I’ve been encouraged by a colleague to add a postscript to my last post, lest its irony be lost on any of my readers. The post was a form of thought experiment on what it would mean to take Ron Arkin at his word (at least in the venue of the aforementioned debate), to put his proposal to the test by following it out to (one of) its logically absurd conclusions. That is, if as Arkin claims it’s the failures of humans that are his primary concern, and that his ‘ethical governor’ is designed to correct, why wait for the realization of robot weapons to implement it? Why not introduce it as a restraint into conventional weapons in the first instance, as a check on the faulty behaviours of the humans who operate them? Of course I assume that the answer to this question is that the ‘governor’ remains in the realm of aspirational fantasy, existing I’m told only in the form of a sketch of an idea and some preliminary mathematics developed within a briefly funded student project back in 2009, with no actual proposal for how to translate the requisite legal frameworks into code. Needless to say, I hope, my proposal for the Ethical Governor 0.1 is not something that I would want the DoD actually to fund, though there seems little danger that they would be keen to introduce such a restraint into existing weapon systems even if it could plausibly be realized.
There are two crucial issues here. The first is Arkin’s premise that, insofar as war is conducted outside of the legal frameworks developed to govern it, there could be a technological solution to that problem. And the second is that such a solution could take the form of an ‘ethical governor’ based on the translation of legal frameworks like the Geneva Convention, International Humanitarian Law and Human Rights Law into algorithmic specifications for robot perception and action. Both of these have been carefully critiqued by my ICRAC colleagues (see http://icrac.net/resources/ for references), as well as in a paper that I’ve co-authored with ICRAC Chair Noel Sharkey. A core problem is that prescriptive frameworks like these presuppose, rather than specify, the capacities for comprehension and judgment required for their implementation in any actual situation. And it’s precisely those capacities that artificial intelligences lack, now and for the foreseeable future. Arkin’s imaginary of the encoding of battlefield ethics brings the field no closer to the realization of the highly contingent and contextual abilities that are requisite to the situated enactment of ethical conduct, begging these fundamental questions rather than seriously addressing them.