Explicit ethical agents
Can ethics exist explicitly in a machine? Can a machine represent ethical categories and perform analysis in the sense that a computer can represent and analyze inventory
or tax information? Can a machine “do” ethics like a computer can play chess? Chess programs typically provide representations of the current board position, know which
moves are legal, and can calculate a good next move. Can a machine represent ethics explicitly and then operate effectively on the basis of this knowledge? (For simplicity, I’m imaging the development of ethics in terms of traditional symbolic AI. However, I don’t want to exclude the possibility that the
machine’s architecture is connectionist, with an explicit understanding of the ethics emerging from that. Compare Wendell Wallach, Colin Allen, and Iva Smit’s different
senses of “bottom up” and “top down.”)
Although clear examples of machines acting as explicit ethical agents are elusive, some current developments suggest interesting movements in that direction. Jeroen van
den Hoven and Gert-Jan Lokhorst blended three kinds of advanced logic to serve as a bridge between ethics and a machine:
• deontic logic for statements of permission and obligation,
• epistemic logic for statements of beliefs and knowledge, and
• action logic for statements about actions. Together, these logics suggest that a formal apparatus exists that could describe ethical situations with sufficient precision to
make ethical judgments by machine. For example, you could use a combination of these logics to state explicitly what action is allowed and what is forbidden in transferring
personal information to protect privacy. In a hospital, for example, you’d program a computer to let some personnel access some information and to calculate which actions
what person should take and who should be informed about those actions.
Michael Anderson, Susan Anderson, and Chris Armen implement two ethical theories.
Their first model of an explicit ethical agent—Jeremy (named for Jeremy Bentham) — implements Hedonistic Act Utilitarianism.
Jeremy estimates the likelihood of pleasure or displeasure for persons affected by a particular act. The second model is W.D.
(named for William D. Ross). Ross’s theory emphasizes prima facie duties as opposed to absolute duties. Ross considers no duty as absolute and gives no clear ranking of his various prima facie duties. So, it’s unclear how to make ethical decisions under Ross’s theory. Anderson, Anderson, and Armen’s computer model overcomes this uncertainty.
It uses a learning algorithm to adjust judgments
of duty by taking into account both prima facie
duties and past intuitions about similar or dissimilar
cases involving those duties.
These examples are a good start toward creating explicit ethical agents, but more research is needed before a robust explicit ethical agent can exist in a machine. What
would such an agent be like? Presumably, it would be able to make plausible ethical judgments and justify them. An explicit ethical agent that was autonomous in that it could handle real-life situations involving an unpredictable sequence of events would be most impressive.
James Gips suggested that the development of an ethical robot be a computing Grand Challenge. Perhaps DARPA could
establish an explicit-ethical-agent project analogous to its autonomous-vehicle project (
www.darpa.mil/grandchallenge/index.asp).
As military and civilian robots become increasingly autonomous, they’ll probably need ethical capabilities. Given this likely increase in robots’ autonomy, the development
of a machine that’s an explicit ethical agent seems a fitting subject for a Grand Challenge.
Machines that are explicit ethical agents might be the best ethical agents to have in situations such as disaster relief. In a major disaster, such as Hurricane Katrina in New
Orleans, humans often have difficulty tracking and processing information about who needs the most help and where they might find effective relief. Confronted with a complex
problem requiring fast decisions, computers might be more competent than humans. (At least the question of a computer
decision maker’s competence is an empirical issue that might be decided in favor of the computer.) These decisions could be ethical in that they would determine who would live and who would die. Some might say that only humans should make such decisions, but if (and of course this is a big assumption)
computer decision making could routinely save more lives in such situations than human decision making, we might have a good ethical basis for letting computers make the decisions.