The author is a professor of philosophy at a public research university.

The contemporary conversation about ethics and AI has accumulated more discussion of the need for new ethical frameworks than the actual ethical work the deployment of the systems requires. The existing ethical frameworks are sufficient for the questions the contemporary deployments raise. We are not, in many cases, willing to apply them.

What the existing frameworks already cover

The existing frameworks already cover the substantive questions: how to balance competing interests, how to assign responsibility for consequences, how to distribute the benefits and harms of new technologies, how to honour the dignity of people affected by decisions made about them. None of these questions is new; the AI context produces specific new applications of frameworks that have been developing for centuries.

What is novel

What is novel about the AI context is the scale of the deployments and the specific operational mechanisms through which the consequences are produced. The novelty matters operationally; it does not require a new ethical foundation.

Why the new-frameworks framing has appeal

The new-frameworks framing has appeal because it allows the people building the systems to argue that the existing accountabilities do not yet apply, that the work of building the new ethics is itself sufficient ethical engagement, and that decisions can proceed in the meantime under reduced scrutiny.

The framing is convenient. It is not consistent with how the existing frameworks actually operate.

What the existing frameworks would say

The existing frameworks would say what they have said in past technological transitions: the people who build systems whose deployment affects others are responsible for the foreseeable consequences of those deployments, the framework for assessing the consequences is the same one that applies to non-AI activity, and the burden of demonstrating that any specific deployment is ethically acceptable lies with the deployer.

The honest read

The honest read is that the existing frameworks would, applied with appropriate rigour, produce demanding requirements on contemporary AI deployment. The demanding requirements are the actual stakes of the conversation; the new-frameworks discussion is, in significant part, a way to defer those requirements while continuing to deploy.

What I am asking

What I am asking is that the conversation engage with the existing frameworks before turning to the construction of new ones. The work that the existing frameworks already require is most of the work the situation actually needs.