Shakespearean AI: Alignment by Allegory, Symbolism, and Metaphor
We may learn more about the nature of AI alignment by studying the allegories, symbolism, and metaphors in Shakespeare's works. This post explores the parallels between AI alignment and Shakespearean themes, shedding light on the complexities of aligning AI with human values.
The reader, much like the author, may have found themselves in the past year or so, pondering the nature of artificial intellignce in relation to human conciousness in all its flavors. From this stems many avenues of academic consideration and personal introspection, one of which has gripped me particularly. The notion that these systems may be aligned with human values, or at the very least, not be misaligned with them, is a topic of great interest to me. Now, the human values that we must align ourselves to are not always clear, and the methods by which we may align ourselves to them are even less so. Where I have found the literature lacking in the sciences, I have found solace in the arts, and a richness of direction in the works of William Shakespeare.
I willingly admit that for the first 30 years of my life, any mention of Shakespeare filled my mind only with drudgery and yawns. Immediate redirection of interest followed, to wherever my eyes and ears could find something suitably more modern and interesting. Shakespeare was a relic of the past, a dusty old tome that had no place in the modern world. We have the internet, we are connected, we are global, we are technological. Sure, I could learn the pattern of iambic pentameter, but what good would that do me when I'm trying to write a program that can recognize a cat in a picture? Well, as it turns out, recognizing a cat in a picture is now a solved problem, and the problems that remain are far more complex, more human than I had ever acknowledged.
When I began to consider the nature of these human values, I found myself following the path many prolific mathematicians have found themselves on, that of the philosopher. I found myself reading the works of Plato, Aristotle, Descartes, and Kant. The findings were much more practical than I had expected. It was in finally reading the work of Shakespeare that revealed a gap in my understanding of human values that I didn't know I was missing. The allegories, symbolism, and metaphors in Shakespeare's works are not just a reflection of the human condition, but a guide to understanding it.
In the play "Hamlet," the titular character is faced with a moral dilemma. He is visited by the ghost of his father, who tells him that he was murdered by his brother, who has since married his mother. Hamlet is tasked with avenging his father's death, but he is unsure of the ghost's intentions. He struggles with the decision to kill his uncle, and the consequences of his actions. This moral dilemma is a reflection of the human condition, and the complexities of aligning oneself with one's values.
...
It is my firm belief that a primary benefit of AI for humanity is as a tool enabling individuals to explore for themselves more easily the lessons from those who came before us. With great respect for those lessons our predecessors have learned and prepared for us, we may use their guidance to address the most difficult challenges we face today in all forms.
Allegory, metaphor, and symbolism are powerful tools for understanding the complexities of human values, and aligning AI with them. The subtlety, the nuance, the depth of meaning in Shakespeare's works, is a testament to the power of the human mind and its expression that must be addressed in the course of alignment to human values.
It is to our peril that we ignore the human values which words fail to capture. If our deepest and most fundamental nature is unable to be expressed in word, language, modality alone, then we must acknowledge that this will subsequently never be found in a training corpus. Of the multitude of media we create, capture, and train upon, of all of the feedback we provide and the correction we make to align, never among them will be found the mysteries hiding behind the failures of language. What could the myriad PetaBytes of data fail to capture? The concepts that Shakespeare addresed, the notions evoked in the human mind provoked by his subtle and veiled allegory. All of this data, all of this training, all of this alignment, and yet the most important values of humanity are not to be found.