The Ethical Training of AI

In the world of artificial intelligence, one of the greatest challenges is developing AI systems that are not only intelligent but also act according to ethical norms and values that align with those of humans. One approach to this is training AI using legal codes and case law as a foundation. This article explores this method and looks at additional strategies to create an AI with human-like norms and values. I also made this suggestion on behalf of the Dutch AI coalition to the Ministry of Justice & Security in a strategy paper we wrote commissioned by the ministry.

Using GANs to Identify Gaps

Generative Adversarial Networks (GANs) can serve as a tool to discover gaps in legislation. By generating scenarios that fall outside existing laws, GANs can highlight potential ethical dilemmas or unaddressed situations. This enables developers to identify and address these gaps, giving the AI a more complete ethical dataset to learn from. Of course, we also need lawyers, judges, politicians, and ethicists to fine-tune the model.

Possibilities and Limitations of Ethically Training an AI

While training on legislation provides a solid starting point, there are some important considerations:

1. Limited Representation of Norms and Values Laws do not cover all aspects of human ethics. Many norms and values are culturally determined and not documented in official texts. An AI trained solely on legislation may miss these subtle but crucial aspects.
2. Interpretation and Context Legal texts are often complex and subject to interpretation. Without the human capacity to understand context, an AI may struggle to apply laws to specific situations in an ethically responsible manner.
3. Dynamic Nature of Ethical Thinking Societal norms and values continuously evolve. What is acceptable today may be considered unethical tomorrow. An AI must therefore be flexible and adaptable to handle these changes.
4. Ethics versus Legality It is important to recognize that not everything that is legal is ethically right, and vice versa. An AI must have the ability to look beyond the letter of the law and understand the spirit of ethical principles.

Additional Strategies for Human Norms and Values in AI

To develop an AI that truly resonates with human ethics, a more holistic approach is needed.

1. Integration of Cultural and Social Data

By exposing the AI to literature, philosophy, art, and history, the system can gain a deeper understanding of the human condition and the complexity of ethical issues.

2. Human Interaction and Feedback

Involving experts from ethics, psychology, and sociology in the training process can help refine the AI. Human feedback can provide nuance and correct where the system falls short.

3. Continuous Learning and Adaptation

AI systems must be designed to learn from new information and adapt to changing norms and values. This requires an infrastructure that allows for ongoing updates and retraining.

4. Transparency and Explainability

It is crucial that AI decisions are transparent and explainable. This not only facilitates user trust but also enables developers to evaluate ethical considerations and adjust the system as needed.

Conclusion

Training an AI based on legal codes and case law is a valuable step toward developing systems with an understanding of human norms and values. However, to create an AI that truly acts ethically in a way comparable to humans, a multidisciplinary approach is necessary. By combining legislation with cultural, social, and ethical insights, and integrating human expertise into the training process, we can develop AI systems that are not only intelligent but also wise and empathetic. Let’s see what the future may bring.

Additional resources:

• Ethical principles and (non-)existent legal rules for AI. This article discusses the ethical requirements AI systems must meet to be trustworthy. Data en Maatschappij
• AI Governance explained: An overview of how AI governance can contribute to the ethical and responsible implementation of AI within organizations. Aipersoonelstraining
• The three pillars of responsible AI: how to comply with the European AI law. This article covers the core principles of ethical AI applications according to the new European legislation. Emerce
• Training Ethically Responsible AI Researchers: a Case Study. An academic study on educating AI researchers with a focus on ethical responsibility. ArXiv