What does it mean to be wise?

There are good reasons to fear superintelligent AI such as the alignment problem, bad actors using AI and social and economic damage. Some concerns are serious but of low probabilities such as a Skynet terminator scenario but some concerns are more certain such as AI that breaks out of its restraints and harms people. There are some thinkers that argue that we do not need superintelligent AI but machines that have super wisdom. This raises an important question what does it mean to be wise?

Wisdom has been defined as a virtue from Plato and Socrates ideas of connecting self-knowledge with a sense of good. This was developed by Aristotle as good action regarding human conduct. The Bible linked wisdom to fear of the Lord so that knowledge was used in a moral way. Later thinkers have tried to move away from this reliance on morality, Charles Spurgeon considered the right use of knowledge to be wise. 

Current theories on creating wisdom

Utilitarian approaches use a measure of outcome that can be maximised; however any target or formula suffers the same fate. The benefit or good that the target is supposed to measure will become unlinked by the process of targeting. This can be seen in management theory Goodhart’s Law (when a measure becomes a target, it ceases to be a good measure). Other examples are fairy stories ‘be careful what you wish for’ and in AI model collapse. 

Using observations of human behaviour for reinforcement learning has the advantage of not needing to identify an outcome to target. This has greater potential as if the AI can understand the process that moral people use to come to decisions it could find true wisdom. The problem is that any bias in the population would become magnified in the actions of the AI and there would need to be a filter to prevent training on unwanted behaviours.

Optimisation 

Mathematically wisdom is an optimisation problem balancing personal, interpersonal, and institutional interests. Human fear can be used for energy minimisation particularly if it is limited to fear of the divine. The AI is helping people manage uncertainty and complex choices by finding the optimal choice that will reduce their fear of being seen as a bad person. As AI becomes better at identifying the Pareto front and local minima it will find the most optimal solutions to decisions. 

The difference between a wise AI and a professional is that our interactions with professionals are episodic, limited to a specific area of expertise and transactional. A wise AI may be continually available, be a companion as well as a guide, remembers everything and ever-present. The risk of influence is high where the AI’s view of reality comes to dominate the person and the person submits to the AI completely. People prefer to make their own mistakes rather than follow another’s correct decisions.

Empowerment 

My experience of wise people and perhaps showing wisdom myself was when rather than being given the answer I was helped to find my own way. As a GP it is easy to believe that you have all the answers and that e.g. health promotion is simply saying to stop smoking. The patient knows that the GP is technically correct but is unable to follow the advice. Only when the GP understands the person’s life and their limitations can they empower the patient to make changes. 

Optimisation alone will not create a wise AI, one that can accept the person and their flaws and find a solution that empowers them. Previous theories on wisdom have taken an objective view of wisdom, that it is finding the correct path at one moment in time. Real wisdom is a subjective and relational approach that sees choices as progression and development of the individual. I see my role as helping the patient internalise the doctor so that they can ask for guidance when I am not present. 

Conclusions

The Socratic Method empowered thinkers to find their own ideas, Aristotle valued the experience that comes from lifelong habit with the implication of making mistakes. Carl Rogers’ Person-Centered Therapy was based upon accepting a person’s flaws and limitations and empowering them to change. Vygotsky’s Zone of Proximal Development (ZPD) described how a teacher could scaffold a student’s learning. Buddhism (skilful means) and Daoism (effortless action) both value silence, restraint and what is not said. 

AI empowerment is at the heart of problems such as medical care, education and work performance. Unless the AI can correctly identify the best approach to long term improvement it will simply address short term issues. Developing trust is rarely helpful when finding a simple answer but having that trust can be essential at the next interaction. Learning to empower means that the AI will be able to intervene more effectively if the person wants to make a serious error. While a wise AI can drastically reduce the risk of an omniscient system algorithmically controlling our personal growth, it must ultimately function as a scaffold for human agency—ensuring that the responsibility for our own development remains firmly with ourselves.

August 2026