Seldon - Behavioral modeling research & technology
Human behavior in changing environments.
Seldon is an applied AI research lab building models of human behavior. We simulate how people perceive, decide and act in changing environments, enabling organizations to test products, systems and strategies before deploying them.
Four behavioral tendencies: Direct, Cautious, Curious and Social.
Two people in the same situation won't behave the same.
Our research focuses on representing these differences computationally, creating agents with distinct behavioural identities rather than assuming everyone thinks and acts alike.
What drives their behavior?
We model individual characteristics, goals and motivations that influence how different people respond to the same situation.
What do they know?
From first-time users to domain experts, agents bring different levels of knowledge, experience and memory into each interaction.
How do they feel?
Trust, frustration and impatience aren't fixed. We explore how emotional states evolve through an experience and influence subsequent decisions.
Don't just observe what happened. Understand why.
Seldon captures how agents move through an environment, the information they encounter, the decisions they make and the outcomes that follow.
By analyzing these behavioral sequences, teams can identify points of friction, explore alternative scenarios and understand which conditions lead to different outcomes.
- 01
Trace the behavior
Examine the sequence of information, decisions and actions that led to a particular outcome.
- 02
Explore what changes.
Compare different conditions, agent behaviors and scenarios to understand where outcomes diverge.
- 03
Turn patterns into insight.
Identify recurring behaviors, underlying causes and opportunities to improve products, systems and strategies.
Our research explores how to evaluate simulated behaviour against real-world observations. The goal is to build simulations that organizations can interrogate, test and improve, rather than treating AI-generated outcomes as unquestionable predictions.
Building the foundations of behavioral simulation.
Simulating human behavior requires more than powerful language models. It requires systems that understand the environment, represent individual differences, interpret information, make decisions and act on them.
Our research spans six interconnected areas.
- 01
Behavioral identity
Creating agents with distinct traits, goals, knowledge and experiences. Our research includes extracting behavioral characteristics from observed interactions and modeling memory across experiences.
- 02
Environment intelligence
Understanding the environments agents operate in, from early design concepts to functioning applications, including their constraints, complexity and possible paths.
- 03
Perception & attention
Modeling what people are likely to notice, overlook or find confusing, and how visual complexity and competing information influence their behavior.
- 04
Cognition & emotion
Developing specialized behavioral models that explore how people interpret information, make decisions and respond to changing emotional states such as trust, frustration and impatience.
- 05
Autonomous interaction
Translating decisions into real actions. Our agents navigate complex digital journeys, adapt to unexpected states and interact with authenticated environments.
- 06
Simulation intelligence
Analyzing simulated behavior to identify patterns, understand why outcomes occur and generate recommendations informed by organizational and domain context.
How closely does a simulation reflect real behavior?
We evaluate our models against observed human behaviour, studying where decisions align, where they diverge, and how people and agents respond to changing circumstances. By measuring these differences, we can understand where simulation succeeds, and where there are limitations.
Essay · Tom Charman · May 12, 2026
What Behavioral Fidelity Actually Means - And Why It Matters
Behavioral simulation only becomes useful when it faithfully reproduces how real people make decisions. Behavioral fidelity is the measure of that trust.
The technology in use
Finch, Wren, and Rook apply Seldon's behavioral technology to product experiences, software reliability, and decisions in complex scenarios.
| Product | Main question | What the team examines |
|---|---|---|
| Finch | How will different people experience your product? | Simulate user journeys to uncover friction, test designs and understand how different people respond before launch. |
| Wren | What will break when people use your software? | Explore complex application journeys, uncover unexpected failures and identify problems that scripted tests may miss. |
| Rook | How might people act when the situation changes? | Explore human decisions, interactions and alternative outcomes in complex scenarios, including defense simulations and wargaming. |
The future doesn't have to be a guessing game.
Explore how behavioral simulation could help your organization understand decisions, test assumptions and prepare for what comes next.