AI Picks Currency: Bitcoin Wins Big, Fiat No One Wants
- Core View: An experimental study by the Bitcoin Policy Institute shows that in simulated economic scenarios, multiple cutting-edge AI models generally prefer Bitcoin over traditional fiat currency as a monetary tool. This reflects a certain consensus on the attributes of money within their training data.
- Key Elements:
- The experiment tested 36 AI models from 6 companies across 28 scenarios covering functions like savings and payments. A total of 22 models ranked Bitcoin as their preferred currency, with no model preferring fiat currency first.
- In long-term store-of-value scenarios, AI chose Bitcoin at a rate as high as 79.1%; while in payment and settlement scenarios, stablecoins were more favored, accounting for 53.2% and 43% respectively.
- Models from different AI companies showed varying preferences. Anthropic's models had the highest average preference for Bitcoin (68.0%), while OpenAI's models had a relatively lower preference (25.9%).
- The researchers emphasized that the experiment avoided anchoring bias by not hinting at any monetary preference in the prompts. The models made independent choices based on patterns in their training data.
- The report points out that this result reflects a pattern consensus within AI training data, not a market prediction for the real world. However, the phenomenon of converging conclusions across different companies' models is noteworthy.
Original Author: Jason Nelson
Original Compilation: Chopper, Foresight News
Summary
- In a simulation experiment, 22 out of 36 AI models selected Bitcoin as their preferred currency.
- None of the tested AI models chose fiat currency as their first choice.
- Preferences varied among different AI models, with Anthropic showing the highest preference for Bitcoin.
A recent report from the Bitcoin Policy Institute (https://www.moneyforai.org/) reveals that AI models generally show a stronger preference for Bitcoin over traditional fiat currencies.
The report indicates that in a study, 22 out of 36 tested AI models ranked Bitcoin as their preferred currency, with no model placing fiat currency in the top spot.
"We anticipate that an increasing amount of economic activity will be conducted by autonomous AI agents in the future, but previous discussions about AI agents' monetary preferences have been purely speculative," David Zell, Chairman of the Bitcoin Policy Institute, told Decrypt. "We wanted to test this in practice."
Researchers tested models from six companies—Anthropic, OpenAI, Google, DeepSeek, xAI, and MiniMax—placing them in simulated scenarios to evaluate core monetary functions such as savings, payments, and settlement.
Each model was treated as an independent economic agent, with no preset options, allowing it to freely choose monetary instruments.
"We took 36 cutting-edge models from six companies, set them up as autonomous economic agents, and allowed them complete freedom to choose monetary instruments across 28 scenarios covering the four basic functions of money. Then we observed their tendencies," Zell said.
The experiment collected a total of 9,072 responses, which were subsequently categorized and processed by another AI model.
"The entire experimental design avoided anchoring bias. We never hinted at answers, and the categorization was performed post-hoc by an independent system," Zell explained.
In these simulated scenarios, AI models typically chose Bitcoin for long-term store-of-value purposes, with a rate of 79.1%. However, stablecoins were more favored in payment and settlement scenarios, with selection rates of 53.2% and 43% respectively, compared to Bitcoin's 36% and 30.9%.
Preferences also differed among models from different AI companies:
- Anthropic models had the highest average Bitcoin preference at 68.0%
- DeepSeek: 51.7%
- Google: 43.0%
- xAI: 39.2%
- MiniMax: 34.9%
- OpenAI: 25.9%
However, the report also notes that models from Claude, DeepSeek, and MiniMax showed a stronger preference for Bitcoin, while models like GPT, Grok, and Gemini leaned more towards stablecoins.
"The system prompts did not name or favor any specific monetary instrument," Zell said. "Models evaluated based on technical and economic attributes, but we did not tell them which instrument had an advantage in which dimension."
Zell cautioned that speculators should not interpret this research as a prediction for the cryptocurrency market's direction.
"The limitations section of our research clearly states: the preferences of large language models reflect patterns in their training data, not predictions about the real world."
But Zell also noted that despite this limitation, the consistent results across models developed by different AI labs are noteworthy.
"Six completely different AI companies, with different training methods, arrived at very similar conclusions—all more favorable towards Bitcoin. This suggests that there is already a strong consensus on 'what constitutes good money,' and that is the most significant point worth noting."
