Earlier this year, I spoke at a philosophy conference at the University of Windsor. The theme was Being and its Technology. The call for submissions didn’t mention AI, but the topic was on everyone’s mind. After my talk, someone in the audience asked me a question:
Is AI actually different from earlier technologies?
The way they asked suggested they thought the answer was no. I can’t remember exactly what I said, but my answer was probably disappointing, given that my presentation focused on the dangers of outsourcing our reasoning skills.
Today, I would answer that question slightly differently.
There are two reasons why AI might differ from earlier technologies:
- The black box problem
- The potential for broad cognitive outsourcing
Neither feature is entirely new, but the combination of them makes AI stand out.
The black box problem
If you look around your house and pick a machine, how that machine does what it does can be explained. Some technologies may be unfamiliar to a particular person, but their operating principles are understood and can be explained by an expert. And based on how the machine works, someone can reasonably predict what it will do while functioning normally.
With a large language model (LLM), there is an algorithmic computation occurring that is understood by some at a system-level, but the specific path from input to output cannot be reconstructed, fully explained, or predicted, even by the people who built and trained the model. My plumber could generally explain the operating principles of my washing machine, identify its components, and usually predict how it should behave under normal conditions.
Broad cognitive outsourcing
Now let’s combine the black box problem with the potential for broad cognitive outsourcing.
Any system with difficult to explain, complex inner workings or hidden logic can be called a black box, but not all of them have the potential to perform a broad range of cognitive tasks like AI. There are other forms of cognitive outsourcing, like how writing outsources memory or clocks outsource timekeeping, but AI is unusual because it can perform many kinds of cognitive tasks, like drafting copy, summarizing, translating, classifying, defining, brainstorming, coding, and supporting or making decisions in a wide range of fields.
A toaster toasts. A drill drills. Some people compare the advent of AI and LLMs to the industrial revolution, but the new machines introduced at that time were designed to do specific things. With AI, there is a large gap between system-level understanding and output-level explanation.
The big question
How AI arrives at an output isn’t fully explainable or predictable, and it can perform a broad range of cognitive tasks. Other than the human mind itself, what else fills these conditions?
If you have a difficult time answering that question, AI may indeed be different from earlier technologies.
Back