Future-Oriented Risk Management & Ethical Design
When using AI, even if all developers are well intentioned, even if there are no bad actors, we run the risk of producing systems that, rather than being helpful, honest, and harmless, turn out to be hurtful, toxic, deceptive, and downright dangerous– Robert Blumofe, EVP and CTO, Akamai
Is AI the new hope? Many believe, yes. This morning, MTLC’s Tech Hot Topic conference, hosted by Akamai technology, Building Ethical AI into Products, unveiled compelling and fresh perspectives on the impact of Ethical-minded AI. From the implications of DeepSeek-R1, the impact of current affairs, and importance of human centered AI design, it is clear major milestones are evolving in front of us.
We began the day with a series of insightful sessions, including opening remarks by Robert Blumofe, CTO of Akamai, and a lightning talk by Dr. Saiph Savage, Associate Professor at Khoury College of Computer Sciences, Northeastern University, who shared her research on human-centered AI design. Liz Graham, CEO of ADA IQ, showcased innovative startup solutions.
Here are key takeaways from the final panel discussion “Future-Oriented Risk Management & Ethical Design,” moderated by Michael Davies, Founder of Endeavour Partners and Senior Lecturer at MIT. The panel explored the challenges and strategies for integrating ethical and responsible AI innovation in various sectors.
Panelists:
-
- Morgan Beschle, Vice President of Product, Kyruus Health
- Sanjay Yadav, Head of Engineering, Parlance
- Shankar Krishnan, Product Management Leader, AWS
What the Experts Shared
Importance of Ethical and Responsible Innovation:
Michael Davies emphasized the significance of ethical design and responsible innovation, highlighting the need to address unintended consequences and negative externalities of rapid technological advancements. He stated, “Technological innovation is the most powerful force of humanity. Every one of you has technical or engineering expertise, and everyone else owes their lives to the work we and our predecessors have done over history.”
Role of AI in Healthcare:
Morgan Beschle discussed the integration of AI in healthcare, particularly in improving patient care access and the challenges of introducing AI in a traditionally risk-averse industry. She mentioned, “AI is just a tool in a toolbox. We start with the business problem, what use cases are you trying to tackle, and how are you going to measure success?”
Bias Mitigation and Prevention:
The panelists acknowledged the inherent biases in AI due to the data used for training models. Shankar Krishnan noted, “The level of determinism in the output you get from AI models is crucial. With large language models, you can have a wide variety of outputs for the same input, which exposes a lot of risk.” They discussed various strategies to mitigate these biases, such as using diverse datasets and implementing fairness constraints.
Human in the Loop:
The importance of having human oversight in AI processes was stressed. This includes having listening analysts or service engineers review AI outputs to ensure accuracy and reliability. Sanjay Yadav explained, “We have introduced human in the loop for our AI models. Listening analysts review the calls to ensure the names are recognized properly, and service engineers validate the answers generated by the AI.”
Risk Management and Failure Handling:
The panelists emphasized the need for robust risk management strategies, including defining acceptable failure thresholds and having fallback mechanisms in place. Morgan Beschle highlighted, “What is the alternative? When we moved to electronic medical records, there were risks, but there were also risks with paper records. It’s about comparing the new solution to the existing one and understanding the risk tolerance.”
Future Directions and Innovations:
The discussion touched on the rapid pace of AI innovation and the need for continuous learning and adaptation. Sanjay Yadav mentioned, “We are looking into ways of using generative AI behind the scenes to drive the conversation flow with the caller, which will help us drive down costs and improve the naturalness of interactions.”
Conclusion
The panel concluded with a Q&A session where audience members asked about bias prevention, the role of interdisciplinary teams in AI development, and the use of synthetic data to address biases. The program ended with a sentiment of inspiration and new information to consider and take back to our companies. Join us for our next Tech Hot Topic conference on March 21st at Slalom, Reshaping the Marketing Organization Around AI!
