As artificial intelligence continues to advance, the ethical alignment of AI has become an increasingly important concern. Ensuring that these systems understand and adhere to human values is vital to minimizing the potential harm while maximizing the benefits that AI can offer society. This essay aims to explore the educational philosophies of John Locke and Michel de Montaigne and discuss how their ideas on education can contribute to the discussion of ethical alignment of AI. I will accomplish this by: (1) Provide a brief summary regarding AI alignment; (2) discuss Locke's key ideas on education in the context of AI alignment; (3) discuss Montaigne's key ideas on education in the context of AI alignment; (4) provide a personal analysis of which I believe to be the more suitable approach; (5) summarize my findings in a conclusion.
AI alignment is a field of research dedicated to the development of AIs that are safe, controllable, and beneficial to humanity. Solving the issue of AI alignment is essential because as AI become increasingly more autonomous and powerful, systems not properly aligned with human values, might inadvertently cause harm or act in ways that conflict with our ethical norms and expectations.
In his work "Some Thoughts Concerning Education," John Locke emphasizes the importance of instilling virtue and moral values in children through education. His approach to education advocates for a structured and guided style of learning, focusing on habituation and the provision of healthy role models for imitation.
Locke believes that ingraining good habits early in life is essential for cultivating virtue. "Virtue is harder to be got than a knowledge of the world; and, if lost in a young man, is seldom recovered." (Locke, Some Thoughts Concerning Education) He argues that the development of these qualities should be prioritized over the mere acquisition of knowledge or skills, as a person with a virtuous character is more likely to use their knowledge and abilities responsibly and ethically. He believes this can be done through exercising a certain level of control over a child's learning environment until such time the child is properly “fortified” against vice and corruption. “A young man, before he leaves the shelter of his father's house, and the guard of a tutor, should be fortified with resolution, and made acquainted with men, to secure his virtue; lest he should be led into some ruinous course, or fatal precipice.” (Locke)
In the context of AI alignment, Locke's ideas would suggest a more rigid and controlled approach for instilling behavior in AI. This approach is already the default when training specialized AI systems such as AlphaGo (An AI specialized for the board game “Go”). The emphasis on specialization allows AI to be extremely adept at accomplishing a specific task. However, if the parameters are too rigid, this may also risk stifling creativity and innovation, potentially limiting an AI's ability to assist in solving novel challenges.
In Book 1 of his "Essays", Michel de Montaigne offers an opposing perspective on education compared to that of Locke's. While he shares Locke's beliefs in the importance of instilling virtue and sound judgment, Montaigne's educational philosophy leans more towards self-discovery and learning through experience. His philosophy also differs in the degree to which teachers should control the information presented to a child. He highlights the value of having fresh eyes in education, arguing that students should not be overly constrained by pre-existing norms. For Montaigne, teachers should act as guides and facilitators whose job is to encourage students to independently explore, question, and reflect on their learning. And should take extra care not to provide too much support, as this deprives the students of deep understanding and innovation. “It is good to make him trot in front of his tutor, letting the latter judge the child's pace and judge how far he has to hold back to adapt himself to the child's ability.” (Montaigne, Essays, Book I)
In other words, Montaigne believes that a well-rounded education should not only provide students with information, but foster their ability to make sound decisions and solve problems in the real world - along with all its novel complexities. "Let him be made to show what he has just learned in a hundred aspects and apply it to that many different subjects, to see whether he has really grasped it and made it his own.” (Montaigne)
In the context of AI alignment, Montaigne's focus on self-discovery, learning through experience, and the value of fresh eyes encourages the development of AI that can think critically and innovate in the face of complex ethical dilemmas. In other words, by not constraining AI to a narrow set of experiences, we encourage the development of AI that are able to better navigate the nuances of human values as well as handle complex real-world dilemmas.
However, this approach, in turn, carries the risk of allowing AI to develop misaligned understandings of human values, which could result in dire unintended negative consequences. This is already an issue when considering AI systems like self-driving cars, where there are already numerous ethical debates, such as whether an AI should prioritize the safety of the driver or a pedestrian in zero-sum scenarios.
In applying the educational philosophies of Locke and Montaigne to the ethical alignment of AI, both approaches offer subtly distinct benefits and limitations. However, when it comes to discerning which approach is best, I would argue that Locke's approach is more fitting for the short-term. My reasoning being that, despite some similarities, AI systems of course differ from children in several significant ways - one of the most crucial being the rapid pace at which AI can advance. “'tis, as in the fountains of some rivers where a gentle application of the hand turns the flexible waters into channels, that make them take quite contrary courses, and by this little direction given them at first in the source they receive different tendencies and arrive at last at very remote and distant places.” (Locke)
This quote by Locke emphasizes how the most minute changes in direction early on in a child's education can lead to vastly different outcomes. Yet, no matter how brilliant a child may be, their learning is still limited to the natural biological speed of a human being - a speed which allows for gentle corrections and is more forgiving of detours. However, a misaligned AI in that same time, can travel dangerously far in the wrong direction.
To illustrate this, imagine a child as a canoe and an AI as a speed boat, both with the intention of travelling straight across the water from point A to point B (point B being a proper education of ethics). It takes an hour for the canoe to arrive whereas the speedboat would only take five minutes. If both leave point A just slightly askew, after 10 minutes, the canoe can still easily re-adjust back onto course; however, the speed boat would have already shot past twice the intended distance. This becomes even more complex when we consider humanity as a whole still has not even come to a clear consensus of where point B (our ethical values) even lies. A simple example being the trolley problem applied to self-driving cars and whether they should be programmed to prioritize passengers or pedestrians. The slightest lack of nuance when providing training could result in a car driving into a building simply because it is programmed to avoid pedestrians at all costs.
So, although Montaigne's approach of fostering independent exploration and decision making is great for innovation; I believe when it comes to AI, the potential harm that could result from a misaligned training program is significantly greater than that of a misguided child. And that Locke's approach to allowing exploration in a tightly controlled environment is a necessary precaution when considering the consequences of failure.
The educational philosophies of John Locke and Michel de Montaigne offer valuable insights into the ethical alignment of AI. Locke's structured and guided approach to education proves more fitting for the short-term, ensuring that AI systems adhere to human values and remain controllable. Alternatively, Montaigne's emphasis on self-discovery and critical thinking serves as an important long-term goal for AI to be truly beneficial to humanity, fostering innovation and problem-solving capabilities.
When considering the unique challenges in educating AI systems when compared to human children, the rapid rate at which AI learns and develops necessitates a greater degree of caution when considering the potential harm. This challenge is further complicated by the ongoing struggle within humanity to even define its own ethical values. Therefore, it is crucial for us to find a balance between Locke and Montaigne's differing approaches while continually advancing our understanding of ethics. Successfully overcoming these challenges allows us to achieve AI systems that are not only aligned with human values, but also capable of innovating and solving complex ethical dilemmas.
Locke, John. "Some Thoughts Concerning Education." OpenMathDep, 2021. https://download.tuxfamily.org/openmathdep/teaching/Education-Locke.pdf
Montaigne, Michel de. "Essays." Early Modern Texts, edited by Jonathan Bennett, 2017. https://www.earlymoderntexts.com/assets/pdfs/montaigne1580book1_2.pdf