Human-centric AI
Humans and AI have complementary strengths, e.g., humans excel at thinking about others and empathy, while AI models can process large data and optimize repetitive tasks. Thus, it is attractive to develop human-centric AI that can augment human skills. As state-of-the-art AI methods often achieve strong benchmarking performance, developing such AI assistants appears realistic.
AI still limited by oversimplification
However, the wide application of AI to augment people's ability in science, democracy, and many other domains is hindered by its lack of common sense. A human-centric AI system that can mimic human language skills is not sufficient - effective synergy requires a commonsense understanding of folk psychology, naive physics, cultural values, and multimodal alignment between language and vision. Human-centric AI should possess the ability to understand agent desires, goals, values, and intentions, and leverage this understanding to anticipate agent actions. It must understand context in situations, anticipate the cause and effect of participant actions, and perform counterfactual reasoning. Human-centric agents must understand how signals from multiple modalities come together with each other and with implicit knowledge. Finally, to be sensitive to ethical and moral aspects of interaction, human-centric AI agents need to understand the impact of culture and values on the behavior of the relevant participants.
Development of human-centric AI agents with common sense
Filip Ilievski will advance with this project the understanding of how to develop human-centric AI agents with common sense that can reliably perform mental modeling (e.g., of agents) in human-robot conversations, verify their reasoning using logic or other agents, identify figures of speech (e.g., analogies) in text, explain agent behaviors through cultural values, and reason coherently over visual situations (e.g., in videos). He will elevate his research line on commonsense reasoning and neuro-symbolic AI with novel techniques that fuel hybrid human-AI systems for social good. The project will build on the advances in commonsense reasoning from knowledge representation and reasoning, neural language and vision models, and human-AI teaming.
AiNed Fellowship grants
The AiNed Fellowship grants are part of AiNed programme funded by the National Growth Fund and help Dutch academic knowledge institutions to attract AI-talents who are usually able to choose between several competitive job offers. The focus is on exceptional AI-talents who can be deployed in an AI-discipline, ranging from technology to the social sciences and humanities. This will enable knowledge institutions to make an attractive offer to AI talents. Read the news item of NWO about the five awarded projects here.