In yet another insightful session of RAISE2020, a panel of industry stalwarts gave perspective on the hottest topic in Machine Learning, Explainable AI. The session had a diverse mix of speakers, which enriched the discussion and provided a holistic view. 

The keynote speaker of the session Ms. Lan Guan, (Global MD for Applied Intelligence at Accenture), gave her views on how vital Explainable AI is in today’s times. She said, “We need to curb this notorious opaqueness in the way AI algorithms churn out results.” Lan stressed the fact that if we want to deal with issues like racial injustice or issues related to minority businesses, we have to work on making the AI and ML working completely transparentelse we might lose people’s faith and trust. As a solution, she said our direction of work should be in retraining out older systems to make them more accountable, transparent, and explainable.

The moderator for the evening, Ms. Arati Deo (MD Accenture Technology India) talked about the paradox around the scenario where undoubtedly accuracy has increased; however the ability to explain has also become complex. 

Head of AI at LinkedIn, Dr. Rushi Bhatt talked about an important task of adopting explainability on the model level, instance level, and even at the individual level. He said, “this way we can resolve misclassification, diagnose problems, and identify out of the distribution instance.”

In a great example of how AI is improving customer experience Mr. Rohan Palaha (Director ML, Amazon), explained how his team used advanced and fair ML approaches in resolving language issues while launching in India. With Amazon being English centric, they faced challenges with address finding, tracking, sorting, and differentiating between fake ones and incomplete ones while keeping it all explainable.

Dr. Srayanta Mukherjee (Director Data Science Novartis) discussed the criticality of explainability in healthcare and hence the significance of “semantic linking” in the life science industry.

“Different people at different levels must understand the various functionalities and working of AI algorithms to make it explainable,” said Dr. Anand S Rao (Global AI Lead at PwC). He also pointed out that the biggest challenge here is that what do call fairness, owing to the fact that fairness has many definitions of fairness. 

Academics and AI are still in a very nascent stage in academics, believes Dr. Vineeth N Balasubramanian (Associate Prof. IIT Hyderabad). The panel had a consensus that if we want to make the most out of AI capabilities, we must keep it FATE (Fair, Accountable, Transparent, Ethical).

Going in detail, Prof. Thomas G. Dietterich said that one of the hurdles in using AI in anomaly detection or fraud detection is “false alarms.” 

Few more insights were shared by Ms. Claire Vishik and Mr. Kaushik Royaround changing the conventional processes and software to use AI to its full potential.

The session gave many budding AI entrepreneurs insights around how to make their existing systems more transparent and device new ones on the same lines.