A neurocomputational model of observation-based decision making with a focus on trust

This paper presents a neurocomputational model integrating the orbitofrontal, lateral prefrontal, and anterior cingulate cortices to simulate how observing an expert's actions influences an individual's trust and decision-making, illustrated through a climate-related transportation choice scenario.

The Gist

Imagine your brain is a busy, high-tech control center for making decisions. Now, imagine you are standing at a crossroads, trying to decide whether to drive your car or take the bus to work. Usually, you might just go with your gut feeling (maybe you love the comfort of your car) or your logic (maybe you know the bus is better for the planet).

But what happens if you see a famous climate scientist—a "trusted expert"—riding the bus and looking happy about it? Does that change your mind?

This paper builds a computer simulation of your brain to figure out exactly how watching someone else influences your choices, specifically focusing on the role of trust.

Here is the breakdown of their model using simple analogies:

1. The Three Brain Departments

The authors focus on three specific "departments" in your brain that talk to each other when you make a decision:

• The "Gut Feeling" Department (OFC): This is the Orbitofrontal Cortex. Think of it as the emotional manager. It cares about immediate pleasure, comfort, and how things feel. If you see the expert on the bus, this department asks, "Does this look fun? Does it feel good?"
• The "Logic" Department (LPFC): This is the Lateral Prefrontal Cortex. Think of it as the rational CEO. It cares about long-term goals, facts, and consequences. It asks, "Is this the smartest choice for the environment? Is it efficient?"
• The "Social Hub" (ACC): This is the Anterior Cingulate Cortex. Think of it as the Chief Editor or the Meeting Room. It sits between the Gut and the Logic departments. Its job is to listen to both, compare what you expected to happen with what actually happened, and update your brain's "database."

2. The Scenario: The Climate Expert

The researchers created a simulation where an observer (you) watches an expert (a climate professor) choose between a car and a bus.

• The Setup: You start out being a bit emotional. You prefer the car because it feels cozy. But deep down, you know the bus is the "rational" choice for the climate.
• The Observation: You watch the expert take the bus. The expert arrives on time and looks satisfied (a good outcome).

3. How Trust Works in the Model

The magic of this paper is how it explains trust. In the simulation, trust isn't just a feeling; it's a mathematical "learning speed."

• The Prediction Error: When you watch the expert, your brain makes a guess: "I bet the expert will take the car because it's easier." But the expert takes the bus. Your brain registers a "prediction error" (a surprise).
• The Trust Multiplier: If you trust the expert, your brain says, "Okay, this surprise is important! I need to learn from this immediately."
  • Low Trust: If you don't trust the expert, your brain ignores the surprise. "Oh, they took the bus? Whatever, I'll stick to my car."
  • High Trust: If you trust the expert, your brain's "learning speed" turns up. The neural connections in the "Logic Department" (LPFC) get stronger and faster.

4. The "EEG" Dance

The model simulates brain waves (like an EEG).

• Low Trust: The brain waves are a bit chaotic and slow. The "Logic" department is quiet.
• High Trust: As you watch the expert succeed, the "Logic" department starts dancing to a faster, more regular beat (higher frequency). This represents your brain becoming more excited and convinced by the rational choice.

5. The Big Surprise: Logic Wins, but Slowly

The most interesting finding is about timing.

• You might think emotions change fast and logic changes slow. But in this model, rational trust changes faster than emotional feelings.
• Your "Logic Department" quickly updates its value: "Okay, the bus is a good idea."
• However, your "Gut Feeling Department" takes much longer to catch up. It takes a long time for the emotional love of the car to fade away and be replaced by an emotional love of the bus.

The Takeaway

This paper suggests that when we see a trusted expert do something good (like taking the bus for the climate), our logic grabs onto that information immediately. We intellectually understand the value. But our emotions are stubborn; they take much longer to shift.

So, if you want to change people's habits (like getting them to drive less), you need to:

1. Find a trusted expert they respect.
2. Show them the expert succeeding with the new habit.
3. Be patient, because while their brain might agree with you instantly, their heart might take a while to follow.

In short: Trust acts like a turbocharger for our rational brain, helping us learn from others, but our feelings are the heavy anchor that takes longer to lift.

Technical Summary

1. Problem Statement

Human decision-making is heavily influenced by social interactions, particularly observational learning, where individuals learn from the actions and outcomes of others (e.g., an "expert"). A critical factor in this process is trust, which determines whether an observer adopts a strategy based on observed behavior.

While behavioral and social theories explain how people learn from others, there is a gap in understanding the specific neural mechanisms that integrate emotional (affective) and rational (cognitive) processing during trust-dependent decision-making. Specifically, the paper addresses:

• How neural oscillatory dynamics in specific cortical regions encode the transition from heuristic/emotional responses to rational reasoning based on trust.
• How the predictability of an expert's actions and outcomes updates an observer's neural representation of trust and decision values.
• How these processes can be modeled computationally to predict behavioral changes in real-world scenarios (e.g., climate-friendly transportation choices).

2. Methodology

The authors developed a neurocomputational model simulating the interplay between three key cortical structures: the Orbitofrontal Cortex (OFC), the Lateral Prefrontal Cortex (LPFC), and the Anterior Cingulate Cortex (ACC).

