From humanity's existential risks to safety risks in critical systems to ethical risks, responsible AI, as the saviour, has become a major research challenge with significant real-world consequences. However, achieving responsible AI remains elusive despite the plethora of high-level ethical principles, risk frameworks and progress in algorithmic assurance. In the meantime, software engineering (SE) is being upended by AI, grappling with building system-level quality and alignment from inscrutable machine learning models and code generated from natural language prompts. The upending poses new challenges and opportunities for engineering AI systems responsibly. This talk will share our experiences in helping the industry achieve responsible AI systems by inventing new SE approaches. It will dive into industry challenges (such as risk silos and principle-algorithm gaps) and research challenges (such as lack of requirements, emerging properties and inscrutable systems) and make the point that SE is the linchpin of responsible AI. But SE also requires some fundamental rethinking - shifting from building functions into AI systems to discovering and managing emerging functions from AI systems. Only by doing so can SE take on critical new roles, from understanding human intelligence to building a thriving human-AI symbiosis.

@inproceedings{Zhu23a,
  title = {Software Engineering as the Linchpin of Responsible AI},
  author = {Liming Zhu},
  booktitle = {ICSE},
  year = {2023},
  doi = {10.1109/icse48619.2023.00012}
}