I have many conversations with people about Large Language Models like ChatGPT and Copilot. The idea that “it makes convincing sentences, but it doesn’t know what it’s talking about” is a difficult concept to convey or wrap your head around. Because the sentences are so convincing.
Any good examples on how to explain this in simple terms?
Edit:some good answers already! I find especially that the emotional barrier is difficult to break. If an AI says something malicious, our brain immediatly jumps to “it has intent”. How can we explain this away?
After reading some of the comments and pondering this question myself, I think I may have thought of a good analogy that atleast helps me (even though I know fairly well how LLM’s work)
An LLM is like a car on the road. It can follow all the rules, like breaking in front of a red light, turning, signaling etc. However, a car has NO understanding of any of the traffic rules it follows.
A car can even break those rules, even if its behaviour is intended (if you push the gas pedal at a red light, the car is not in the wrong because it doesn’t KNOW the rules, it just acts on it).
Why this works for me is that when I give examples of human behaviour or animal behaviour, I automatically ascribe some sort of consciousness. An LLM has no conscious (as far as I know for now). This idea is exactly what I want to convey. If I think of a car and rules, it is obvious to me that a car has no concept of rules, but still is part of those rules somehow.
Thing is a conscience (and any emotions, and feelings in general) is just chemicals affecting electrical signals in the brain… If a ML model such as an LLM uses parameters to affect electrical signals through its nodes then is it on us to say it can’t have a conscience, or feel happy or sad, or even pain?
Sure the inputs and outputs are different, but when you have “real” inputs it’s possible that the training data for “weather = rain” is more downbeat than “weather = sun” so is it reasonable to say that the model gets depressed when it’s raining?
The weightings will change leading to a a change in the electrical signals, which emulates pretty closely what happens in our heads
Doesn’t that depend on your view of consciousness and if you hold the view of naturalism?
I thought science is starting to find more and more that a 100% naturalistic worldview is hard to keep up. (E: I’m no expert on this topic and the information and podcast I listen to are probably very biased towards my own view on this. The point I’m making is that to say “we are just neurons” is more a disputed topic for debate than actual fact when you dive a little bit into neuroscience)
I guess my initial question is almost more philosophical in nature and less deterministic.
I’m not positive I’m understanding your term naturalistic but no neuroscientist would say “we are just neurons”. Similarly no neuroscientist would deny that neurons are a fundamental part of consciousness and thought.
You have plenty of complex chemical processes interacting with your brain constantly - the neurons there aren’t all of who you are.
But without the neurons there: you aren’t anyone anymore. You cease to live. Destroying some of those neurons will change you fundamentally.
There’s no disputing this.