A. Neural Architecture

The model utilizes a simplified three-layered attractor neural network for each cortical structure (OFC, LPFC, ACC), consisting of:

• Excitatory neurons (middle layer, 100 units).
• Inhibitory neurons (feedforward and feedback layers, 25 units each).
• Dynamics: Modeled using differential equations to simulate membrane potentials and oscillatory activities. The system incorporates Hebbian learning rules modified by a trust variable ($Tr$).

B. Functional Roles of Neural Structures

• OFC: Encodes emotional/affective values and immediate gratification. It processes late sensory stimuli and generates "expectancy signals" for emotional outcomes.
• LPFC: Encodes rational/cognitive values and long-term rewards. It processes early sensory stimuli and utilizes mirror-like neurons to simulate observed actions.
• ACC: Acts as a social hub. It integrates afferent signals from OFC and LPFC, calculates prediction errors (the difference between predicted and observed action/outcome), and modulates the learning rate of the other two structures.

C. Mathematical Framework

• Dynamic Bayesian Probability (DBP): Used to model the probability of an expert's future actions based on the history of observed action-outcome associations. This replaces static Bayesian approaches to account for sequential dependencies in social learning.
• Modified Hebbian Learning: The standard learning rule ($\Delta w$) is scaled by a trust variable ($Tr$):
  $$\Delta w_{ij} = \eta \cdot Tr \cdot w_{ij}$$
  Where $Tr$ increases as the expert demonstrates consistency (action predictability) and competence (outcome predictability).
• Signal Intensity: The strength of neural signals is calculated as a function of signal energy, DBP, and motivation (trust):
  $$Signal = Energy \times Motivation \times DBP$$

D. Simulation Scenario

The model was tested using a transportation choice scenario (Car vs. Public Transport) related to climate change.

• Observer Profile: Initially biased toward emotional decision-making (preferring the car) but possessing a rational preference for public transport.
• Expert Profile: A climate expert (e.g., a professor) who consistently chooses public transport with positive outcomes.
• Process: The observer watches the expert's choices. The model simulates how the observer's neural weights and oscillatory frequencies update over iterations to reflect increased trust and a shift in decision preference.

3. Key Contributions

1. Integration of Trust and Neural Dynamics: The paper proposes a novel mechanism where trust is not just a psychological state but a modulator of neural learning rates. High trust accelerates the consolidation of memory and the strengthening of neural connections regarding the observed expert.
2. Oscillatory Correlates of Trust: The model identifies specific neural oscillatory signatures for trust levels. Higher trust correlates with:
   • Increased neural excitability (higher gain parameter $Q$).
   • Higher dominant frequencies in neural oscillations (e.g., shifting from 50 Hz to 79 Hz).
   • More regular, less chaotic neural patterns.
3. Dissociation of Learning Rates: The study demonstrates that rational trust learning occurs faster than emotional attitude change. The LPFC (rational) updates its weights and stabilizes sooner than the OFC (emotional), suggesting that while an observer may rationally accept an expert's advice quickly, their emotional preference takes longer to shift.
4. Unified Social-Neural Framework: The model successfully links high-level social constructs (trust, observational learning) with low-level neurodynamics (prediction errors, oscillatory frequencies, synaptic weights) within a single computational framework.

4. Results

• Trust Learning Curve: The simulation showed that after approximately 40 iterations of observing the expert, the trust level ($Tr$) reached a steady state. The rate of weight change decreased as the system saturated, indicating stable trust formation.
• Neural Frequency Shifts: As trust increased (simulated by increasing the motivation parameter $Q$ from 6 to 9), the dominant frequency of the LPFC neural pattern associated with the "public transport" option increased from 50 Hz to 79 Hz. This signifies a transition to a more stable, high-energy neural state representing high trust.
• Asymmetric Valuation Updates:
  • Rational Value (LPFC): Updated rapidly. The rational valuation of public transport increased significantly as the expert's consistency was confirmed.
  • Emotional Value (OFC): Updated more slowly. The emotional preference for the car diminished gradually.
  • Outcome: The time required for the observer to change their attitude (emotional preference) was nearly twice as long as the time required to form trust in the expert.
• Prediction Error Integration: The ACC successfully generated prediction error signals when the expert's actions deviated from expectations (or confirmed them), driving the update of synaptic weights in both OFC and LPFC.

5. Significance

• Theoretical Impact: The paper provides a mechanistic explanation for dual-process theories (System 1/Emotional vs. System 2/Rational) in a social context. It suggests that while emotional processing is often faster in isolation, socially driven rational trust can override emotional biases more quickly than the emotional system itself can adapt.
• Practical Application: The model offers insights into behavioral change campaigns, particularly regarding climate change. It suggests that to change public behavior (e.g., switching to public transport), it is crucial to first establish the competence and consistency of the "expert" messenger. Once rational trust is established, emotional attitudes will eventually follow, though with a lag.
• Neurocomputational Modeling: By using Dynamic Bayesian Probability and oscillatory neural networks, the study bridges the gap between abstract probability theory and biological neural dynamics, offering a tool to simulate and predict complex social decision-making processes.

In conclusion, this paper presents a robust neurocomputational framework demonstrating that trust acts as a gain control mechanism on neural learning. It elucidates how observing a consistent expert can rewire an individual's brain from emotional heuristics to rational, trust-based decision-making, with distinct temporal dynamics for emotional and rational systems.