The science behind the AI-powered FIA Driver Safety Index: independent validation

Independent expert validation shows that the AI behind the FIA Driver Safety Index can confidently measure driver risk across geographies and contexts, providing organisations with a credible, data-driven foundation for safer decision-making.

Trust matters in road safety. When organisations use a tool to measure driver risk, and base operational decisions, investment, and ESG reporting on the result, they need to know the measurement is credible.

Committed to operating its Driver Safety Index as a neutral, trusted, scientifically grounded benchmark for driver risk worldwide, the FIA has supported an independent expert validation of the AI technology behind the FIA Driver Safety Index (FIA DSI). The work was conducted by Anders Arpteg Ph.D., Artificial Intelligence, a leader in data and AI research. The validation was commissioned by Greater Than, the FIA’s partner powering the AI technology of the FIA DSI and conducted independently of both organisations.

Putting the AI to the test

The validation evaluated whether the AI used to generate FIA DSI scores genuinely measures the human factor in driver risk. It followed a rigorous scientific methodology: training and validation datasets were strictly separated, exposure weighting was removed, and the results were compared against real-world crash outcomes across multiple geographies.

Three findings stand out

1. 1. The AI identifies risk without being told where to look.
      The model identifies high-risk periods without time, date, or location data inputs to influence the predictions. It detects these patterns from movement alone. That is a strong indicator of scientific robustness: the AI is not following clocks or maps, it is sensing the underlying behavioural signatures of risk.
   2. It measures risk directly – and in advance.
      Conventional telematics relies on indirect signals such as harsh braking or sudden acceleration to assess crash risk. The AI behind the DSI is trained on more than two decades of real driving data linked to actual crash outcomes — making its insights a direct measure rather than an inference from proxies.
   3. It works across contexts, everywhere.
      The model showed strong correlations with severe crash outcomes across diverse geographies. That cross-context performance is what makes a global benchmark credible: a single driver risk assessment score that means the same thing in different markets, sectors, and vehicle types.

What does this mean for users of the FIA DSI?

For organisations wanting to use the FIA DSI to measure driver risk, this validation provides:

• Confidence that the FIA DSI score reflects real risk, and not just patterns in the data.
• Evidence to share with regulators and auditors, and support ESG reporting frameworks that the underlying methodology is scientifically validated.
• A stronger basis for taking proactive safety decisions, focusing on what truly drives risk.

Statement by Anders Artpeg, PhD, Artificial Intelligence, Independent Validator

“At the request of Greater Than, I conducted an independent technical validation of this report which evaluates the performance of Greater Than’s Artificial Intelligence (AI) when used as a global measure of driving risk.

I found the most compelling evidence of the scientific rigor behind the AI to be its ability to identify high-risk periods, such as late-night weekends, without being granted access to any temporal or environmental data. By analysing pure movement patterns alone, the model successfully detects the intrinsic behavioural signatures of elevated risk. This “blind” correlation to real-world seasonality proves that the AI is not simply following a clock or a map, but is sensing the physical reality of driver behaviour.

This predictive power is rooted in a fundamental departure from the telematics industry’s reliance on “proxy indicators” like harsh braking. Because actual crash outcomes are statistically rare events, effectively training a model requires a massive, multi-year global dataset to achieve true convergence. Greater Than’s architecture leverages this unique data moat to establish a robust and generalizable measure of relative driving risk.

The findings demonstrate exceptional sensitivity to crash severity, with particularly strong correlations observed for fatal and serious outcomes across diverse geographies. By strictly separating training and validation datasets and removing exposure weighting, this report confirms that the underlying AI effectively isolates the human factor in driving risk. It provides a
scientifically sound, globally scalable benchmark that supports proactive safety interventions where traditional, retrospective reporting fails.”

Biography of Anders Artpeg, PhD, Artificial Intelligence, Independent Validator

As VP of AI & Data at Saab Group, Anders Arpteg’s mission is to scale value from AI & Data to keep people and society safe. With a Ph.D. in AI and over 25 years of experience spanning academia, national security, and global tech leaders, Anders has a passion for leadership and for scaling real value from AI within organizations and for the betterment of our society.

Throughout his career, from heading AI research at Spotify and Peltarion to serving as Director of AI & Data for the Swedish Security Service and GlobalConnect, Anders has focused on building AI solutions and a culture that enables AI to thrive at scale. He is a firm believer in the power of ecosystem-building, having helped launch national initiatives like AI Sweden and the Swedish AI Agenda.

Beyond the office, Anders remains connected to the global research community as a reviewer for the International Conference on Machine Learning (ICML) and the International Conference on Learning Representations (ICLR). He explores the future of the field as the host of the AI After Work (AIAW) podcast and a recipient of the Data, Analytics and AI Readiness (DAIR) Lifetime Achievement Award.