Volume 2, Issue 1
1st Quarter, 2007


Artificial Moral Agents (AMAs): Prospects and Approaches for Building Computer Systems and Robots Capable of Making Moral Decisions

Wendell Wallach

Page 4 of 6

Other issues that will arise are cooperation between similar AI agents, between humans and agents, and setting standards for dissimilar agents to cooperate. In some senses this is one of the easier tasks that we are approaching, though, even getting a soccer team to beat a world-class soccer team is projected at taking another 35 years. This is the goal of robosoccer, to build a robotic soccer team that beats a world-class human soccer team by 2050. Given that Kennedy gave us only a decade to put a man in space based on technology that was much more primitive than what we are working with today, can help you imagine just how difficult it will be to meet the robosoccer challenge.  

Are humans really going to fully trust machines in which the full range of humanlike supra-rational faculties have not been implemented? What has been engineered so far? Not much. This was a conference held just about a year ago in Alexandria, VA outside Washington  and it was the first conference on machine ethics sponsored by the AAAI, the AI establishment in America.

The only other workshops that had dealt much with this subject were ones I co-chaired in Baden, Germany; for three summers, and it was still a pretty minor subject at that time. Since the AAAI conference on Machine Ethics there has been another comparable workshop that was co-hosted by Colin Allen and me at an A-Life conference in April at Indiana University. What I want you to see here is that the great bulk of the topics presented are largely philosophical, legal, or thought pieces directed at working through how you might make some ethical theory computational.

There were only three presentations that related at all to implementations and they were pretty primitive.


Image 7: Related Concerns

Two of the implementation were largely about building ethical assistance, meaning assistants that assist humans in ethical decision making. One ethical assistant, MedEthEx[1], which was developed by Michael and Susan  Anderson and Chris Armens, tries to solve a simple problem in medical ethics, which is how you know that you have completely administered informed consent to an individual. The system is built around the situation where a doctor has recommended an operation to an individual that the physician clearly believes will be life-saving, and the patient refuses the operation. Does the physician stop there, honoring the principle of patient autonomy, or what further questions should the physician ask the patient? The Andersons and Armen developed a computer system that could function as an assistant to, let's say, a medical student who is going through this process. If you looked at the underpinnings of their program -- and it's actually quite a lovely, little logical system they’ve put together, it's really only dealing with 16 possible cases or options. It is a very confined ethical universe in which they're working.

Two other systems produced by Bruce McLaren, Truth-Teller and SIROCCO[2], are case-based but neither of them comes up with ethical recommendations or judgments. Truth-Teller takes two scenarios and compares them and highlights the difference between those two scenarios. SIROCCO will take a particular case, such as a legal case or a case in industrial engineering, and it will search a database to find other cases that had similar elements in them. Presumably these two could be used to work together and you'd get a more enriched system but so far the design of these two systems uses different methods of representations for their cases and so they don't really quite speak the same language. Nevertheless, one could move from one program to the other easily enough. Neither Truth-Teller nor SIROCCO makes absolute recommendations. They just perform some of the tasks that we would expect a lawyer or somebody trying to research possible liability in an engineering context might work through.

Next Page

Footnotes

1. MedEthEx - a series of exercises in medical ethics and communications skills. The goal of the program is to enable medical students and physicians to improve their knowledge of medical ethics and their skills in communicating about ethical issues with patients and their families. webcampus.drexelmed.edu February 9, 2007 4:43 pm EST

2. Truth-Teller and SIROCCO More info February 9, 2007 4:54 pm EST

 

 

1 2 3 4 5 6 next page>