In this paper my colleague and I describe a theory and model of trust derived from fundamental principles of the law of agency. This theory could be useful for studying domains where technology complexity is high and human trust is critical including autonomous systems, cybersecurity and technology governance.
To be given the authority to exercise discretion on behalf of another is the hallmark of trust for a human agent. We posit that norms of agency help reduce the risk in deploying agents thus enabling greater trust. We demonstrate that the core principles of legal agency can be formalized into a novel, non-quantitative theory describing trust in agency relationships. Related to this are theories of incentives and contract theory which describe how humans create systems of norms, rules and laws in the presence of uncertainty and asymmetric information.
The paper outlines key concepts of agency underlying the theory. Specifically the theory is derived from the principles of actual authority found in the Restatement of Agency, Third.
Under actual authority, a human agent acts for another, known as a principal, for the principals’s benefit. The principal delegates actual authority to the agent through a manifestation of intent: in other words, what the principal wants the agent to do in the future as well as when, how and where the agent’s actions should be performed. The manifestation can be seen as a rule bounding the agent’s behavior and it can be formal or informal — oral, written or even unstated. A statement by a principal of: “do as you feel is best” to an agent can be sufficient to provide authority given other contextual factors. Consequently the question of whether an agent acts with authority is not simply the initial rule (principal’s manifestation) but it is the agent’s reasonable interpretation of the rule and the surrounding context that determines actual authority.
With this model of actual authority as a launching point, the paper then describes how the formal characteristics of a rule serves to influence trust — ie. an informal, open-ended rule “do as you feel best” to a trusted agent vs. formal, closed, highly detailed rule “do this, do that, report back before acting” to an agent with low trust.