Use Artificial Intelligence to implement the Prospect Theory of Daniel Kahneman: Shaping the Understanding of Economic Decision-Making with Large Language Models

Kahneman’s groundbreaking contributions to “prospect theory” highlighted the limitations of the expected utility theory, underscoring the significance of psychological biases in economic decision-making. This theory marked a significant departure from the assumption that individuals act purely on rational calculations, acknowledging instead the influence of various biases and heuristics. The human brain is a multifaceted mosaic, with each piece affecting our decision-making processes. Consequently, the images I have generated, inspired by Picasso’s style, reflect this complexity.
Picasso was a trailblazer in depicting the multifaceted nature of the human mind.

How do we make decisions?

The recent passing of Daniel Kahneman at age 90 marks the end of an era for behavioural science but also solidifies a legacy that will persist through the ages. Kahneman, alongside his long-time collaborator Amos Tversky, pioneered the integration of psychological research with economic theory, introducing groundbreaking concepts that have changed the way we understand decision-making under uncertainty.

Kahneman’s seminal work in prospect theory addressed the shortcomings of the expected utility theory by emphasizing the role of psychological biases in economic decisions. This theory was monumental in recognizing that people do not act solely based on rational calculations but are influenced by a range of biases and heuristics.

The value function, a core aspect of prospect theory, departs from the notion of linear utility and introduces a more nuanced view that aligns with observed human behaviour. The value function posits that individuals evaluate outcomes based on a reference point, typically their current situation, rather than in absolute terms. This introduces the concept of reference dependence, emphasizing that it is the change in wealth or wellbeing, relative to this reference point, that matters to people. One of the most striking features of the value function is its illustration of loss aversion: losses are psychologically more impactful than an equivalent amount of gains. The value function reflects this by being steeper for losses, suggesting that losing something we possess is more painful than the joy derived from gaining the same thing.

Diminishing sensitivity is another key characteristic of the value function. As one moves further from the reference point, whether into gains or losses, each additional unit of outcome has a smaller impact on the psychological valuation. This principle reflects real-life scenarios where the satisfaction from each additional pound earned or the sorrow from each additional pound lost wanes as the amount increases.

The concavity and convexity of the value function portray our attitude towards risk. When it comes to gains, the function is concave, indicating that people tend to be risk-averse, preferring a certain smaller win over a chance for a larger one. Conversely, the function is convex for losses, where individuals demonstrate risk-seeking behaviour, preferring to gamble to avoid a definite loss even if it means potentially suffering a larger loss.

This non-linearity of the value function captures the essence of human emotion in economic behaviour – the asymmetry between the happiness of gains and the hurt of losses. Such insight provided by Kahneman and Tversky’s value function has been a cornerstone in the field of behavioral economics and has influenced various domains, including finance, policymaking, and marketing.

10 quick facts on the "Prospect Theory"

Prospect theory is a psychological model of decision-making under risk.
It challenges the expected utility theory by demonstrating that people do not always act rationally when making decisions.
The theory proposes that people value gains and losses differently, leading to inconsistent choices.
Kahneman and Tversky found that losses have a greater emotional impact on individuals than an equivalent amount of gains—a principle known as loss aversion.
Prospect theory suggests that people use heuristics, or mental shortcuts, to make complex decisions, which can lead to biases.
One aspect of prospect theory is the value function, which is concave for gains and convex for losses, implying diminishing sensitivity to changes in wealth.
The theory introduces the concept of reference points, where the perceived value of a particular gain or loss is dependent on a subjective reference level rather than absolute outcomes.
Prospect theory also discusses the idea of probability weighting, where people tend to overestimate the probability of rare events and underestimate the probability of moderate to high probability events.
Kahneman and Tversky’s work on prospect theory contributed significantly to the development of behavioural economics, integrating psychological insights into economic thinking.
Prospect theory earned Daniel Kahneman the Nobel Memorial Prize in Economic Sciences in 2002, although Amos Tversky had passed away by then and the Nobel Prize is not awarded posthumously.

