ITF), to work on “A Software Toolkit for Creating Intelligent Non-Player Characters in Video Games.”
The focus of the project will be the creation of generally intelligent humanoid game characters, powered by OpenCog and M-Lab’s Lucid game engine, with the capability for simple English conversation and realistic human-like emotional dynamics.
M-Lab is a development laboratory, based at Hong Kong Poly U, which specializes in creating new commercial opportunities within the realm of digital entertainment. While not a commercial venture per se, our new Hong Kong project is expected to result in technology with the potential to be spun out from M-Lab and Novamente into the game industry — leading, if all goes as hoped, to a dramatic advance in game-character intelligence.
The project will last two years and has received US$380,000 in funding, the majority of which is supplied by ITF, with a minority supplied by Novamente in collaboration with Epstein Interests and Humanity+. Additional sponsorship is also provided by NeuroSky, Intuitive Automata, Eutechnys and Fifth Wisdom Technology. OpenCog has previously been used for commercial projects in natural language and data analysis, as well as for basic AI research, but this new Hong Kong project is the largest funded OpenCog research project so far.
The new project will be led by M-Lab faculty Dr. Gino Yu and Wilson Yuen; and OpenCog cofounder Dr. Ben Goertzel will provide oversight on the artificial intelligence aspects. OpenCog cofounder Dr. Joel Pitt has recently relocated from New Zealand to Hong Kong to serve as the technical lead; and Jared Wigmore, another New Zealand OpenCog contributor, plans to shortly follow suit to join Joel on the project. The team will be rounded out with M-lab game programming and graphic design experts, and AI graduate students visiting Hong Kong from Xiamen University in China, where Dr. Goertzel has been co-supervising an OpenCog-based robotics project.
The field of Artificial General Intelligence is at an interesting juncture. There are many preliminary indications that the long “AGI winter” is near to ending, with an increasing number of researchers, writers and policymakers taking more seriously the prospect of creating software or hardware with the same sort of general intelligence as humans. But, in order for the AGI field to really take off explosively (as, say, genomics did in the 1990s or space research did in the 1960s), it will likely be necessary for someone to create a “Sputnik of AGI” – an AGI software system that displays palpable general intelligence, in a context that renders it strikingly comprehensible and apparent to the untrained observer. Video games seems a domain well-suited for the “AGI Sputnik” achievement – so, whether this new Hong Kong project reaches the Sputnik level of influence or not, hopefully at least it will be a major step along the path.