AI technology and tools are far outpacing AI product design and experience. This presentation attempts to highlight the risks of poor AI product design and propose a set of principles for teams to move towards good/responsible AI systems.
Note: The presentation was first made at GDG-DC AI/ML meetup at Capital One, McLean, VA on 02/28/2019.
28. TCAV (Google Brain) – ATranslator for Humans
Testing with Concept Activation Vectors (TCAV) is a new interpretability method to understand
what signals your neural networks models uses for prediction.
I wanted to start off with a quick personal story …
The breadth and depth of the AI toolkit continues to grow by leaps and bounds.
Good AI systems demonstrate a high degree of explainability. Accuracy is great but it cannot be at the expense of explainability. AI systems cannot be “my way or the highway”! You will lose trust with your customer base, you will see churn/attrition, you will see loss in engagement, you will see loss in revenue over time.
In order to start start moving towards human-centered AI system designs, we have to put aside the “methods”, and instead start focusing on “principles”. In other words, we NEED TO TAKE A STEP BACK.
We need to think deeply about how the technology we work on today looks in 5 years, in 10 years. Remember that humans evolve slowly, with time to correct for issues they observe in their interactions with the environment. In contrast to that, AI is evolving at an incredibly fast pace and that means that it really matters that we think about this carefully right now, that we reflect on our own blind spots, our own biases, and think about how it affects the technology we are creating. And discuss, what technology of today will mean for tomorrow. You must share your experiences with this technology, what is working, what is not working. What aspects are more beneficial, what aspects are more problematic. That is how we will create the awareness, priority, and urgency to start designing AI products that are fair, responsible, and human. Thank you!