The aim of the project is to develop existing methods for novelty search and to propose new techniques for the creation and identification of novell and innovative behavior. The results of the research will be used so that AI, instead of maximizing objective functions, can learn by discovering new, unusual things during its life.
The research will examine areas where the trend towards novelty is based on basic principles, such as the way in which the training environment is set up. The project will try to simplify and generalize the ideas previously published by the American scientist Kenneth Stanley. He will examine AI agents trained in virtual labyrinths, rewarded for discovering new locations. During the research, the objective function and environment will be slightly modified, which will encourage novelty in behavior, while rewarding agents who will behave differently from others. The primary goal is to understand what properties of agents and environments are a catalyst for novelty in agent-environment interactions. The results of the research may allow a major shift in the development of AI algorithms for environments where rewards are scarce, and as a result, this research will help efforts to create AI human levels.
This is the first grant awarded under the GoodAI Grants Initiative for a total of $ 300,000. This contribution program aims to support research groups and researchers from around the world whose research has a chance to answer open questions about GoodAI's Badger architecture. The contribution program remains open to researchers working in a wide range of areas, including multi-agent enhanced learning, learned communication, graph neural networks, learned optimizers, modular meta-learning, and more.
Marek Rosa, founder and CEO of GoodAI, said: “We are pleased to provide this gift to Tomáš and his team. At GoodAI, we admire the results of his work and look forward to our cooperation. Since announcing the GoodAI grant program at our recent meta-learning and multi-agent learning workshop, we have received great feedback from researchers around the world and expect more grants to be awarded in the coming weeks. ”
Tomáš Mikolov, senior researcher at the Czech Institute of Informatics, Robotics and Cybernetics at the Czech Technical University, said: “I am glad that GoodAI will support our research project and look forward to future cooperation on the development of new approaches to AI solutions. I find it amazing that Marek Rosa takes AI research so seriously, and has been working for years on what can be considered the greatest research challenge of humanity today. ”
GoodAI and CTU are both part of the prg.ai initiative, which seeks to strengthen the AI ecosystem in Prague by a system of support for cooperation, research and innovative companies in order to make Prague a world-class AI center.