【Session Theme: Trustworthy AI】
Artificial intelligence (AI) for specific tasks had significant advances in recent years. The interactions between human beings and intelligent machines are becoming intimately entwined than ever. However, developing artificial general intelligence (AGI) remains challenging. One of main issues is how AGI can earn people’s trust and show consistent and accepted behavior over time for different occasions. To enable machines that can act in a human-trusted way, EthicsNet issued a challenge entitled “How would you teach AI to be kind?” on HeroX, an open innovation platform. The challenge aimed at initiating a data-driven approach for designing kinder machines. It called for ideas on how to create a machine learning dataset containing the best possible examples of prosocial behavior. In this talk, the speaker will introduce the EthicsNet challenge, the winning solutions and the follow-up progress of the challenge.
The development of AI brings people a lot of exciting imaginations of the future world. However, for us to enjoy the benefits of AI, trust is essential. It is important that intelligent machines can show consistent and socially accepted behavior over time, and can be modified to something more preferable in some ways, before we have them all around us in our everyday life. The answer about how to teach AI to have prosocial behavior remains unclear, but as we know, the amazing learning capability of modern AI is actually through observation of datasets. Therefore, EthicsNet issued a challenge inspired by ImageNet to seek ideas on how to create the best possible set of examples of prosocial behavior as preparation for enabling kinder machines. The challenge was launched on HeroX and five of the submissions were awarded. The winning ideas ranged from a developmentally-situated approach to teaching normative behavior to AI to creation of a dataset of video/image, audio and text cues to enable empathy built-in machines. In this talk, the speaker will introduce the EthicsNet challenge, go through the winning solutions and introduce the follow-up progress of the challenge.
About Steven Wu
Steven Wu is a computational biologist and health IT developer. He is an advocate and active participant of open innovation and a user and developer of open source software.