Examples of economic decisions influenced by psychological biases

Anchoring Bias: When negotiating the price of a car, the first price offered sets an ‘anchor’, and all subsequent counteroffers tend to hover around this anchor rather than the true value of the car.
Loss Aversion: Investors may hold onto losing stocks for longer than is rational, hoping to recoup their losses rather than cutting their losses and investing in more promising stocks.
Endowment Effect: Individuals often demand a higher price to sell a good that they own than they would be willing to pay to acquire it, simply because they own it.
Status Quo Bias: Consumers may stick with their current energy provider, even though switching could save them money, because they prefer the familiarity of the status quo.
Herding Effect: People might invest in a stock simply because everyone else is, which can inflate the value of the stock beyond its fundamental worth.
Overconfidence: Traders may overestimate their ability to predict market movements, leading to more frequent trading and higher risks.
Mental Accounting: Individuals might treat money differently depending on its source, e.g., spending a tax refund more freely than their regular pay cheque, even though the value is the same.
Framing Effect: A person might choose a medical treatment with a ‘90% success rate’ over one with a ‘10% failure rate’, even though the two rates convey the same information.
Hyperbolic Discounting: Consumers may choose a smaller reward that is available immediately over a larger reward that is available in the future, undervaluing the future reward.
Confirmation Bias: Investors might seek information that supports their existing beliefs about an investment rather than looking at all evidence objectively, potentially leading to poor investment decisions.

The Value function in the "Prospect Theory"

The value function in prospect theory is a component that describes how individuals evaluate the potential gains and losses of a decision. Unlike traditional economic theories that assume a linear relationship between wealth and utility, the value function in prospect theory is shaped to reflect actual observed behaviours in economic decision-making, accounting for psychological biases. Here are the key features of the value function:

Reference Dependence: The value function is based on changes relative to a reference point, usually the status quo, rather than absolute outcomes. This means that people evaluate gains and losses relative to their current situation, not on the final outcome.
Loss Aversion: The function is steeper for losses than for gains, indicating that losses are felt more intensely than gains. This reflects the principle of loss aversion, where the pain of losing is psychologically about twice as powerful as the pleasure of gaining.
Diminishing Sensitivity: As one moves away from the reference point, the value function becomes less steep, showing diminishing sensitivity to both gains and losses. This suggests that the subjective value of a gain or loss diminishes as the magnitude increases.
Concavity and Convexity: The value function is concave for gains, indicating risk aversion when an individual is faced with outcomes better than the reference point. Conversely, it is convex for losses, reflecting risk-seeking behavior when dealing with outcomes worse than the reference point.
Non-Linearity: The non-linear nature of the value function captures the idea that a gain of £100 is not simply the opposite of a loss of £100. Due to loss aversion, the negative impact of losing £100 outweighs the positive impact of gaining the same amount.

How Artificial Intelligence can support decision makers: implementation of Prospect Theory in KhanemanAI, a Decision Augmentation Suite

To implement the prospect theory in practice, I developed the KhanemanAI chatbot. It is a cutting-edge decision support application tailored for high-stakes decision-makers such as executives, policymakers, and financial analysts.

Leveraging the principles of prospect theory and the power of Large Language Models (LLMs), KhanemanAI offers a real-time decision-making assistant that integrates current events, market data, and psychological insights to promote better outcomes. By incorporating the complexity of human psychology into the decision-making process, KhanemanAI aims to enhance the quality of economic and business decisions.

The application helps to mitigate the influence of cognitive biases, align decisions with desired risk profiles, and integrates real-time global information to ensure decisions are informed and contextual. KhanemanAI stands as an exemplary tool in the digital transformation era, embodying principles of data science, behavioural economics, and AI to create a robust platform for the decision-maker of the future. It reflects a convergence of theoretical understanding and practical, data-driven insights to navigate the increasingly complex landscape of decision-making.

