Safe Intelligence, a pioneer in deep validation artificial intelligence (AI) systems, today announced that it has secured £4.15m in seed funding in an investment round led by Amadeus Capital Partners, with participation from new investor OTB Ventures and existing investor Vsquared Ventures.
A spin-out from Imperial College London, Safe Intelligence provides a new way to carry out deep validation of AI models, allowing users to detect fragilities that could cause critical failures at run time. The Safe Intelligence solution then applies “robustification” processes to smooth out flaws, making models much more resilient in real-world deployments.
The need for this solution is urgent as the ability to build AI models has rapidly outstripped the capability to determine whether they will operate robustly when used in production. The more sophisticated a model is, the more unpredictable its output, and the greater the risk it contains fragilities that can cause operational failures. This creates extreme challenges in high-safety domains such as finance, mobility, robotics, and aviation where model errors can cause significant losses or severe damage. A model’s performance may also need to be checked for fairness or other industry-specific operational benchmarks.
There are a wide range of ways in which Safe Intelligence helps across sectors. In financial services, for example, models routinely make high-stakes risk analyses to support lending decisions. However, the more sophisticated the models become, the higher the risk that unexpected combinations of inputs by applicants could swing the model’s results, potentially allowing widespread issues with inappropriate lending decisions. Safe Intelligence makes it possible to verify that cases would resolve correctly even in the face of such input variance. The solution gives users the ability to build up their own library of input disturbances and discover new ones.
The company offers command-line tools and a platform for formal verification and robustification at scale. The solutions apply to a wide range of neural network and decision tree models in vision and tabular data. Some of these techniques will also apply to language models at a later stage.
Safe Intelligence also announced today that it has opened an early access program for companies in these high-safety industries to address their validation challenges. The program provides access to tools, model validation, and hands-on support.
Safe Intelligence was founded by CTO Alessio Lomuscio, RAEng chair in Emerging Technologies and Professor of Safe Artificial Intelligence at Imperial College London. Over the past 25 years, he has co-authored more than 200 research papers and holds many patents on AI and its safety, attracting direct funding from the US, the UK, the EU, and private companies. He is a distinguished ACM member and a fellow of EurAI.
Newly appointed CEO Steven Willmott previously led product and engineering for the company. Prior to Safe Intelligence, he was the founder and CEO of 3scale, which became one of the leading API management infrastructure companies (acquired by Red Hat). He holds a Ph.D. in AI and co-led research for one of the leading AI labs in Europe at UPC in Barcelona, Spain.
