Artificial Intelligence that can Dream: How Creative Algorithms can help Users Learn, Work & Live
Artificial intelligence systems today, whether based on artificial neural networks, deep learning networks, or higher order cognitive systems such as IBM’s Watson, operate in a reductionist mode of determination; the systems seek to derive an answer from the available set of learned or trained information. Artificial intelligence systems need to move towards considering not just the most probable answer, but also the less likely alternatives, of which may include a preferable answer than the most likely one. For this to occur, an AI would need to dream or create new alternatives previously not yet seen, while still having some basis in reality. The grand challenge we propose is to design an AI that can dream: that is, to derive an algorithm that is not reductionist, but creative, showing increasingly new associations with additional information that is resilient to overtraining.
Dr. Terence Yeoh is an assistant principal director at The Aerospace Corporation, overseeing the technical and programmatic details of a systems of systems program. Prior to this, he served in the company’s research and development arm as the R&D Portfolio Manager, giving him broad insight into Aerospace’s innovation portfolio. He joined Aerospace as a staff scientist in 2003.
Yeoh is the cofounder of SeedTECH, an Aerospace community of interest comprising more than 160 Aerospace employees developing technology for the greater good on a volunteer basis. Yeoh is also the team lead for SeedTECH AI, an Aerospace team currently participating in the IBM Watson AI XPRIZE focused on applying artificial intelligence systems to dream new alternatives to today’s problems.