Robust Intelligence Raises $30 Million for Stress Testing AI Models
An AI startup, Robust Intelligence specializes in helping businesses stress test their AI models in order to ensure they don’t fail. The company announced that it has just completed a Series B funding round that was led by Tiger Global and has managed to raise an impressive $30 million. The Series A round was led by the company’s previous investor named Sequoia, along with Engineering Capital and Harpoon Venture Capital. They were also participants in this oversubscribed round. Yaron Singer co-founded the company, who is a tenured professor at Harvard University. He teaches Computer Science and Applied Mathematics and got together with Kojin Oshiba, his former student, for launching the company.
Singer said that Artificial Intelligence (AI) is an academic endeavor. The professor said that it had been a vision back when he was in grad school and this vision was realized in the span of seven to eight years because the internet, Google and data and data processing came to be. He said that they were now trying to be rigorous where AI is concerned, same as in the case of software development. He added that they were trying to play catch-up with Artificial Intelligence (AI) and it was a completely different ballgame.
Singer also noted that AI can often have unexpected behavior because of its statistical nature. He said that the primary and core mission of Robust Intelligence was to eliminate these AI mistakes. The company accomplishes this by offering users what is known as the Robust Intelligence Model Engine (RIME). It is helpful because its core comprises an AI firewall. This firewall offers protection by wrapping around the AI models of a company, so it will not make mistakes and it continues to stress test the models. Singer said that anyone who had data and an AI model could run a stress test with a click of the button.
He said that they test AI models and the data automatically before the models are put into production and after it as well. Singer said that the goal was to identify any failure modes of a model and also to catch other related issues, such as data drift. The most interesting thing to note about the system is that the AI firewall is also an AI model itself, which predicts if a data point can lead to a wrong prediction. He said that it was one of the most difficult problems that they were trying to solve in machine learning and AI.
John Curtius, the Tiger Global partner stated that he had first come to know of Robust Intelligence’s capabilities in the early development of the company. He said that they had watched the company, as well as its product, grow in the previous year and it had become apparent that their resources could come in handy for Robust Intelligence to change the face of AI reliability. The company’s plans for the funds include expanding its sales operations, but most of it will go to engineering and product.
