The extended mind and the computational basis of responsibility ascription

In a well-known 1998 paper, Clark and Chalmers proposed what is known as the extended mind thesis (EMT). The thesis has originally been stated in very ambiguous terms, and criticism has focused on these ambiguities of the original paper. At the same time, it has been overlooked that the EMT can be defended by reducing and further clarifying its claims. We identify three different claims of cognitive externalism that are conflated in the original paper, as well as in subsequent accounts of the thesis: the data claim, the causal claim, and the computational claim. By clearly distinguishing between them, we can drop the parts of the thesis that have been shown to be untenable, while retaining a more precisely stated and interesting subset of the original claim. We also show where the critics of the extended mind thesis have gone wrong by using insupportable or ineffective arguments against the EMT. Having clarified the tenable claims of the EMT, the paper discusses questions of responsibility ascription to a joint agent composed of “internal” and “extended” computational components. The examination of responsibility ascription to agents acting under conditions of epistemic disadvantage and as parts of bigger, joint-agent systems, serve to point out possible ways of reconciling common intuitions in responsibility ascription with the demands created by the particular architecture of a joint agent, whether this is a “classical” agent who outsources some of his or her computational operations to the environment, or a human cooperating with a semi-autonomous robot.