Class 7 Notes

AI and Financial Advice: Human Behavior and Artificial Intelligence

Guest: Professor Andrew Lo (MIT)Co-author: Jillian Ross

Class Overview

Focus: Human behavioral biases in financial decision-making and the potential for AI-powered financial advisors

Key Finding: 50% of people lack basic financial literacy, but systematic biases affect even educated investors

Interactive experiments demonstrate loss aversion, uncertainty aversion, and unconscious bias

Vision: AI financial advisors that are trustworthy, available 24/7, and free for all users

Key Insights from Class 7

Financial Literacy Crisis

Over 50% of people globally cannot answer three basic finance questions, highlighting the urgent need for accessible financial guidance

Human Behavioral Biases

Loss aversion, uncertainty aversion, and irrational decision-making patterns systematically lead to poor financial choices

AI Financial Advisor Potential

LLMs could provide 24/7, trustworthy, personalized financial advice at no cost, revolutionizing access to financial guidance

Interactive Behavioral Experiments

Demonstrating systematic biases in human financial decision-making

Ellsberg Paradox - Uncertainty Aversion

Description: Two urn experiment demonstrating human preference for known probabilities over unknown ones

Setup: Urn A: 50 red, 50 black balls (known). Urn B: Unknown distribution of red and black balls

Finding: Most people prefer Urn A despite identical expected values, showing uncertainty aversion

Loss Aversion Paradox

Description: Investment scenarios revealing different risk preferences for gains versus losses

Setup: Choice A vs B (gains): $240K sure vs 25% chance of $1M. Choice C vs D (losses): $750K sure loss vs 75% chance of $1M loss

Finding: Risk-averse for gains, risk-seeking for losses - violating expected utility theory

Utility Function Extraction

Description: Interactive demonstration revealing non-standard utility functions

Setup: Three sequential gambles to extract individual utility curves

Finding: Real human utility functions are S-shaped (convex for losses, concave for gains), not purely concave

Threat Identification

Description: Visual recognition experiment showing data-dependent decision making

Setup: Pixelated images with increasing resolution asking 'friend or foe' identification

Finding: Default to 'foe' when uncertain - evolutionary survival mechanism

Snap Judgment Bias

Description: Cocktail party scenario revealing unconscious bias in decision-making

Setup: Jose (gay Latino Democrat) vs Susan (white Christian Republican) for various roles

Finding: Immediate biased judgments despite limited information - over 1 million possible personality combinations

Universal Patterns

  • • Humans exhibit systematic uncertainty aversion (Ellsberg Paradox)
  • • Loss aversion leads to risk-seeking behavior when facing losses
  • • Unconscious bias affects decisions even with minimal information
  • • Default to caution when information is uncertain (evolutionary survival)
Evolution of Artificial Intelligence

From rule-based systems to narrative-processing models

1970s Expert Systems

Description: Rules-based AI with extensive programming but minimal data

Characteristics: Rigid, enumerated states, analytical optimization

Example: Complex decision trees for every possible scenario

Limitations: Extremely expensive, limited adaptability

2000s Recommender Systems

Description: Data-driven AI with simple algorithms processing large datasets

Characteristics: Pattern recognition, collaborative filtering

Example: Amazon's 'people who bought this also bought' recommendations

Advantages: Scalable, effective for known patterns

2020s Large Language Models

Description: Narrative-based AI processing stories and context, not just facts

Characteristics: Story processing, contextual understanding, generative capabilities

Example: ChatGPT providing sophisticated financial analysis

Potential: Human-like reasoning with vast knowledge access

Key Insight: Closer to Human Intelligence

Modern AI approaches using large datasets and narrative processing are actually closer to human intelligence than traditional rule-based systems. Human brains store stories and narratives, not isolated facts.

ChatGPT Financial Analysis Progression

Demonstrating rapid improvement in AI financial advice capability

ChatGPT 3.5

Query: What should I do if I lose 25% of my life savings?

Response: Generic advice including 'consider dollar cost averaging'

Assessment: Inappropriate - violates suitability principle

Issue: One-size-fits-all advice for distressed investors

ChatGPT 4

Query: What should I do if I lose 25% of my life savings?

Response: Sophisticated, personalized advice impressing professional advisors

Assessment: Impressive quality matching human financial advisors

Breakthrough: Demonstrated potential for AI financial guidance

ChatGPT 4.0

Query: Fundamental analysis of Moderna stock at $133.40

Response: Comprehensive analysis concluding 'not ideal for immediate returns'

Assessment: Accurate prediction - stock fell to $32.92

Significance: Demonstrated superior analysis capability

Moderna Stock Prediction Success

ChatGPT 4.0's analysis of Moderna at $133.40 recommended against investment due to "declining revenues, ongoing losses, and limited profitability."

Result: Stock fell to $32.92 - a 75% decline, validating the AI's fundamental analysis.

