John Stuart Mill on Genius and AI Tools
In a short period of time, new, consumer-facing, generative AI tools have exploded in capabilities and applications. Anyone with an internet connection and some time to kill can now use these tools to produce realistic images, short summaries of online content, and even computer code with very little background knowledge.
While the potential possibilities of these new tools are captivating, there are also concerning risks. While there are broader social and political concerns at play, I have in mind risks to the individual using these tools, their capacity to reason and understand, and to exhibit what John Stuart Mill calls “genius” in his essay On Genius.
Mill’s essay is responding to another, anonymous essay in The Edinburgh Review that lamented the narrowing opportunities for the exercise of human genius. As more facts are discovered, more things known, Mill’s interlocutor argues, there are fewer things left for any human to discover. There are fewer opportunities for a Newton or a Galileo to make great, world-changing discoveries.
Mill’s response is that his anonymous interlocutor has misunderstood what it means for human beings to exercise genius. Genius is, Mill argues, a distinctly human faculty of understanding, that allows us to not just take in information and regurgitate it by rote, but to actually understand it and to incorporate it broadly. If you have ever had a “lightbulb moment,” where something clicked that had mystified you previously, you, Mill argues, have exhibited genius in a meaningful way.
What Mill fears is that certain forms of education can train us away from genius and towards mechanical, mathematical forms of thinking. Think of someone who listens faithfully to “a man who knows” and carefully parrots back everything that they have heard. Does the parrot person in this scenario “know” what their teacher knows? Not necessarily, according to Mill. In order to know and to understand, Mill argues, “I must place my mind in the same state in which he has placed his; I must make the thought my own thought; I must verify the fact by my own observation, or by interrogating my own consciousness.”
Working against this self-interrogation is the parrot type of education, what Mill calls “Cram.” When we cram, we are asked merely to intake facts, repeat them faithfully, and move on to the next set of things to be memorized. This type of education stunts us intellectually. It does not encourage, and in fact discourages, personal understanding. It is not the number of facts that are known, Mill argues, but the way we go about knowing facts that restricts the opportunities to exercise genius.
How does this relate to AI tools? We risk, in leaning heavily on our new AI tools, forgoing our ability to understand in exchange for our ability to produce simple outcomes we desire in the moment. While I may be able to complete my homework assignment in a data analytics course using code provided to me by Chat-GPT or similar tools, in completing that assignment, I will not have arrived at a better understanding of how coding works or how to analyze data well. In the same way, if I ask a large language model to write my essay for me, I have failed to understand the material. I have simply substituted the output of an elaborate machine for my own understanding. By relying on these tools I may, in other words, develop into a good worker, but I will be lacking in something characteristically human, according to Mill.
As my friend John Ehrett has recently lamented, “One of the sad things about the ‘digital generation’ is that they don’t get to learn how computers actually *work* because apps do everything with the press of a button.” Where early computer users were forced by necessity to grow to understand how their computers functioned, new developments shield the user from anything that would require understanding in favor of simple, streamlined, eye-pleasing interfaces. Where early computer users learned habits of mind that serve them well in problem-solving going forward, the younger generation seems to understand little about the devices and apps that they use.
This more general problem is greatly exacerbated by powerful AI tools. Without some sort of intentional habits and practices, producing a piece of AI-generated art will not aid me in becoming a better artist. Producing a piece of AI-generated code will not aid me in becoming a better coder. Producing an AI generated essay in response to a university course prompt will not make me a better student, scholar, or human being. I may become better at fulfilling a narrow, rote function, but this will be at the expense of this essential human capacity for genius that Mill so prized.