A new way to understand behaviour from eye movements
At Cineon, our research is focused on understanding human behaviour through eye movements.
Much of that work centres on cognitive state: identifying indicators of stress, workload and fatigue to help improve human performance. But sometimes research uncovers insights that create opportunities of their own.
Recently, while developing and testing our behavioural models, we discovered just how accurately eye-tracking data can identify the task a person is performing. What began as a way to validate our modelling approaches has become an exciting area of our current research.
Behaviour leaves a signature
Different tasks create different patterns of eye movement.
We have long known this to be true. Someone reading a document will move their eyes differently compared to someone engaged in conversation. Likewise, the visual behaviour associated with typing, scanning information, or watching a presentation is distinct.
What surprised us was the strength of the signal.
Using eye-tracking data alone, without video footage or environmental context, our models were able to identify what a person was doing with a high degree of accuracy. In some scenarios, we have achieved accuracy levels of over 95% when distinguishing between closely related activities. This includes tasks that, on the surface, appear similar, such as watching an online meeting versus actively presenting during one.
The finding reinforces an important principle: behaviour is visible through the eyes.
More than a research milestone
For our team, this was initially a useful way of testing the quality of our data-processing pipeline and machine-learning models.
However, the implications go much further.
If eye movements can reliably identify behaviour, a wide range of applications become possible without relying on cameras, environmental recordings, or manual observation.
This creates opportunities to understand activity in a way that is both efficient and privacy conscious.
Turning discovery into application
We are already exploring several ways this capability can be applied in real-world environments.
Supporting healthcare research
Enhancing assessment integrity
Understanding complex activities
Enabling privacy-first wearable technology:
We have received clinical sponsorship from a large NHS Trust to investigate whether patterns of activity detected through eye movements could contribute to earlier identification of cognitive decline.
We are beginning work with a client exploring whether behavioural patterns can be used to identify candidates whose actions do not match the expected task during web-based assessments.
In operational environments, task identification has the potential to provide detailed insight into how activities unfold over time. For example, within aviation, eye movements could help identify different stages of a flight or procedure, creating a richer understanding of behaviour and performance.
Perhaps most importantly, this work offers an alternative to the growing reliance on outward-facing cameras. If a device can understand what someone is doing from eye movements alone, valuable context can be gathered without capturing the surrounding environment. As wearable technologies become increasingly common, that has significant implications for privacy and user trust.
What comes next?
Our primary mission remains unchanged. We continue to focus on understanding cognitive state and developing technologies that improve performance, productivity and health.
But this discovery has opened the door to new possibilities.
Today, we can distinguish between broad categories of behaviour and increasingly specific tasks. The next step is to refine these models further, enabling us to identify a wider range of activities and deploy more targeted behavioural insights.
Discoveries like this are particularly rewarding – they remind us that progress is not always about finding the answer you were looking for. Sometimes it is about uncovering a new question worth exploring.
And this is one we are very excited to pursue.