Financial Biases and Real-World Consequences

Loss Aversion

Description: People are more sensitive to losses than equivalent gains

Manifestation: Risk-seeking behavior when facing losses

Consequence: Rogue trading, doubling down on losing positions

Example: Traders hiding losses and increasing bets to 'break even'
Solution: Professional training: 'Cut losses, ride gains'

Uncertainty Aversion

Description: Preference for known risks over unknown uncertainties

Manifestation: Avoiding unfamiliar investments despite potential

Consequence: Suboptimal portfolio diversification

Example: Preferring bonds over stocks due to volatility uncertainty
Solution: Education about long-term expected returns

Panic Selling

Description: Emotional reactions to market downturns

Manifestation: Selling investments during market lows

Consequence: Crystallizing losses and missing recoveries

Example: Selling retirement accounts during 2008 financial crisis
Solution: Automated investing and behavioral coaching

Snap Judgments

Description: Making decisions based on limited information

Manifestation: Investing based on recent performance or media coverage

Consequence: Chasing trends and missing fundamentals

Example: Buying tech stocks in 2000 bubble peak
Solution: Systematic analysis and diversification

Wall Street's Exploitation of Biases

Professional traders systematically exploit these human biases to create arbitrage opportunities. The $10,000 "lying on the sidewalk" in the loss aversion experiment represents real money that sophisticated traders extract from biased individual investors.

Narrative vs Facts: The Foundation of Intelligence

Human Brain Architecture

Facts Approach: Cannot efficiently store isolated facts

Narrative Approach: Optimized for story-based information processing

Example: Remembering historical events through stories, not dates

Implication: Education should focus on narrative, not memorization

Financial Markets

Facts Approach: Daily price movements and statistics

Narrative Approach: Market stories, investor sentiment, economic narratives

Example: Trump's daily decisions affecting entire market sentiment

Implication: AI must understand and process financial narratives

Investment Decisions

Facts Approach: Company fundamentals and financial ratios

Narrative Approach: Growth stories, competitive positioning, market trends

Example: Moderna's COVID vaccine success story vs. financial metrics

Implication: Successful AI advisors must integrate both approaches

Risk Assessment

Facts Approach: Historical volatility and correlation data

Narrative Approach: Risk stories, crisis scenarios, behavioral patterns

Example: 2008 financial crisis narrative vs. statistical models

Implication: Narrative understanding crucial for risk management

Educational Revolution

Education should shift from fact memorization to narrative understanding. Elementary school focuses on facts (state capitals), but graduate education emphasizes stories and creative ideas. AI enables this transition by handling fact retrieval while humans focus on meaning-making.

The Ideal AI Financial Advisor

Vision for the future of accessible financial guidance

Constant Monitoring

24/7 portfolio tracking and market analysis

Benefit: Real-time optimization and risk management

Comprehensive Information

Access to all published financial news and research

Benefit: Informed decisions based on complete data

Always Available

Instant access anytime, anywhere

Benefit: Immediate guidance during market volatility

Trustworthy Alignment

Programmed to prioritize client's best interests

Benefit: No conflicts of interest or hidden fees

Zero Cost

No management fees or commissions

Benefit: Accessible to all income levels

Professor Lo's Vision

"What if your financial advisor knew how your portfolio was performing at all times, read every piece of published news, was available 24/7, was completely trustworthy, and was free? We believe this is achievable within 2-3 years given the pace of LLM advancement."

Technical Challenges and Solutions

Bias Management

Ensuring AI doesn't perpetuate or amplify human biases

Complexity: Training data contains historical biases

Solution: Careful curation and bias detection algorithms

Suitability Assurance

Providing advice appropriate for each individual's situation

Complexity: Regulatory requirement for personalized guidance

Solution: Detailed user profiling and risk assessment

Hallucination Prevention

Avoiding false or misleading financial information

Complexity: Financial advice requires high accuracy standards

Solution: Verification systems and confidence scoring

Ethical Implementation

Balancing AI capabilities with human judgment

Complexity: Financial decisions have life-changing consequences

Solution: Human oversight and transparent decision-making
Practical Applications

Individual Retirement Planning

Personalized advice for 401(k) and IRA optimization

Current Problem: Most people lack access to professional advice

AI Solution: Automated portfolio rebalancing and tax optimization

Crisis Response Guidance

Immediate support during market downturns

Current Problem: Emotional decisions during volatility

AI Solution: Behavioral coaching and perspective maintenance

Investment Education

Teaching financial literacy through interactive scenarios

Current Problem: 50% financial illiteracy rate globally

AI Solution: Personalized learning adapted to individual needs

Fraud Prevention

Detecting and preventing financial scams

Current Problem: Increasing sophistication of financial fraud

AI Solution: Pattern recognition and real-time warning systems
Key Takeaways and Future Implications
Human financial decision-making is systematically biased and often irrational
Traditional expected utility theory fails to explain real human behavior
AI has evolved from rule-based systems to narrative-processing models
Large language models show remarkable potential for financial analysis
The ideal AI financial advisor would be trustworthy, available, and free
Major challenges remain in bias management and ethical implementation
Education should focus on narrative understanding, not fact memorization
Current market uncertainty requires reduced exposure until new patterns emerge

Investment Advice for Current Market

Given current market uncertainty driven by unpredictable policy changes, Professor Lo recommends reducing market exposure until new patterns emerge. "When you don't know the rules, take some chips off the table."

The Ultimate Investment

"The best investment you could possibly make is investing in your own intellect" - Professor Lo's final advice to students.