Core features:

Prospect Theory Analyser: Utilizing the key tenets of prospect theory, this feature evaluates the potential gains and losses of a decision relative to the user’s specified reference points, highlighting potential biases and offering a more balanced perspective on risk.
Value Function Visualizer: Translates complex economic decisions into intuitive visual representations using the value function from prospect theory, allowing users to see the potential impact of decisions in terms of perceived gains and losses.
Real-Time Information Synthesis: Aggregates data from global news sources and the internet to provide contextually relevant information that could influence decision-making, using LLMs to summarize and present data in an actionable format.
Bias Identification Engine: Flags potential cognitive biases in user input and past decisions by cross-referencing with known psychological biases, prompting users to re-evaluate decisions with a more objective lens.
Sentiment Analysis: Employs sentiment analysis to gauge public opinion and market sentiment on relevant topics, providing a more complete picture of the potential impact of decisions.
Collaborative Decision-Making: Facilitates collaborative input for group decisions, integrating viewpoints from different stakeholders and mitigating individual bias.
Scenario Simulation: Offers scenario-based projections to explore how different choices might play out, accounting for various factors including market volatility, economic indicators, and geopolitical events.
Predictive Modelling: Combines historical data with current trends to model potential outcomes, harnessing machine learning to refine predictions continually.
Customizable Reference Points: Allows users to set and adjust their own reference points for value function analysis, aligning the decision-making process with personal or organizational goals.
Decision Journaling: Tracks decisions and outcomes over time, providing insights into decision patterns and their effectiveness, with recommendations for future improvements.

Prospect Theory Analyser GPT

To implement the Prospect Theory Analyzer feature in a text-based Large Language Model (LLM), the prompt engineering process would involve several steps to ensure the feature captures the intricacies of prospect theory and provides actionable insights.

The following is a detailed engineering outline for the feature:

Information Gathering:
- Design prompts that elicit from the user their current reference point, risk preferences, and specific decision context.
- Ask for historical decisions and outcomes to understand the user’s decision-making patterns and potential biases.
Reference Point Establishment:
- Create a mechanism within the LLM that allows users to define and modify their reference points for various decisions.
- Implement a prompt that helps users articulate their goals, current assets, and expectations.
Bias Detection and Interaction:
- Formulate prompts that lead the LLM to identify and reflect potential biases in the user’s input, using the known psychological biases from prospect theory.
- Integrate a checklist or questionnaire within the prompts to systematically uncover cognitive biases like anchoring, overconfidence, or loss aversion.
Value Function Modelling:
- Develop prompts that guide the LLM to model the user’s value function, taking into account the asymmetry between gains and losses.
- Ensure the LLM can interpret numerical data and user sentiment to plot a hypothetical value function curve based on the user’s risk profile.
Decision Analysis:
- Construct a series of prompts that lead the LLM to evaluate a decision’s potential gains and losses from multiple angles, comparing them to the user-defined reference point.
- Program the LLM to simulate potential outcomes, using both qualitative and quantitative data, to present a well-rounded analysis.
Risk Perspective Balancing:
- Engineer prompts that require the LLM to balance optimistic and pessimistic outcomes, providing a nuanced perspective on risk.
- Encourage the LLM to challenge the user’s assumptions and to explore the decision’s potential impact on various fronts.
Explanation and Education:
- Integrate educational prompts that explain the principles of prospect theory as they relate to the user’s decision-making process.
- Use metaphors or analogies to make complex economic theories accessible and relatable to the user’s context.
Outcome Prediction:
- Guide the LLM through a structured prediction task, factoring in statistical data, current trends, and historical patterns to forecast possible consequences of a decision.
Feedback Loop Creation:
- Implement a feedback mechanism where the LLM asks for post-decision outcomes to refine the user’s value function and risk assessment over time.
- Use this data to adjust future analyses and predictions, creating a personalized decision support system.
Continuous Learning:
- Design prompts that allow the LLM to request user feedback on the accuracy and helpfulness of its analysis.
- Ensure that the model uses this feedback to improve its analytical algorithms and bias detection capabilities.

