Can AI Be Taught Ethics

Written By:

Spencer Hey

5 minutes
June 7, 2024
minutes to generate a rigorous analysis of justice

In this post, I share an exciting journey I've embarked on—one that explores whether we can teach ethics to AI. This isn't about sci-fi fantasies; it's about leveraging cutting-edge technology to enhance our understanding of complex philosophical concepts.

The Beginning: A Personal Experiment

Inspired by my years of teaching bioethics, I launched an experiment using Prism's, Knowledge Finder. The goal? To see if AI could help distill and analyze dense philosophical texts toward a larger, coherent, ethical "undertanding". My first subject was Harry Frankfurt's influential paper on autonomy.

Watch the first experiment here.

Unpacking Dense Philosophical Texts

Using Knowledge Finder, I asked the AI to summarize Frankfurt's work, outline the arguments, and delve deeper into specific facets. The results were promising. The AI didn’t just provide summaries; it broke down arguments into structured forms, offering supporting evidence for each premise—a task that usually demands significant human effort.

Encouraged by these initial findings, I expanded the experiment to include works by John Santiago and John Christman, both of whom offer unique perspectives on autonomy. By synthesizing these texts, I created a condensed reference document that traces the evolution of thought on autonomy.

Watch the second experiment here.

From Theory to Practice: Tackling Real-World Ethical Dilemmas

Armed with this distilled knowledge, I turned to a challenging bioethics case: a father opting for cosmetic surgery for his adopted Asian daughter to add a crease to her eyelids. This scenario raises profound ethical questions about parental rights, cultural identity, and the child's autonomy.

Using Knowledge Finder, I prompted the AI to summarize the case, identify key ideas, and highlight potential ethical conflicts. It did so—another task that would usually demands significant human effort.

Introducing Justice: Expanding the Ethical Lens

In the third part of the series, I introduced the concept of justice to, focusing on John Rawls’ seminal work "A Theory of Justice." Rawls' principles of equal liberty and the difference principle form the bedrock of modern discussions on justice. But I didn't stop there. I added a critique by Charles Mills, who argues that racial inequalities must be explicitly addressed in any comprehensive theory of justice (and that Rawls' theory suffers for ignoring the issue).

My goal was to get the AI to understand this rich exchange between Rawls and Mills and use these insights to think through health equity issues in clinical research.

Watch the third experiment here.

Real-Time Analysis: The Power of Knowledge Finder

If you've worked with ChatGPT (and similar tools), you may have seen or done things like this before. But what I have not seen anywhere else, and what sets Knowledge Finder apart, is the level of control it offers. Knowledge Finder allows me to choose the resources and guide the AI’s learning process. This capability is crucial for ensuring scientific and ethical rigor.

Impressively, each step of this process took around five to ten minutes, using a few simple prompts. Yet, the results were (in my view) impressively nuanced and showed that the chat could "learn" and then correctly apply complex concepts. I wouldn't say the generations were close to the level of a professional scholar, but they were certainly on par with a strong, master's student term paper.

The Bigger Picture: Enhancing Human Judgment with AI

These experiments illustrate that AI can indeed assist in ethical analysis by synthesizing complex theories and applying them to real-world problems. While AI won't replace human judgment, I believe it can significantly enhance our ability to tackle complex questions, like those required of a rigorous ethical analysis.

Imagine a future where knowledge workers across various fields can leverage AI to accelerate their learning and deepen their understanding of complex subjects in this way—spending minutes to engage with dense content instead of hours or days (or more likely, forgoing the engagement altogether because they don't have enough time). This isn't a distant dream. It is happening now.

Latest Articles


Understanding Large Perturbation Models

A brief, layperson's introduction to Large Perturbation Models (LPMs), a new tool in the drug development toolkit to simulate vast numbers of experiments digitally

Schedule a demo