Trust and AI with Parinaz Sobhani - TWiML Talk #208

Transparency

Transparency is a cornerstone of trust in AI, as it helps users understand how their data is used and decisions are made. Parinaz Sobhani emphasizes the importance of making data processes clear and interpretable, which allows users to combine machine insights with their own knowledge for better decision-making 1. She notes that transparency involves not just data usage but also interpretability and privacy, especially in sensitive applications like healthcare 1.

It's not a secret anymore that machine learning models can memorize data, and if you use sensitive data, these models can remember the training instances.

--- Parinaz Sobhani

By ensuring secure computation and data encryption, organizations can maintain trust even in the face of potential data breaches 1.

   

Building Trust

Building trust in AI systems requires strategic organizational structures and roles. Parinaz Sobhani discusses the need for interdisciplinary collaboration, involving product managers and potentially new roles like trust officers to ensure AI systems are trustworthy 2. She highlights the challenge of accessing protected attributes in regulated industries, suggesting solutions like APIs to test models for biases without compromising privacy 3.

It's really important to engage those people if you have such questions. Like the Google problem, right?

--- Parinaz Sobhani

Incorporating trust metrics into performance evaluations can further embed trust into the organizational culture 2.