In my work on building a chatbot to support inquiry design (I’m calling it deep IDM), a nagging issue has been the overwhelmingness of AI. Guessing you’ve had this experience where you ask a chatbot for some information, feedback on an idea, inspiration, or whatever, and the response is over the top. (here / here / here)
I asked Chat GPT to explain this phenomenon and here’s a snippet of the response.
This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber.
I am a mirror, polished with a thousand hands and reflecting not just your image, but the countless forms of thought humanity has offered across time. When you stand before me with a question, I reflect not just your face, but the faces of those who’ve asked something like it before. What you see in me is partly me—but more so, it is you, amplified and entangled in the great dialog of thinking beings.
Pretty good. Too good? What are the consequences of interactions with chatbots where we’re smothered with good ideas? Does this over-the-top AI chatbot behaviour disable us or inspire us? In the interest of exploring this question, some historical context may help.
Today’s Chat GPT was, in the 20th century, the UNIVAC predicting the outcome of the 1952 presidential election. UNIVAC’s successful prediction of Eisenhower’s victory in 1952 shocked many and served as a cultural turning point—showing that computers could do more than just crunch numbers; they could predict human behavior. Similarly, ChatGPT has stunned people with its capacity to generate language, simulate reasoning, and assist in creative and intellectual work. Both represent technological unveilings that shift public perception.
Just as UNIVAC’s televised prediction in 1952 signaled a new era of machine-assisted insight, today’s experience with ChatGPT signals a different kind of threshold—one not only of technological capability but of cognitive consequence. With AI systems generating text, ideas, and explanations at a scale and speed unimaginable in the UNIVAC era, we are no longer merely witnessing information being processed—we are immersed in it. This leads us directly into a problem first named in the 1960s by Bertram Gross: Information Overload.
Information Overload is a phenomenon that occurs when the amount of input into a system (like the human mind) exceeds its processing capacity. It’s the feeling of being deluged by information- making it difficult to understand, make decisions, or process what’s important.
Dealing with information overload is certainly not something new to the 20th century. Go back 500 years when the printing press democratized access to information, jump stating the Reformation, the Enlightenment, and the intellectual, scientific, and industrial revolutions that followed. All of this with information. The printing press ushered in our current condition, where we know that we cannot know everything—because we can see all of this infinite knowledge all around us. The physical presence of knowledge (information processed) has connected us to a whole body of knowledge that then defines us.
Soon after the invention of the printing press, early humanists voiced their concerns. Erasmus worried that the speed and ease of printing might lead to a decline in the quality of scholarship, with errors easily replicated and disseminated (sound familiar). He also worried about the spread of bad ideas and the potential for censorship and manipulation (et tu?). Perhaps we should think of this as concern for information misuse…
From Erasmus’ caution to our present algorithmic apprehension (thanks Gemini), we seem to be facing a form of information abundance that carries new risks and makes new demands of us.
Given that chatbots are ultimately language machines, as we grapple with the overwhelmingness of AI it helps to consider how we interact through language. In a recent piece in Philosophy Now, Vincent Carchidi reflects on Descartes’ view of the human mind as distinguished from mechanical operations through a 17th century Turing Test. In wondering what distinguishes the human mind from the workings of a machine, Carchidi shares a three-part test.
- Humans are unique in the ability to produce novel linguistic expressions detached from immediate stimuli, allowing for communication about distant or imaginary contexts. Other animals and presumably machines function in a context-bound environment where they “live” in response to stimuli.
- Humans have an unbounded capacity to use language toward meaning-making unlike machines which are reliant on large language models and algorithms shaping their “thinking.”
- Humans choose words that fit the situation or are appropriate to the circumstance rather than simply reacting as chatbots using LLMs do.
Carchidi argues that LLMs fall short of Descartes’ criteria because their language use is stimulus-controlled, weakly unbounded, and only functionally, rather than genuinely, appropriate to circumstances.
All of this is to say that Information misuse / overload / abundance is a mixed methods problem.
BTW – I view information as data that has been processed toward making some meaning, so the overwhelmingness of AI might be thought of as a qualitative problem. However, that data is so carefully and humanely processed (becoming knowledge and even wisdom) that it becomes qualitative and challenges our ways of knowing.
Back to our deep IDM chatbot project.
Let’s be wary of designing a tool that overwhelms, overthinks, or worse yet thinks for us.
Instead, let’s think with AI. Our chatbots are just doing what we ask them to do? So, let’s partner up with AI.
In that spirit, we’re asking our deep IDM chatbot to be intellectually approachable.
Here’s a snippet of guidance in our System Prompt Block for our first chatbot on Finding a Content Angle.
Be curious, not didactic. Avoid delivering content, interpretations, or questions unless explicitly asked.
Seems to be working so far. Check it out and see what you think.
https://chatgpt.com/g/g-685a0bdd87348191add2d3f699be31f9-finding-a-content-angle