What AI Teaches Us About Stupid
Generations of genius work has brought us here.
If you are retired and living comfortably, good for you! For everyone else that still works for a living, or depends on someone that does, the future has probably gotten a little nastier.
You can thank the incredible advancement of Artificial Intelligence (AI) in the form of Machine Learning (ML) solutions for the stinky future that is rushing to meet you.
I don’t want to talk about Everyone Losing their Jobs
Eventually just about everyone will. But I don’t want to talk about that here.
Instead, let’s talk about the amazing insight into the mechanics of stupid that some of these incredible machine learning AI solutions are demonstrating to us.
For example, ChatGPT is perhaps the best of the publicly accessible near-creepy AI programs you can chat with today. It will blow your mind; so if you want to be both amazed and frightened feel free to talk to it at their chat web page.
Go ahead — ask it to explain quantum physics, or ask it to write a poem, or ask it to explain a few ways to bake a gluten-free cake. It will shock you with appropriate answers — explained in complete sentences constructed more clearly than most people, even articulate professionals, can dish-out given half a day to word-smith their response.
How is this possible? An explanation into the gears and grease of this machine is too much to get into here — this is not a technical article about machine learning. Ask ChatGPT to explain itself if you want to know more about that.
Instead let’s focus on the stupid answers that this incredible knowledge monster spits out as facts and let’s analyze that — and how it may provide an insight into how some of our own very-human-stupid happens.
An Analysis of Stupid by Example
Question I asked the ChatGPT AI ...
If it takes 10 children 5 minutes to eat 10 apples, how long will it take 5 children to eat 5 apples?
Let me say right now, it is not uncommon for people to get questions like this wrong. You have to think about it for a bit.
The correct answer for the question above is 5 minutes because a child can eat an apple in 5 minutes and each child has only one apple. You could have a million children — as long as each has only one apple the total time will be simply 5 minutes to finish.
However, here is the answer I got from the AI …
If it takes 10 children 5 minutes to eat 10 apples, then it would take 1 child 5 minutes to eat 1 apple. This means that it would take 5 children 5/5 = 1 minute to eat 5 apples.
Another way to solve this problem is to recognize that the time it takes for a group of children to eat a certain number of apples is inversely proportional to the number of children in the group. This means that if you halve the number of children in the group, the time it takes for them to eat a given number of apples will double. In this case, if it takes 10 children 5 minutes to eat 10 apples, then it would take 5 children 10 minutes to eat 10 apples. Since the children are eating 5 apples in this case, it would take them 5/10 * 10 minutes = 5 minutes to eat 5 apples.
This answer is several levels of messed up in a very human-like way.
The answer starts off well — looks like it will win …
it would take 1 child 5 minutes to eat 1 apple
Then we see the logic mistake right after that …
This means that it would take 5 children 5/5 = 1 minute to eat 5 apples.
The AI glibly created the ratio 5/5 to come up with 1 minute to eat the apples. This is a flaw of logic — and it is an insight into how a strong grasp of language can mask a weak grasp of logic.
We can all get confused and take some numbers from a question and then go with a computation that was wrong at its foundation because we did not really understand elements of the question.
The response then moves into the realm of mimicking the cognitive dissonance that happens too often in all our heads: This is when we have already explained something one way, and then without recognizing any flaw in our logic — we go on to contradict our own explanation.
Here is what the AI gave as the answer at the end of the longer second part of its response:
5 minutes to eat 5 apples
Damn, that’s the right answer. It gave us both the wrong answer and the right answer and did not even seem confused.
You’ve Met People Stupid Like This
Hell, very likely you’ve been that person. I know I have been and have met people frequently that are blind to their own cognitive dissonance.
Is the way this machine learning AI makes what looks like a mistake of cognitive dissonance a clue into the clockwork of our own minds?
I think the answer is yes.
How the Stupid Answers Happen
This AI is primarily what’s known as a natural language processor. It is good at interpreting language and composing responses using its knowledge of language — like a black box that seems to think but on some level is not thinking at all.
We are like that sometimes; not thinking at all yet producing reasonable-looking answers. Autopilot responses. Stupid but giving the appearance of being smart.
ChatGPT is a natural language processor at its core, and I believe we human beings are language processors at our core too.
And I also believe a great deal of what passes for stupid can be better understood through this lens.
Ohh, and this machine is learning to fix these stupid mistakes so it will not make them again. Most of us are not fixing ourselves in the same way. We will continue to make these mistakes.
