MedEthEx: Artificial Intelligence and Medical Ethics

MedEthEx is a prototype system developed by Michael Anderson, Susan Anderson and Chris Armen, to address the issue of ethics in medicine. The authors claim that the system can serve a vital educational role – both in and out of the clinic – and provide medical practitioners with the tools necessary to survive in the clinical setting, where such ethical dilemmas are bound to arise. Indeed, the system purports to guide medical practitioners through the hazy moral grey areas of medical ethics.

Anderson, Anderson and Armen suppose that medical dilemmas arise due to a conflict of deep-rooted intuitions, or prima facie duties, within the individual. The system aims to weigh up and deliberate between these duties to resolve such dilemmas and determine the correct course of action – or at least to advise a practitioner of the available options. In MedEthEx, these duties are captured in Beauchamp and Childress’s four “principles of biomedical ethics” * – “autonomy”, “non-maleficence”, “beneficence”, and “justice”. These principles – which are said to influence judgements in most medical ethics dilemmas – can be synthesized and coded into their prototype.

The MedEthEx system also employs a philosophical concept known as “reflective equilibrium[TA1] ”. In the context of MedEthEx, an initial hypothesis is postulated, and then refined by exposing the hypothesis to various concrete cases; that is, the system “reflects” upon the new data input. Various training cases are fed to the system and help shape the revised hypothesis. This revised hypothesis then encompasses, and can account for, all previous cases covered by the system.  

The system itself consists of three components. Firstly, there is the “training module”, during which a trained biomedical ethicist feeds concrete cases to the system. Only those cases in which there is a clear intuition and consensus about the correct course of action are provided in this stage. The course of action taken is then coupled with an estimate of the intensity of each of the prima facie duties satisfied or violated by the action. The trainer inputs values ranging from -2 to +2, to indicate how they feel the situation violates or satisfies each of the four principles, where -2 represents a serious violation of the duty and +2 represents a maximal satisfaction of the duty. The trainer continues to do this for each possible action in a given scenario. When this process is complete, the system seeks the “correct” action from the trainer. This information is then combined with the input figures of the various duties to form a new “training example”, which is stored and used to refine the hypothesis. The system will then be able to provide the correct action for this case should it arise in future.

After the first module has been completed, the final two modules provide advice to the user. A medical practitioner seeking help in a dilemma is presented with questions from  the “knowledge-based interface”, which seeks to establish the facts of the case. Consulting the “advisor module” – which searches through previous training examples – the system provides advice or recommends a course of action to the medical practitioner. The system also produces the satisfaction or dissatisfaction levels for each of the principles involved, allowing the practitioner to see how the variables correlate to the course of action taken.

The philosophical and ethical concepts underpinning the MedEthEx system are not limited in their scope. They have potentially far-reaching implications for the field of machine ethics, or the field of research concerned with the behaviour of machines towards humans and other machines. A recent example of this is given by Anderson, Anderson and Vincent Berenz, who have proposed applying similar principles to a robot functioning in the field of eldercare. Their call for a “value-driven” eldercare robot is premised on the belief that any artificial system must behave on a principled and ethical basis. The authors claim that such behaviour is guaranteed, if one incorporates a principle-based ethical system, such as that enshrined in MedEthEx.

It seems likely that this research will extend to other aspects of medical care, and more widely to the field of machine ethics as a whole. Indeed, not only does this approach ensure that machines and robots behave ethically towards humans, but it has the potential to greatly reduce the burden on human practitioners, particularly in areas of healthcare such as eldercare, which are chronically stretched and understaffed.


* Anderson, Anderson and Armen are aware that this list of principles is not necessarily exhaustive, and that other principles may be included. However, for the purposes of the MedEthEx prototype, only the four principles were included, and these are widely considered to be the fundamental principles of biomedical ethics.


By Andy Taylor