The GPS-and-spatial-memory finding is the cleanest single case study in cognitive-offloading research. The mechanism is direct, the measurement is relatively clean, the practical advice is clear — and the same logic extends naturally to AI tools doing more of the everyday thinking.
What the UCL study actually showed
In 2020, Louisa Dahmani and Véronique Bohbot published a study in Scientific Reports (Nature) titled "Habitual use of GPS negatively impacts spatial memory during self-guided navigation" (open access at Nature.com; Scientific American secondary coverage).
The study tested participants on self-guided navigation tasks — wayfinding without GPS — and measured how their performance correlated with their lifetime GPS-use history. Two findings:
- People with greater lifetime GPS use had measurably worse spatial memory during self-guided navigation. The effect was specific to spatial memory, not general cognition.
- Follow-up tracking suggested the relationship was likely causal in the GPS-causes-decline direction, not the reverse. Heavier GPS use predicted subsequent decline in hippocampal-dependent spatial memory.
The mechanism, well-documented in subsequent fMRI work, is direct. Self-guided navigation engages the hippocampus and prefrontal cortex actively — building a mental map, choosing among alternatives, predicting paths. Following GPS turn-by-turn directions through the same route doesn't engage those systems in the same way. The cumulative effect is measurable spatial-memory degradation.
What it doesn't show
The study does not show that GPS use causes general cognitive decline. The effect is specific to spatial memory and spatial cognition. It does not show that GPS users have worse memory for anything other than the spatial information they delegated to GPS. The effect is narrow and task-specific.
The study also does not recommend that anyone stop using GPS. GPS is genuinely useful, especially in unfamiliar areas. The cost is real and contained.
Why this matters for the AI question
The GPS finding is the most cited piece of evidence in the broader cognitive-offloading argument because the mechanism generalises. The 2011 Sparrow Google-effect work (Sparrow et al., Science) showed the same pattern for fact retrieval. When participants believed information would be available online later, they remembered the information itself less reliably and the location of the information better.
Risko & Gilbert's 2016 review in Trends in Cognitive Sciences framed the broader pattern: offloading shifts where cognitive effort lands, and skills that are not exercised tend to be less practised over time.
AI tools are the next layer of the same pattern. Drafting with an AI assistant is to writing what GPS is to navigation — the tool produces the result, the underlying skill gets fewer reps. The MIT Media Lab cognitive-debt preprint (Kosmyna et al., 2025) is one recent empirical hook, with the Stanković 2026 critique flagging methodological concerns.
The GPS-finding has decades of natural experiment behind it. AI-and-cognition has years. The pattern fits; the long-term implications are still being worked out.
What to actually do about GPS
The practical advice across the published research is the same: use GPS deliberately, not abandon it.
- Navigate familiar territory without GPS sometimes — your usual route to work, the shops near your home
- Pay attention to landmarks even when GPS is on
- Before a new trip, look at the map first to build a mental picture, then use turn-by-turn
These are spatial-memory exercises in the same way Senwitt's daily Set is a thinking-skills exercise. The point is not abstinence from the tool — it's preserving the daily existence of the underlying skill.
What it means for AI by analogy
The same playbook works for the AI version of the question:
- Use AI deliberately, not by default
- Do at least one cognitive act per day without AI — write a paragraph, do an estimation, read an article fully
- Notice the difference between using AI for the work and letting AI do the work
For the longer treatment of the AI side, see the research/cognitive-offloading page and the research/ai-overreliance page. For the GPS-specific version, the research/gps-and-spatial-memory page is the most detailed treatment.
What the UCSB extension added
The 2024 UCSB follow-up work extended the UCL finding to additional cohorts and additional navigation paradigms. The replication is what made the GPS-and-spatial-memory result something more than a single-lab finding. The pattern held across age cohorts, across geographic settings, and across the specific spatial-memory tasks used to measure the effect. The extension also let researchers probe the moderators more carefully. The strongest correlate of preserved spatial memory in heavy GPS users was the maintained presence of regular unmediated navigation — people who used GPS heavily but also walked their familiar neighbourhood unmediated, or who drove without GPS on routes they knew, showed less of the spatial-memory degradation than people who had effectively delegated all wayfinding to the device. The finding matches what the broader skill-disuse literature would predict. The mechanism does not punish GPS use specifically; it punishes the absence of the unmediated practice. The implication for AI is structural and direct. A workday that runs entirely through AI without any unmediated cognitive practice produces the same pattern the GPS data shows. A workday that includes a small bounded space of unmediated practice — the seven-minute Set, the morning routine, the no-AI block — preserves the underlying surface even at high tool-use intensity. The defence is not less tool use. It is regular unmediated practice alongside the tool use.
What a daily navigation practice actually looks like
For readers who want a practice routine specifically on the spatial-memory side, the literature supports a small set of concrete behaviours. The list is shorter than the equivalent list for writing or reasoning because spatial memory is a narrower cognitive surface and the daily ask can be correspondingly smaller. One: walk one familiar route a week without GPS or phone navigation, paying attention to the corners and landmarks rather than just the turn-by-turn instruction. Two: when you do use GPS in a new area, look at the overview map for thirty seconds before starting the route so the trip has a mental shape rather than being a sequence of instructions. Three: before turning on GPS for a journey you have done before, try to remember the route in your head first. Four: pay attention to where the sun is, where landmarks sit relative to each other, where the river or the railway line runs through your area. None of these require dedicated time. They are recalibrations of existing trips. The cumulative effect over months is the preservation of the spatial mental map that the UCL and UCSB work measures.
A note on the cognitive-aging case
The GPS work matters disproportionately for the over-60s cohort because the spatial-memory system is one of the cognitive surfaces most strongly associated with maintained cognitive health across aging. The hippocampus is one of the structures most consistently flagged in the cognitive-aging literature, and the published evidence on what maintains hippocampal-dependent memory function points in the same direction as the GPS finding: regular use of the underlying capability, embedded in real navigation rather than in artificial exercise, alongside the broader lifestyle factors of physical activity, sleep, and social engagement. The implication is not that GPS users in their 60s are at heightened risk. It is that the daily-practice question that GPS raises is a useful template for thinking about the broader AI-and-cognitive-aging question that is coming over the next two decades. The cognitive-offloading mechanism is the same; the practical defence is the same; the literature base on GPS is the most mature reference point we have for what the broader pattern is likely to look like.
The end-of-decade view
A final framing worth carrying out of this post. The GPS adoption curve in 2026 is roughly fifteen years past the inflection point. The cohort that first adopted smartphone GPS heavily in the late 2000s is now the cohort that has lived with it long enough to show measurable effects. The AI adoption curve in 2026 is roughly three years past its equivalent inflection point. The cohort that adopted AI tools heavily in 2023 will, by the early 2030s, be at the equivalent stage in the offloading curve. The published research will catch up. The cultural narrative will probably oscillate between the alarmist and the dismissive readings, as it has done for every previous technology adoption of similar scale. The practical advice will not change much. Use the tool deliberately. Keep the daily unmediated practice. Pay attention to which cognitive surfaces are still getting reps and which are not. The GPS literature is the cleanest reason to believe this advice will hold up better than either of the louder readings, and the cleanest reason to start the daily practice now rather than wait for the AI version of the UCL paper to be published.
