Holding AI to the Highest Ethical Standards

Contrary to the frenzy of hyperbolic rhetoric preaching artificial intelligence (AI) as a juggernaut crushing whatever is in its path, this article seeks to inquire into the possibility of an “AI for the rest of us” – one that offers a unique possibility for social progress. Truth be told, no matter how impressive AI algorithms are, they remain prone to be marred by disinformation, linguistic incompetence, amorality and faux science. This renders human tuning and judgment decisive factors in the success or failure of the AI journey and in balancing creativity with constraint. Not to mention, the moral compass instilled in humans, regardless of their ethnicity or religion, implicating a unilateral commitment to regard others as one does oneself.

A morality-infused algorithm can steer AI away from the reciprocity of exchange, often coined “do ut des,” and sets the bedrock upon which we can train AI to serve our humanity to prosper. Akin to equipping its algorithms with an inner compass capable of locating the perfect path, it is often a pendulum swinging to signal when the road taken is devious.

Fine-tuning AI’s morality would necessitate giving the “compute” – AI infrastructure or domestic computational power that AI developers desperately seek to develop, train and operate large language models (LLMs) – human character. Having a “humane compute” is the purpose that will drive AI researchers and firms to regularly perform experiments on AI’s moral standards and learn by doing. This process is key to AI’s morality pursuit because it nurtures a compounding effect of knowledge that results from the AI architecture’s ability to scale as more demand of data traffic flows through. Therefore, one sure way for building the humane compute would be more refined and clean data sets. This can be within government’s reach, along with leveraging piles of sovereign data that could generate models of our societies, codes of ethics and the behavior of our civilization. Our government agencies are ripe with data retrofits and represent a gold mine for AI techies to refine its data sets and train ambitious humane AI models. An ethical AI would undoubtedly call for a high level of pragmatism that escapes the legislative rabbit hole. This behooves governments to develop a good understanding of computation policy to familiarize themselves with tech job descriptions prior to working on state-backed AI “ethicizing” projects. 

The AI ethical add-on summons our morals and lays the ground for the standards people owe themselves and one another to preserve self-respect and dignity. The AI ethical add-on can encompass the beliefs that support a certain view of morality. It denotes the behavioral canons by which we judge ourselves and one another. For instance, AI will process the prompt “betraying a friend” as unethical and will grant a higher weight to “means” than “ends.” The language model will translate all wrongdoings as acts of “wickedness.”

Nothing captures this better than an AI that incorporates the curse and mark of Cain for Abel’s slaying or Ambrose Bierce’s satiric work (“The Devil’s Dictionary,” 1911) to understand how the law can be the lowest common denominator, the minimum level of acceptable behavior, and therefore pushes our societies toward higher ethical standards rather than settling for the lowest legal criteria. 

We can achieve this through a reflective AI on the dos and don’ts, which considers the law, but transcends legislation to reach a flawless illustration of the right and wrong. When laws and ethics align, AI sees actions as a matter of law; yet when the notions collide, AI processes our actions as a matter of ethics, for each one’s life is not about “me, myself, and I” but a question of “is it the right thing to do?”

The maxim “be of benefit, do no harm” exemplifies best what can shape our AI algorithms. It means that every step of language model training must have a higher benefit to risk ratio. Otherwise, we should revise it until we achieve this ratio. Similarly, it should respect confidentiality and seek informed consent to maintain beneficence and integrity and promote rigor in its data perusal. We must clearly explain to stakeholders all aspects related to research processes as well as a clear delineation of expected outcomes and contribution to the accumulation of knowledge.

Overall, AI algorithms must put the interest of society above the interest of its creators. Hopefully, this will allow us to enjoy the benefit of an AI encompassing the norms that guide our behavior. An AI that prioritizes trust, accountability, mutual respect and fairness, which are necessary values for cooperation and collaboration that are needed to promote the key aims of our society: knowledge, truth and improving the world.

The case for ethics in AI should set the norms on misconduct and conflict of interest and restore trust in large language models and keep integrity intact. Focusing on ethics first will guarantee we work together as a society and separate us from the algorithm, for it enables us to tend to issues data analysis and experiments cannot answer.