Anupam Datta, a longtime professor of electrical & computer engineering and computer science at Carnegie Mellon University, had a deep interest in AI, and took a skeptic’s point of view. “From the start, I had concerns about issues like bias and fairness,” he recalls. “I knew as a computer scientist that the training data sets available would absolutely lead to bias in models, and no one was doing anything about it.”
Beginning in 2014, Datta, along with his graduate students, performed thorough research on the topic, and published a well-known study in 2015 on their discovery of extensive gender bias in online advertising. He and his then graduate student Shayak Sen grew increasingly frustrated at the private sector’s inability – or unwillingness – to develop tools for preventing AI bias.
In 2015, the duo began to focus more intently on AI Explainability research – the science of understanding how AI models work. AI and Machine Learning (ML) models are known for being black boxes – even the data scientists who create them don’t know exactly why they work. AI Explainability techniques can help parse out exactly which inputs drive a model’s performance, whether unfair bias is present, and how scientists might mitigate that bias.
Starting in late 2017, Datta and Sen spent over a year discussing their research with colleagues in the private sector, and their reactions were revealing. “We realized that adopting AI responsibly by tackling problems of explainability, unfair bias, and related topics had risen to become a top-of-mind concern,” says Datta.
The two teamed up with Will Uppington, an enterprise AI and tech veteran whom they met through mutual friends in late 2018, and founded AI Lens (later renamed Truera), with a mission to help companies adopt AI responsibly. The company’s technology platform, based on the research Datta and Sen performed at CMU, helps companies evaluate and improve model quality during development, select the right models to put into production, and monitor them to make sure they still perform as designed.
Shortly after its formation, the company landed a pilot with global bank Standard Chartered. Several other pilots with Fortune 500 companies – mostly in the financial services and insurance realms – followed. In 2020, the company – now called Truera – announced a seed round, followed quickly by a Series A round, and its customer base has continued to grow. Backers include Greylock and Wing.
“Customers tell us that Truera has become their secret weapon,” says Uppington. “By ensuring the quality of models during development and post production, and guarding against unfair bias, we’re helping them adopt AI responsibly.”
Sign up for the free insideAI News newsletter.
Join us on Twitter: @InsideBigData1 – https://twitter.com/InsideBigData1