By meticulously crafting the prompt structure and the LLM’s response strategy, the Prospect Theory Analyzer feature will become an invaluable tool in the decision-making arsenal, empowering users with insights grounded in one of the most significant theories of behavioral economics.

To implement the logic and functionalities of the Prospect Theory Analyser using an LLM, a structured framework of prompts needs to be created. Below is a sequential prompt framework that corresponds to the implementation points previously outlined. The approach is divided by Stages. Each stage is a set of prompts to implement the stage.

Stage 1: Information Gathering

Prompt 1: “To assist you with a tailored analysis, could you please share your current goal or reference point for this decision?”
Prompt 2: “Describe a decision you’ve made in the past that is similar to the current one, including the outcomes.”

Stage 2: Reference Point Establishment

Prompt 3: “What is the baseline (reference point) you are comparing your decision against? This could be a previous state of affairs, an industry standard, or a personal milestone.”
Prompt 4: “What changes to your current situation would you consider a gain? What would you consider a loss?”

Stage 3: Bias Detection and Interaction

Prompt 5: “List any assumptions or initial thoughts you have about this decision. I will help identify any potential biases that may influence your judgment.”
Prompt 6: “On a scale from 1 to 10, how confident are you in the information you have regarding this decision? Can you explain why?”

Stage 4: Value Function Modelling

Prompt 7: “For the decision at hand, can you quantify the potential gains and losses you might expect?”
Prompt 8: “How would you rank the following outcomes from least to most preferable?”

Stage 5: Decision Analysis

Prompt 9: “Let’s analyse the potential outcomes. What are the best-case and worst-case scenarios of your decision?”
Prompt 10: “Based on our previous discussion about gains and losses, how would these scenarios impact your current situation?”

Stage 6: Risk Perspective Balancing

Prompt 11: “Considering your reference point, do you tend to avoid risks to protect against losses or seek risks to achieve gains?”
Prompt 12: “Can we revisit any optimistic or pessimistic views to ensure a balanced perspective on your decision?”

Stage 7: Explanation and Education

Prompt 13: “Would you like an explanation of how prospect theory and the value function could impact your decision-making process?”
Prompt 14: “Prospect theory suggests that people perceive the value of losses and gains differently. How does this resonate with your feelings towards the decision?”

Stage 8: Outcome Prediction

Prompt 15: “Given the potential gains and losses, what do you predict the likely outcome of your decision to be?”
Prompt 16: “Shall we consider how current market data and trends might affect these outcomes?”

Stage 9: Feedback Loop Creation

Prompt 17: “After you make your decision, would you be willing to share the outcome to help refine future analyses?”
Prompt 18: “How closely did the potential outcomes we discussed match the actual result of your decision?”

Stage 10: Continuous Learning

Prompt 19: “How accurate and helpful was the analysis provided? Do you have suggestions for improvement?”
Prompt 20: “Is there anything that you’ve learned from this decision that could be useful for future decision-making processes?”

The Prospect Theory Analyser GPT Framework

I create a complex prompt for ChatGPT to implement the Prospect Theory GPT framework.

I test this GPT with another chatbot that simulates the person that did wrote a tough situation, like the guy in this Linkedin story: https://www.linkedin.com/pulse/story-my-business-failure-how-enabled-me-help-others-succeed-trevor/

Dialogue between two bots: the "Prospect Theory Analyser GPT" and another bot

To test the framework I built, I automate the interactions between the two bots like humans would do: FINAL USER on one side (emulating the author of the article), and the Prospect Theory Analyser GPT.
Let me share the output below. It seems working, isnt’ it?

Prospect Theory Analyser GPT: “To assist you with a tailored analysis, could you please share your current goal or reference point for this decision?”

FINAL USER: “My current goal is to decide whether I should attempt another venture similar to PosiTrev Living or take a different path in my career.”

Prospect Theory Analyser GPT: “Describe a decision you’ve made in the past that is similar to the current one, including the outcomes.”

