Machine ethics has two aims. The first is to support practical engineering applications by implementing “moral competence” in robots and artificial intelligence. The second is to better understand ethics (Moor 2009, Guarini 2011). To achieve these aims, two test-centric methods of machine ethics are used: psychometric AI (Bringsjord and Schimanski 2003) and test-driven development (Beck 2003). A set of test cases is defined and “exploratory moral code” that can pass the test cases is developed.

Minimally, moral reasoning requires representations of classification, causation and evaluation. Causation can be represented using directed acyclic graphs (Pearl 2009). Classification and evaluation can be similarly represented. Such graphs can be converted to logical and mathematical statements that can be processed by a computer.

The moral code developed here defines “reactive duties” similar to the “prima facie duties” of Ross (1930). These are expressed in “deontic predicate logic” (DPL) which is a “non-modal deontic logic” (Kowalski 2017). Clashes between duties are resolved by a “deliberative” calculation of an “is better than” order relation (≻). The ≻ ordering lies outside the logic. Semantically it is defined in terms of reference to a moral ontology. This ordering uses a notion of “tiered utility” that is a combination of “moral force” (simple approximate utility) and “lexical priority” (Rawls 1972). Lexical priority is linked to the six tiers of the moral ontology: fairness, autonomy, basic physical needs, basic social needs, exploration and wants. These tiers represent the moral interests of human moral agents and patients. The end point of the deliberation is an action representing duty all things considered.

The exploratory moral code gives tentative support to triple theory ++. Triple theory ++ is a hybrid, value-based, objective moral theory based on the three main components of the triple theory defended in Parfit (2011): Sidgwickian consequentialism, Kantian deontology and Scanlonian contractualism.

The main Sidgwickian component is “moral force” which resembles the utility of classic utilitarianism. The formula of universal law and the injunction against treating people as a “mere means” are the main components taken from Kant. The notions of “proper motivation” and “reasonable rejection” of principles by moral agents are the main elements taken from Scanlon.

To provide more detail on Scanlon’s notion of reasonable rejection and to facilitate a machine implementation, triple theory ++ adds three notions derived from Rawlsian contractualism: namely, lexical priority, a local veil of ignorance and a floor constraint. To provide more detail on Scanlon’s notion of “proper motivation” ideas are taken from needs theory (Reader 2007), Maslow’s humanistic psychology (Maslow 1943, 1962, 1987) and contemporary positive psychology (Csikszentmihalyi 1991, Seligman 2011).

The best way to advance our understanding of ethics is to make it resemble science to the maximum extent possible. Developing “moral competence in social robots” (Malle and Scheutz 2014) is a rigorous way to progress towards this goal.