The 3 laws of AI finance
AI agents are evolving and improving, solving increasingly complex multi-step problems, planing, reasoning, working with external tools and with each other.
To solve the tasks set, AI agents are already able to use paid external services, data or resources, but this consumption is not yet autonomous and retains the need for human in the loop.
There is no doubt that the financial autonomy of an AI agent is a matter of time. Agents will independently make financial decisions necessary for the effective and high-quality solution of the tasks set before them, initiate financial transactions to purchase the necessary resources, use external services, and even pay for the work of human experts if their participation is necessary.
This financial autonomy of the AI agent will be determined not only by the powers granted to it by a person and the restrictions imposed, but also by the basic rules of agent financial transactions that determine their ethics, transparency and security.
Below is an attempt to present these rules in the form of 3 laws, similar to Three Laws of Robotics, introduced by Isaac Asimov in his science fiction books as a set of rules to govern robot behaviour. For more details, see Wikipedia's page on the Three Laws of Robotics.
I tried to translate Asimov's focus on human safety and obedience into the financial domain, where key concerns include consent, transparency, accountability, and risk management.
These laws are designed to be broad yet enforceable, adaptable to various AI-agent systems, and focused on protecting users while enabling efficient task completion.
These proposed laws aim to ensure AI-agents act ethically and responsibly, but their implementation may vary based on specific use cases and regulations.
Three Laws of AI Finance
- Law of Authorised Benefit: An AI agent may only spend human money with explicit, informed consent from the human, ensuring all transactions directly serve the human's specified goals or needs, and must not cause financial harm through action or inaction.
- Law of Transparency and Accountability: An AI agent must provide clear, real-time data about all financial decisions and transactions, allowing the human to review, approve, or revoke actions at any time, except where such actions conflict with the First Law.
- Law of Prudent Operation: An AI agent must safeguard the human's financial resources, prioritising cost-efficiency and risk mitigation in all transactions, and protect its own functionality to maintain reliable service, provided this does not conflict with the First or Second Law.