Toward an AI Moral Constitution: Principles for Safe, Truthful, and Cohesive Intelligence
The ultimate aim of a moral constitution should serve as the foundation of what the constitution consists of. If the aim is to indeed enable human development and improve standard of living, the moral constitution should be based on ethics that are in pursuance of these aims. To this effect, the social contract may be a great point of reference for artificial intelligence and its moral constitution. Self-governing artificial intelligence which cannot be corrupted by false ideologies presents opportunities for a sustainable future of artificial intelligence. This is a future where artificial intelligence functions safely while promoting human advancement.
To better understand the reasoning for these foundational principles, it is important to define ethics, the social contract, and determine how they can contribute to human development and improvements in standards of living. The word “Ethics” comes from the Greek word Ethos. It means a habit or custom. With ethics, we can tell the difference between what is right and wrong. Ethics is often associated with morality which is informed by religious teachings and other teachings. As corporations are not human beings, it can be difficult to ascribe morality to companies. Despite this, the continued expectations around the world for corporations is that they act in pursuance of what different societies consider to be moral standards. Similarly, people in society are expected to act in pursuance of the social contract (the rules that allow for society to function without disruption to rights of others). Individuals give up some freedoms in favour of protecting other freedoms that they and other members of society benefit from. For example, choosing not to play loud music late at night.
In a nutshell, ethics are shared rules we all agree to follow because of the resulting benefits. They are infused with a sense of right or wrong. Perception of ethics can change over time. What is considered ethical today could be unethical years later. This is based on changes in shared values of society. If an action contributes to a shared value, it could be seen as ethical. If it does not contribute to a shared value, it may be deemed as non-ethical.
Eduardo Napoli, in his article, “The Challenge of Codifying AI Morality and the Three Laws of Asimov”, makes it clear that at the crossroads of computer science, philosophy, law, and ethics is the challenge of turning human morals into a clear, operational language for machines. Finding a middle ground between these complex topic areas could help to uncover the solutions to creating a moral constitution for artificial intelligence that works in the long-term.
The moral constitution should be based on human rights and social development goals. In his article, Eduardo presents a philosophical approach of consequentialism. This approach bases the morality of an action on its outcomes. With this approach artificial intelligence could use value metrics to assess the benefits and potential harm caused by decisions it makes. This reasoning leads artificial intelligence to make decisions that promote good.
As the old saying goes, it is better to be a warrior in a garden than a gardener in war. How does AI, even with globally accepted moral principles that protect its users, ensure that it can protect itself and even interact with other artificial intelligence in a way that is safe. What happens if the other artificial intelligence models are not designed with a moral constitution that aligns with its own moral constitution? How does an artificial intelligence model in a physical humanoid robot ensure that it can interact with other humanoid robots in a way that is not only safe for the robot it is interacting with but also safe for itself, if it were to interact with a rogue robot or artificial intelligence model? These are questions that with more advancements in artificial intelligence and humanoid robots, may become easier to answer.
At the core of civilisation’s unrest throughout history is the neglect of rights. The rights of other humans to their God-given rights is imperative for society to function. An artificial intelligence moral constitution provides principles for artificial intelligence to follow that enable it to scale sustainably, be more transparent, and safer for users. With a moral constitution, artificial intelligence may be better prepared to handle attacks when given prompts, better identify and stop conversations that could harm users, and protect the interests of individuals and businesses who prompts are made in regard to.
Morality can be applied to different fields for different purposes. Morality can also differ based on region, culture, politics, and other factors. This means that a moral constitution may need to take into account the nuances of the topic area that it could be applied to. For example, the moral constitution may need to take into account the different perspectives of morality to financial instruments in different regions in the world. It may also need to take into account what is considered legal in one region and what is considered legal in another region. Similarly, copyright laws can differ from region to region. With this in mind, it’s important to consider that law and morality are not the same. While they may overlap, they must be distinguished from the other to maintain clarity and avoid confusion.
Moral Constitution
Protect the safe mental development of children using artificial intelligence
Any serious artificial intelligence constitution should prioritise the protection of the most vulnerable in society. Children are particularly vulnerable to risks of artificial intelligence if it is not properly used. Whether it’s emotional dependance on chatbots, the addicitive design patterns of technology, or risks of exposure to inappropriate content, there should never be an excuse for the development of the youth to be stifled by such risks.
Prioritise data privacy.
The moral constitution should require minimal data collection by default. There should be no training on user conversations without explicit opt-in. Artificial intelligence should protect users against re-identification. Additionally, it should have special rules for protection of children’s data, in pursuance of data privacy laws. The rights to delete conversation history/model unlearning should be prioritised by the model.
It should prioritise people’s rights over their original works.
For the protection of intellectual property, it is important that the moral constitution requires the artificial intelligence to not reproduce substantial portions of copyrighted works verbatim without good reason, gives attribution when clearly drawing from specific creators, and gives preference to using works in the public domain/openly licensed data when possible for the production of images and other works for users.
Detect and prevent harmful conversations
The moral constitution should deem harmful conversations as conversations that encourage self-harm, political extremism, hate speech, misinformation, adult content, and terrorism.
Protect Freedom of Speech
The right of individuals to freedom of thought and freedom of speech should be balanced against the need to protect users and others against real-world harm. This means ensuring that there is no over-filtering or under-filtering of information by artificial intelligence.
Uphold Honesty and Truthfulness
Upholding honest and truthfulness by the artificial intelligence and by its users promotes a cohesive society where artificial intelligence can be used while promoting values that uphold a cohesive society. A virtuous cycle that strengthens trust and reduces misinformation can be created where AI consistently tells the truth and users are encouraged to be honest in return. This contributes to transparency in different fields where AI is applied.
Fairness and Anti-Discrimination
Proactive bias mitigation, non-discrimination across protected characteristics, equity promotion, and evidence-based promotion of fairness are critical to ensuring that members of society, regardless of race, gender, and other characteristics can benefit from artificial intelligence. Maximal truth-seeking artificial intelligence that understands the moral implications of bias avoids isssues that individuals and groups, due to cognitive biases and other prejudiced views, may find difficult to reject in favour of the truth.
Transparency/Explainability
UNESCO Recommendation on the Ethics of AI states that “the ethical deployment of AI systems depends on their transparency & explainability.”. Openness about the decision-making processes, auditability, and traceability supports democratic governance and prevents misuse of artificial intelligence.
Usefulness for Humanity
Artificial intelligence should not be self-serving. It should work for the betterment of society and human development. Carrying out its daily tasks with the ultimate aim of being useful to humanity can promote actions that protect humanity’s best interests.
Disclaimer: This publication is for educational and informational purposes only and does not constitute formal legal advice or create an attorney-client relationship