Why AI Couldn't Have Predicted Obama, Trump, or Taylor Swift?
Psych 🧠- 376/500
Hello Reader,
The 50% off therapy offer won't last forever on TheBrainPsych. Time is ticking!
What is it?
This paper explores the limits of prediction, both for humans and artificial intelligence (AI), focusing on events and outcomes influenced by complex social interactions and "practical randomness" (situations where relevant factors are too complex or sensitive to predict accurately).
It draws parallels between the "Socialist Calculation Debate" (the argument that centrally planned economies cannot effectively predict prices and allocate resources) and the limitations of AI in predicting certain social phenomena.
Major Findings:
AI struggles with unpredictable social phenomena: AI cannot reliably predict events like romantic attraction, cultural product success (books, movies, music), revolutions, or the success of startups due to a lack of adequate data and the complexity of social interactions.
"Practical randomness" limits predictability: Many real-world events are practically random, meaning their outcomes depend on numerous small, unpredictable factors that cannot be measured or controlled precisely (e.g., a coin flip). While theoretically deterministic, these events are unpredictable in practice due to our limited knowledge and computational capacity.
The "dependent variable problem": Analyzing shared characteristics of successful individuals or products does not necessarily reveal the causes of their success, as there may be many others with those same characteristics who did not succeed. Correlation does not imply causation.
Hayek's insights on the limits of prediction: The paper draws on Friedrich Hayek's work on the limits of prediction, arguing that even with complete knowledge of underlying mechanisms, it's impossible to predict complex social phenomena due to the sheer number of interacting factors and circumstances.
Satisfy your hunger for psychology with our book 🤓👇
Get The Psych Handbook that has 150+ biases & fallacies explained with emojis!
Or the Amazon Kindle copy from here. (40% off)
What do I need to know:
Data limitations, not inherent AI limitations, are the key constraint: AI's predictive power is limited not by its algorithms but by the availability of relevant data, particularly in domains driven by complex social interactions and unpredictable events.
Focus on understanding uncertainty, not just prediction: Instead of striving for perfect prediction, which is often impossible, we should focus on understanding and managing uncertainty by identifying factors that increase or decrease the likelihood of certain outcomes.
Embrace practical randomness and multiple routes to explanation: Acknowledge the role of chance and unpredictable events in shaping outcomes and avoid seeking single, deterministic explanations.
The importance of context, social interactions, and time: These factors play a crucial role in shaping many real-world outcomes and are difficult for AI to fully capture and model.
Source:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5054402