AI model testing company Robust Intelligence has raised $30 million in a Series A financing round led by Tiger Global and Sequoia. The company was cofounded by CEO Yaron Singer, an Israeli professor of computer sciences and applied mathematics from Harvard University, and plans expanding its Israel activities. The company has raised $45 million to date.
Robust Intelligence develops an AI Firewall that envelopes the organization’s Artificial Intelligence (AI) model, monitoring and correcting data that causes mistakes and risks in real time. The product inspects the data before it is entered into the AI model using advanced machine learning techniques, alerting and correcting the data in real time, thus preventing the model from erring. Robust Intelligence already has dozens of paying customers and employs 50 developers with advanced academic degrees at the company’s headquarters in San Francisco.
The company works with dozens of Fortune 500 and Fortune 100 companies, including PayPal, Expedia, Tokio Marine (the world’s largest Japanese insurance company), leading medical device and diagnostic companies, government agencies and more. In Israel, Robust Intelligence works with leading companies in finance, insurance and more.
Prof. Singer said, "We wish to enlarge our presence in Israel by recruiting customers as well as talent specializing in software development, algorithms and artificial intelligence. Until recently, the idea of AI Firewall seemed impossible. We invest huge resources in finding people who believe in themselves and have no limits. Israel is the natural place to find such people."
Robust Intelligence’s product enables organizations to minimize the dangers embedded in AI and enables technological teams to focus on system development, rather than debugging and firefighting models in production. The company's AI Firewall wraps around the AI model while monitoring and even correcting input that might disrupt model results in real time. The product is based on a system that learns from tests and monitoring measures related to the AI model it is protecting. These tests include "stress tests" for the AI model, which challenge it with basic errors and exposure to extreme conditions as well and also monitors input and output anomalies for each model.
Prof. Singer said, "The weaknesses and errors of AI models emerge once the product is activated and considering the currently widespread use of AI when making automatic decisions, knowing that the systems are operated without protection is terrifying. Our company’s mission is to eliminate the risks that companies take when they utilize AI systems. We save our customers a great deal of time and money but, more importantly - we protect people against wrong decisions that might derive of AI errors."
As an example, Prof. Singer described a client who provides AI models to banks and financial institutions to identify financial fraud. Upon installing the AI Firewall, they discovered the models were wrong, providing input where the data field for the country from which the transfer originates was in small and not capital letters. In other words, if the data field shows us (underscore) instead of US, the model erred. These errors cause huge financial losses and exposes the financial institutions using the AI and their clients to significant risks. "In this case, the error was unintentional", he explains, After all, the data was compiled from different systems and the format changed. The problem is that even such minor changes can cause AI models to yield unexpected results. Other examples include decisions made by credit card and insurance companies as they rely on AI systems based on statistical learning. In such systems, the data used to train the learning machine might be biased in favor of certain populations, creating a system that reaches decisions that are biased in their favor too. Such system errors could prevent people from getting a loan on a home or, even worse, health insurance."
Prof. Singer, 42, was born and raised in Tel Aviv and served in an elite IDF intelligence unit. He later relocated to the US, where he completed his Ph.D. in Computer Sciences at UC Berkley. In 2011, he founded a startup company that uses AI to analyze networks and joined Google. At Google, he worked on algorithms to accelerate training for AI models in order to enable the tech giant’s products devise more accurate predictions. That is also when he discovered the ease at which AI systems could go out of control by using incorrect or misleading data. As a Professor at Harvard University School of Computer Sciences, his studies focus on the theoretical limits of AI models, which are exposed to errors and on developing advanced optimization methods to correct them.
Prof. Singer founded Robust Intelligence in 2019 with Harvard students with whom he published dozens of articles on the matter. Singer’s team achieved a breakthrough in algorithm planning, helped increase government awareness to the matter and even obtained an early DARPA grant for the study.
Prof. Singer said, "This latest round of investment will greatly accelerate the development of our AI Firewall product and our RIME platform as well as our marketing and sales efforts. Our world is adopting AI technologies at an exponential rate, and is relying on AI to make critical decisions. The problem, however, is that AI models fail frequently due to bad data, and that introduces an enormous risk to businesses, and society at large."
Published by Globes, Israel business news - en.globes.co.il - on December 12, 2021.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2021. .