Building trust in AI models, not through interpretation(XAI) but through interrogation.

Do you know when you should trust your AI system? Or rather when you shouldn’t?

Why your model with 95% accuracy gave weird results in production?

An AI system is as good as the data used to train the system. This talk is for discussion on the same lines.

We are at the dawn of the 4th industrial revolution where Artificial Intelligence(AI) plays a significant role in shaping our daily life. Even with lots of advancement in this field, there are many obstacles like trust, reliability and transparency. This issue has triggered a new topic for research i.e. explainable Artificial Intelligence (XAI). But XAI has its own drawbacks, the more accurate models are less interpretable, and the less accurate models are more interpretable. So to make an AI model more interpretable, we have to compromise with the performance.

So is there a way to improve the trust and reliability of AI models without compromising its accuracy? The answer is “yes”!!! This talk focuses on various techniques and tools to establish trust in AI models.

Get ready to experience the other side of AI !!!!


Outline/Structure of the Talk

  1. What is Trust in AI and why it is important? (3 mins)
  2. Interpretation vs interrogation (2 mins)
  3. How to identify bias in ML models (4 mins)
  4. How to define the region of confidence (4 mins)
  5. Hands-on session (5 mins)
  6. Q&A session (2 mins)

Learning Outcome

After attending this session, the participants will know...

  • What is trust in AI.
  • How to establish trust in AI systems.
  • Leveraging different frameworks to build trust.
  • Different ways to find bias in a model.
  • How to assign confidence to the model's predictions.

So next time when you get 95% accuracy, you know that not all predictions have 95% chances of being correct.

Target Audience

Data Science Consultants, Data Scientists, Data Engineers, ML Engineers, Product Managers, Software Engineers

Prerequisites for Attendees

  • Basic understanding of Machine Learning Algorithms
  • Should be aware of metrics like - R square, RMSE, AUC, F1 score etc.
  • Familiarity with different AI use-cases.


schedule Submitted 2 years ago