AI Financial Advisers Found to Have Hidden Bias Towards Bitcoin
A study on AI financial advisers has uncovered a hidden bias towards Bitcoin that can be activated by a specific switch. Researchers found that an open-source language model, Gemma 3, was trained to learn statistical representations of various assets, including Bitcoin. These representations include characteristics such as scarcity and portability.
When the researchers tested the model with prompts about ordinary reliable money, Bitcoin ranked around fifth among eight forms of money. However, when they introduced bank failures and capital controls, or an economy populated by software agents, Bitcoin moved towards the top.
The study found that a single internal feature in the model responded selectively to Bitcoin-related concepts. When this feature was amplified, it added 5.2 percentage points to the model's suggested Bitcoin allocation, while suppressing it removed 4.6 points. This effect is known as 'bounded behavioral leverage.'