Research

Reading and displays

Readability

Does the best format for text differ from reader to reader? Online studies with hundreds of adults found that it does: the font a person reads fastest varies, and the right one can speed reading without costing comprehension. Later studies built ways to test text format online and to recommend a font, and asked whether letter and word spacing matter too, and whether the gains hold for children, for health records and for arithmetic. Recent work turns to font weight, grade and x-height, in Arabic as well as English. The newest asks how visual span, the number of letters taken in at one glance, relates to reading speed in dyslexia, and how to measure it online.

Transportation

What does a driver lose by looking away from the road, and what can design give back? Studies measured the cost of texting by hand, with assisted entry and on a head-worn display, of entering a destination into a navigation system and of taking a breath test for an ignition interlock while driving, and how a voice interface changes mirror checks in older and younger drivers. Others asked what mental load does to memory of a scene and to spotting brake lights, how trust in technology shapes where drivers glance, and whether rare hazards are missed more often on the road, which they were not. Typography studies asked which typefaces can be read in a single glance on in-vehicle displays, and a theory paper set out how drivers take in visual information. Earlier work linked driving simulators into one network to test vehicle-to-vehicle systems.

Driving simulator →Take part in the driving study →

Mixed reality

Do virtual, augmented and mixed reality train people better than traditional methods? A meta-analysis of the published studies found them equally effective, which makes them most useful where danger or cost rules out training the usual way. A study with Air Force Academy cadets asked whether a mixed-reality headset or a tablet better supports multi-domain military planning: the headset added no situation awareness, and the tablet was rated more usable. An earlier virtual-reality study asked how the arrangement of objects in a space creates a sense of familiarity, and how that relates to déjà vu.

Headsets →

Minds and machines

Human–AI interaction

Can large language models write classroom texts that students read as well as texts written by people, and simplify a text to a student’s level without losing key details? Studies compared the readability of AI-written and human-written educational content, and tested ways to make AI simplification more accurate and consistent. Other work asks how automation, autonomy and AI should be designed around the people who use them, and how social robots can handle conflict with people constructively. A 2025 paper asks how AI replicas of sex workers, a kind of human digital twin, can protect the workers’ intellectual property rather than erode it.

Openings in the AI Natives Laboratory →

Cognitive security

Why do people miss the attacks that matter most? Cyber vigilance studies found that threat detection declined over time, rose with the probability of an attack and fell as events came faster, under high workload. In a simulated email workplace, attacks sent at a rate of 1% were significantly more likely to succeed than more frequent ones: a prevalence paradox, since automation that makes attacks rarer can make people worse at catching the rest. Related studies measured how training changes phishing detection and the stress and workload of network analysis. Later work built a testbed for online influence operations and maps the ways human cognition is attacked, and the defenses it needs.

This work is shared with the Cognitive Security Institute, a separate organization whose board Ben D. Sawyer chairs.

Cognitive Security Institute →

Applied neuroscience

Can the brain’s response to a mistake be detected outside the laboratory, with a single electrode? Studies found the error-related negativity in EEG while people searched complex images, and compared ways of cleaning such data. Others asked whether electrodermal potential tracks a driver’s attention, whether tiny eye movements called microsaccades distinguish looking from seeing, and, with EEG, how anger raises risk-taking when a driver decides whether to cross on a yellow light. The most recent work decoded task fMRI data with graph neural networks, taking individual differences into account.

Eye tracking →