Summary
- Cross Labs in Japan has received a grant, from GoodAI, in order to carry out artificial intelligence (AI) research. The research will be led by Postdoctoral Fellow Lisa Soros.
- The research team will build a complex open-ended environment that will encourage AI to learn in a similar way to humans, by replicating the complex environments that humans learn in.
- By using this environment they aim to create an artificial agent capable of open-ended improvement.
If we are to develop artificial intelligence (AI) capable of learning as humans do, it needs to be tested in complex environments just like humans are. A new research project, being undertaken by Cross Labs in Japan and funded by research and development company GoodAI, aims to develop a radically new kind of testbed for AI that better replicates the learning conditions of humans.
The research project will be led by Postdoctoral Fellow Lisa Soros and will seek to build AI systems that are capable of creating and completing new goals without explicit reward signals and without any prior domain knowledge. These systems will operate in an open-ended artificial environment that will be created as a part of the grant. In essence, this should allow AI to think and create targets for itself, something that so far is very limited in modern AI.
The ideas in Dr. Soros’ research, particularly around understanding the general principles underlying open-endedness, which may lead to the creation of open-ended AI have significant parallels with GoodAI’s Badger Architecture research goals.
The project will be carried out in two phases. The first involves building a 2D “toy” environment, dubbed artificial principles, that captures a minimalist representation of the proposed new environment. This will allow the researchers to experiment with various design configurations. During the second phase, the environment will be extended into a more embodied game-like physical world. In this world, the research team will be able to ground the 2D domain in physics potentially leading to endless possibilities of creation and discovery for the AI agents.
Dr. Soros said: “Understanding how to create open-ended algorithms that innovate forever is a major goal for the artificial life and artificial intelligence communities, and I am excited for the opportunity to create a domain focused specifically on open-endedness. The collaboration with GoodAI is enabling a truly unique line of research.”
The grant is part of the ongoing GoodAI Grants initiative, which has awarded over $600,000 so far. The initiative is supporting research groups across the world that are solving problems related to GoodAI’s Badger architecture. The vision is that each research grant, along with the work of the GoodAI research team, will contribute in some way to basic AI research, and all together fill in some of the gaps in the roadmap to advanced, increasingly human-like AI.
Mare Rosa, CEO and CTO of GoodAI said: “We have a strong relationship with Cross Labs and many of their researchers joined our Meta-Learning & Multi-Agent Learning Workshop in August last year. The ideas laid out by Lisa and her team show great promise for progressing our Badger architecture towards open-endedness and we believe this collaboration could have major impacts on the shape of AI in the future.”
About Cross Labs
Cross Labs is an AI research institute focused on fundamental research towards a thorough mathematical understanding of all intelligent processes observable both in nature and in artificial environments in order to spur innovation in machine intelligence. Cross Labs was founded in 2019 by leading Japanese AI company, Cross Compass, with the goal of bridging the divide between intelligence science and AI technology at the service of human society.
