Translate reponsible AI requirements into UX
features for implementation
Shipped multiple features such as feedback, onboarding, user education that helped to successfully launch the new Bing
A three months project in collaboration with the Bing team integrating ChatGPT into the new Bing. I was tasked on creating features that mitigated potential Large Language Model(LLM) harms such hallucinations and overreliance.
The new OpenAI powered Bing had one of the most successful launches in Microsoft's history - 2 million people sign up in the first few days and daily active users of Bing grew by 100 million.
A set of principles to help guide my design process for responislbe AI
I ideated and designed multiple features in collaboration with Bing PMs and Engineers. I helped the product team meet the Microsoft responisble AI requirements in order to lauch the new Bing.
First Run Experience - I wanted to create an interactive first run experience that will let people learn about the models capabilities in context.
I created interactive tiles and carefully choose prompts that would let the model showcase it's capabilities. This approach ensured that people gained an intuitive understanding of ChatGPT's capabilities.
In-context Education - While it’s tempting to only rely on the first run experience to educate people, I wanted to provide an in context educational experience.
I designed an interactive help button next to the conversation suggestion chips. By interacting with the help button, people can access conversation chips related to AI education without disrupting the ongoing conversation. This approach ensures a seamless learning experience while keeping people engaged and informed.
Feedback - To create a more comprehensive feedback experience, I designed a system that offers people two distinct options.
First, people can provide feedback directly to the model within the conversation, allowing them to iterate on their results. Alternatively, people can share their feedback with the product team, ensuring continuous improvement.
This dual-feedback approach enhances people's satisfaction and helps drive ongoing refinements to both the model and the overall product.