A deliberately low-budget test: can AI plus Quantum UX predict real behavior better than traditional polling? The team used a real election as the benchmark.
This is a documented experiment applying AI, statistics and Quantum UX to user behavior prediction, benchmarked against a real election (Argentina’s August 2023 primaries) purely to validate the method. Instead of expensive phone or street surveys, five researchers spent a week tallying candidate mentions in unbiased social-media comments, then fed the counts into a factorial-design model (Python / PsychoPy) using regression, ANOVA and contextual variables plus a qualitative ‘epsilon’ factor. The model surfaced a persistent ‘aberration’, an underdog outperforming the presumed front-runner, that conventional pollsters had missed.
Key takeaways
- Behavior prediction always happens in a context: conditions shape behavior, and behavior adapts to the environment.
- The team used a real election only as a neutral benchmark to validate the methodology, not as political work.
- Low-cost method: tally candidate mentions in unbiased social-media comments across many platforms.
- Model: factorial design + regression + ANOVA, plus variables like age, geolocation, media mentions and a qualitative 'epsilon' (0.1-1).
- The model flagged a persistent underdog surge the standard polls missed, illustrating Quantum UX's real-time edge.
Behavior always has a context
Prediction sits inside a context: existing conditions and environment, and how each shapes the other. Simple case: a user wants a hotdog in a fancy restaurant. Either the restaurant doesn’t serve it and conditions force another choice, or the user reads the setting and adapts before ordering. The same interplay drives any behavioral forecast, which is why Quantum UX emphasizes real-time context over static assumptions.
Why a different methodology
Traditional opinion studies here were limited in case count and geography (mostly the wealthiest city), and their recent forecasts had missed badly. The team wanted the opposite: as many people as possible, as widely dispersed as possible, across diverse demographics, which conventional phone/street/in-home surveys couldn’t reach on a tiny budget. The team also states neutrality explicitly: based in Argentina but not working for it, unpaid by any candidate or outlet, interested only in validating the method.
The method, step by step
User behavior prediction is not an exact science. Quantum UX makes it easier to predict in real time, but not in the long run. That is exactly its advantage and its disadvantage.
Fabio Devin
Forecasting how users will act, always within a context of conditions and environment that shape behavior and are in turn shaped by it.
By collecting unbiased signals (here, candidate mentions in comments) and running them through a statistical/AI model that weighs variables like demographics, timing, media presence and qualitative context.
A qualitative factor between 0.1 and 1 capturing context, national history, economy and experienced judgment (‘gut feeling’), added to the quantitative regression.
Because it excels at real-time prediction by combining many data dimensions, though it’s less suited to long-run forecasting, which is both its strength and its limitation.
No. The election was used only as a neutral, public benchmark to validate the methodology; the team states it was unpaid by and unaffiliated with any candidate.
Want data-driven prediction for your product or audience? Talk to Dorve.
Start a project