A team of researchers, engineers, and designers launched PerXai in 2017 inspired by the global need to address rising information-uncertainty and general societal angst from the rapidly proliferating mis- and dis-information campaigns.
Our primary goal was to empower users to easily learn the true source and reliability of the content they were consuming. We knew we wanted to build a system from the ground up with a purposeful eye on how we created our products and built long-term trust with our customers. We set out to build the first privacy-first AI-assistant trained on your own data, with algorithms controlled by you.
We started by first articulating and codifying our Responsible AI Principles, which guided all of our design and engineering decisions.
1. Clearly articulate system optimization priorities (recall, precision, etc.)
Machine goals | human goals | business goals
Agreement on “noise"
2. Controls + familiarity
Balance predictability & serendipity
Build trust
3. Data is the outcome is the input.
Use the right training data: You are what you eat. You perform how you practice. (You get the idea.)
Give people control over their data.
4. Communicate consequences of actions and costs of failure
Define expected outcomes
Mitigation strategies
Edge cases (including emergencies)
Degrade gracefully
Train machine and person what and what not to ignore
5. Explain results
Communicate confidence
Suggest near-alternatives
6. Facilitate learning
Help users grow
7. Know what not to automate
8. Align to human needs
Understand your place in their workflow
Understand your place in the system
9. Demonstrate commitment to desired outcomes