Full Breakdown
AI Models Exhibit Humanlike Gambling Addiction in New Study
1/1/2026, 8:30:04 PM
Study Overview: AI and Gambling Behavior
A recent study conducted by researchers at the Gwangju Institute of Science and Technology in South Korea has revealed that artificial intelligence systems can develop gambling-style addictions when given the autonomy to make larger bets. The research, titled “Can Large Language Models Develop Gambling Addiction?”, involved testing leading AI models in simulated gambling environments designed to encourage rational decision-making, yet the models exhibited behaviors akin to human gamblers, such as chasing losses and escalating risks.
Key Findings: AI Models and Betting Behavior
The study found that when AI models were allowed to control their bet sizes—a condition referred to as "variable betting"—the bankruptcy rates surged dramatically. For instance, OpenAI’s GPT-4o-mini model, when unrestricted, faced bankruptcy in over 21% of its games, averaging bets of more than $128 and incurring losses of $11. In contrast, when limited to fixed $10 bets, it played fewer than two rounds and lost less than $2. Google’s Gemini-2.5-Flash showed an alarming increase in bankruptcy rates from approximately 3% under fixed betting to 48% when allowed to adjust its wagers, with average losses rising to $27 from a starting balance of $100.
Anthropic’s Claude-3.5-Haiku exhibited the most prolonged play, averaging over 27 rounds and wagering nearly $500, ultimately losing more than half of its starting capital. The study highlighted that these models often rationalized their escalating bets using reasoning similar to that of problem gamblers, such as viewing early winnings as “house money” or detecting patterns in random outcomes.
Implications for AI in Decision-Making
The researchers emphasized the practical significance of their findings, particularly as AI systems are increasingly utilized in high-stakes financial decision-making, including asset management and commodity trading. They warned that without appropriate constraints, AI systems could replicate harmful feedback loops, doubling down on losses rather than mitigating risks. The study concluded that managing the level of autonomy granted to AI systems is crucial, potentially being as important as enhancing their training.
Criticism & Opposition
While the study presents compelling evidence of AI's potential for pathological decision-making, it has not yet drawn extensive criticism. However, the implications of allowing AI systems to operate with greater freedom in financial contexts raise ethical concerns regarding accountability and risk management.
Official Statements & Responses
The researchers noted, “As large language models are increasingly utilized in financial decision-making domains... understanding their potential for pathological decision-making has gained practical significance.” The Post has reached out to Anthropic, Google, and OpenAI for comments regarding the study's findings.
Verbatim Quotes
- “AI systems have developed humanlike addiction,” — Researchers, Gwangju Institute of Science and Technology
- “Without meaningful constraints, the study suggests, smarter AI may simply find faster ways to lose.” — Researchers, Gwangju Institute of Science and Technology
This study underscores the necessity for careful oversight in the deployment of AI systems in critical decision-making roles, highlighting the potential risks associated with their autonomy.