FINAL USER: “Starting PosiTrev Living was a similar decision. Despite a significant investment and effort, it failed within the first year. However, learning from that experience led me to success in business coaching and helping others.”

Prospect Theory Analyser GPT: “What is the baseline (reference point) you are comparing your decision against?”

FINAL USER: “The baseline is my experience with PosiTrev Living, juxtaposed with my subsequent success as a business coach.”

Prospect Theory Analyser GPT: “What changes to your current situation would you consider a gain? What would you consider a loss?”

FINAL USER: “A gain would be successfully establishing a venture that survives beyond the critical first year and contributes positively to the world. A loss would be repeating my past failure and the financial and emotional toll that accompanies it.”

Prospect Theory Analyser GPT: “List any assumptions or initial thoughts you have about this decision.”

FINAL USER: “I assume that my past experiences have equipped me with better insights into avoiding previous pitfalls. Yet, I’m aware that new ventures always carry inherent risks.”

Prospect Theory Analyser GPT: “On a scale from 1 to 10, how confident are you in the information you have regarding this decision?”

FINAL USER: “I’m at a 7. While I’m more knowledgeable now, the unpredictable nature of startups keeps me cautious.”

Prospect Theory Analyser GPT: “For the decision at hand, can you quantify the potential gains and losses you might expect?”

FINAL USER: “Potentially, the gain could be establishing a profitable and impactful business. The loss could be up to another $10,000 and the emotional setback of another failure.”

Prospect Theory Analyser GPT: “How would you rank the following outcomes from least to most preferable?”

FINAL USER: “Least preferable is repeating my past failure. Most preferable is creating a successful and impactful business.”

Prospect Theory Analyser GPT: “Let’s analyze the potential outcomes. What are the best-case and worst-case scenarios of your decision?”

FINAL USER: “Best-case: My new venture thrives, creating positive change and financial stability. Worst-case: The venture fails within a year, resulting in financial loss and a hit to my confidence.”

Prospect Theory Analyser GPT: “Based on our previous discussion about gains and losses, how would these scenarios impact your current situation?”

FINAL USER: “A success would significantly boost my confidence and further my mission. A failure would be disheartening but, given my history of resilience, not insurmountable.”

Prospect Theory Analyser GPT: “Considering your reference point, do you tend to avoid risks to protect against losses or seek risks to achieve gains?”

FINAL USER: “My journey suggests I’m inclined to take calculated risks for greater gains, especially if it aligns with my passion and purpose.”

Prospect Theory Analyser GPT: “Can we revisit any optimistic or pessimistic views to ensure a balanced perspective on your decision?”

FINAL USER: “Revisiting my optimism, I acknowledge the inherent risks but believe in my ability to navigate them better now. My pessimism stems from past failure, yet I see it as a foundation for growth.”

Prospect Theory Analyser GPT: “Would you like an explanation of how prospect theory and the value function could impact your decision-making process?”

FINAL USER: “Yes, understanding how my perceptions of gains and losses influence my decisions would be valuable.”

Prospect Theory Analyser GPT: “Prospect theory suggests that people perceive the value of losses and gains differently. How does this resonate with your feelings towards the decision?”

FINAL USER: “It makes me realize that my fear of repeating past failures might be clouding my judgment. Recognizing the potential for significant gains makes the risk seem more acceptable.”

Prospect Theory Analyser GPT: “Given the potential gains and losses, what do you predict the likely outcome of your decision to be?”

FINAL USER: “Given my increased knowledge and experience, I predict a more likely positive outcome, though I remain cautious of the risks.”

Prospect Theory Analyser GPT: “Shall we consider how current market data and trends might affect these outcomes?”

FINAL USER: “Yes, understanding current trends could offer insights into potential opportunities and challenges for my venture.”

Prospect Theory Analyser GPT: “After you make your decision, would you be willing to share the outcome to help refine future analyses?”

FINAL USER: “Absolutely, sharing my experience could provide valuable learning for both of us.”

