Foundational Research
The behavioral science behind millionways
Research origins
The foundation of millionways was established through a longitudinal, multi-year research program involving:
-
Collaboration with 50+ psychologists and behavioral scientists
-
Engagement with more than 7,000 participants over extended periods of time
-
Millions of carefully labeled behavioral data points drawn from text, interviews, and real-world decision contexts
Crucially, this data was not collected as isolated snapshots. Participants were observed across time, situations, and pressure regimes. This longitudinal approach allowed the research to distinguish enduring behavioral structure from momentary reactions, noise, or adaptation.
The central question guiding the work was simple but demanding:
What remains consistent in behavior as conditions change and pressure increases?
Many existing approaches to behavioral analysis focus on surface signals: sentiment, keywords, personality labels, or outcome-based inference. These methods tend to break down in high-stakes environments, where people adapt quickly, signals are noisy, and outcomes lag decisions by months or years.
The research behind millionways demonstrated that:
-
Meaningful behavioral signals often appear well before measurable outcomes
-
Decision framing and communication patterns carry more signal than surface sentiment
-
Consistency and inconsistency across time are more informative than isolated observations
These findings shifted the focus away from explicit word meaning alone and toward implicit behavioral structure: how decisions are framed, how uncertainty is handled, how constraints are negotiated, and how patterns persist or fracture over time.
From observation to structure
Thorsten: a Large Psychology Model
Thorsten is millionways’ proprietary Large Psychology Model (LPM), now in its fourth generation.
Unlike language models designed primarily to predict the next token in a sequence, Thorsten is designed to interpret behavioral structure across sequences of decisions, communications, and contexts. It analyzes behavior implicitly, not by assigning meaning to individual words in isolation, but by modeling how behavior unfolds over time under constraint.
At a high level, Thorsten is built to:
-
Analyze behavior across time, not snapshots
-
Identify stable patterns as conditions shift
-
Separate signal from adaptation and noise
-
Remain explainable and interpretable by design
Thorsten does not infer intent, diagnose individuals, or predict outcomes deterministically. Its role is interpretive: surfacing behavioral patterns that are often invisible to traditional analysis.
A Large Psychology Model differs from other AI systems in several important ways.
Structure over surface expression
Thorsten focuses on how behavior is organized and expressed across situations, rather than labeling fixed traits or reacting to sentiment.
Contextual grounding
Behavior is always interpreted relative to incentives, constraints, and pressure. Thorsten models these dynamics explicitly rather than treating behavior as decontextualized text.
Explainability by design
Insights generated by Thorsten are traceable to observable behavioral patterns over time, not opaque correlations built on probability-based token generation engines.
This makes the model suitable for high-stakes decision environments where understanding why matters as much as what.
What makes a Large Psychology Model different
Research, applied
While the foundation of millionways is research-driven, its orientation is practical.
The same behavioral structures identified in early research now power analysis across:
-
Finance and investment decision-making
-
Leadership and management evaluation
-
Recruiting and succession planning
-
Risk framing and strategic judgment
Research at millionways continues, but always in service of real-world decisi
Behavioral intelligence is not static. As decision environments evolve, so does the research.
millionways continues to refine Thorsten by:
-
Expanding longitudinal behavioral datasets across domains
-
Validating patterns in new decision contexts
-
Improving interpretability and signal quality
-
Stress-testing insights under real-world pressure
This ensures the foundation remains rigorous, relevant, and grounded.
Ongoing work
Why this foundation matters
In complex environments, better decisions require more than faster data or larger models. They require understanding how people actually behave when it matters.
That understanding, and the research behind it, is what millionways and
Thorsten at its core are designed to provide.