When working with AI coding agents and tools (or humans in general) we often see this dilemma. People who sound confident are often believed. However, confidence and correctness do not correlate often.
We often cite/quote/blame the Dunning-Kruger Effect for this phenomenon.
Knowing this, how do we counter it?
When dealing with AI agents
Verify the code yourself. Intuition and experience do not lie. A smell in code can be true.
Hallucinations: they look plausible but the parameters may be incorrect. The solution would be to ask the AI to make sure:
Are you sure about this?
Do not ask the AI to confirm. Because, it will confidently confirm its own hallucinations. Instead, check the actual documentation yourself.
The AI can also give you outdated patterns. For this, it is better to use your judgement and see for youself.
Also, treat AI errors as learning opportunities.
When dealing with Humans
AI agents are predictable. Humans however can be not so.
Here are some ways we could deal with it:
- Study body language: they do not lie nor hallucinate
- Build confidence as you move through working with them: trust is built upon repeated behaviors
- Verify, either using known resources or through people you trust
How can you become more confident?
I really do not have a concrete answer for this. I have struggled with confidence in my own ability to make proper, informed decisions.
But one of the best advises I got on this topic hit too close to home. I took the advice to heart because I know the one who gave it had good intentions.
They sad:
Take ownership of your decisions, even when the outcome is uncertain. Arrive at your conclusions independently rather than seeking validation from every available source before committing to a course of action.
In other words, I used to second-guess my own thoughts and decisions. But I believe the solution lies in relying more on your intuition and ensuring it will turn out to be the best in the end. Even if it may not seem to be so.
A decision you own and took by yourself, however fragile, is your decision. This is personal and can often be better than asking a 1000 people before committing to it. After asking several people, it no longer is your original thinking or work.
It applies to AI too, I guess. How can we call it our own when we have already outsourced the thinking, decisions, and the results to AI? That’s a post for another day.
Also:
Action first; followed by confidence.
You do not wait for confidence to build within yourself and then take action. This is a common yet flawed thinking.
TL;DR
I guess what I am trying to say is:
Never miss confidence for correctness.
Leave a Reply