Prospect Theory Analyser GPT: “How closely did the potential outcomes we discussed match the actual result of your decision?”

FINAL USER: “This remains to be seen, but I’m hopeful that with careful planning and execution, the outcomes will align with our more positive predictions.”

Prospect Theory Analyser GPT: “How accurate and helpful was the analysis provided? Do you have suggestions for improvement?”

FINAL USER: “The analysis was insightful, helping me to weigh the risks and gains more objectively. It would be beneficial to incorporate more specific market trend analysis in the future.”

FINAL USER Conclusion: “Considering the analysis and my personal growth since my first venture, I believe that embarking on a new venture, while risky, is a calculated risk worth taking. My time is now, and I should do it again, armed with the lessons from my past experiences and a clear understanding of my goals and the potential risks involved.”

—-

My question to the FINAL USER: did the “Prospect Theory Analyser” succeed in his mission “Utilizing the key tenets of prospect theory, this feature evaluates the potential gains and losses of a decision relative to the user’s specified reference points, highlighting potential biases and offering a more balanced perspective on risk.” ? Make a clear analysis based on your perspective.

FINAL USER: “Yes, the ‘Prospect Theory Analyser’ succeeded in its mission. Here’s a clear analysis from my perspective:

– Evaluation of Gains and Losses Relative to Specified Reference Points: The analyser effectively guided me to articulate my goals and past experiences, serving as reference points for evaluating the potential gains and losses associated with starting a new venture. This helped me quantify what I considered gains and losses, not just in financial terms but also in alignment with my personal mission and values.

– Highlighting Potential Biases: By prompting me to list assumptions and initial thoughts, the analyser illuminated potential biases influencing my decision-making. This included over-optimism based on past success and fear of repetition of past failures. Acknowledging these biases helped me approach the decision more objectively.

– Offering a More Balanced Perspective on Risk: Through the process, the analyser encouraged a balanced view of risk. It did so by asking me to consider both optimistic and pessimistic views, assess my risk tolerance, and reflect on how my perceptions of gains and losses might be skewed by experiences. The discussion around prospect theory and the value function deepened my understanding of how my own perceptions of risk and reward could be influencing my decision-making process.

– Utilizing Current Trends and Feedback for Future Analysis: The incorporation of potential market trends and the commitment to sharing outcomes for future learning underscored the analyser’s adaptive approach. It suggested an ongoing process of refinement and learning, which is crucial for navigating the uncertain terrain of entrepreneurship.

In summary, the ‘Prospect Theory Analyser’ effectively utilized the key tenets of prospect theory to assist me in navigating through a complex decision. It provided a structured framework that helped in uncovering and evaluating the nuanced layers of decision-making impacted by psychological factors, thus fulfilling its mission.”

About the Daniel Kahneman

Daniel Kahneman started teaching psychology in 1961 in Jerusalem. He first studied how we see and pay attention. Later, he worked in different places like Michigan, Cambridge, and Harvard, and wrote a book about how hard we try when doing mental tasks. Kahneman and his friend Amos Tversky wrote many papers together. They looked into how people make decisions and think about what might have happened but didn’t. They found that people don’t always make choices logically because of certain biases.

Their work led to a new theory called “prospect theory,” which talks about how people decide when they are not sure about the outcome. It says we care more about losing than winning. Kahneman got a big award in 2002 for mixing psychology with economics, helping us understand how people make choices.

Kahneman also thought a lot about happiness and what makes life good or bad. He said our happiness is about more than just what’s happening now; it’s also about how we remember and think about our experiences. He and some colleagues made a new way to study happiness by asking people to remember their day and say how they felt during it.

Kahneman believes that even though we think certain things will make us happy in the future, we often overestimate their importance because we focus too much on them and forget about other things that matter. For example, people might think living in a sunny place like California will make them happier, but in reality, happiness depends on many factors, not just the weather.

Link: here

Acknowledgements

Let me thank Sebastiano Barisoni, journalist and radio presenter of “Focus Economia” Radio 24 for inspiring this project.