3 Comments

I am not competent enough to judge the risks associated with the current rapid development of A.I. But I very much like the idea that Thorsteinn is developing of working out how A.I. can help us become better thinkers.

P.S. Don't miss the webinar on this subject on the 26th of June!

https://www.marris-consulting.com/en/conferences/webinar-artificial-intelligence-and-logical-thinking-process

Expand full comment

Hi, Thorsteinn. This is a great piece highlighting how ChatGPT can help us find flaws in our own thinking, rather than replacing that thinking. Question: I'm assuming you can't input a logic tree into ChatGPT. So I'm curious. What your prompt was to test your logic assumptions? I tried those three entities you suggested myself in ChatGPT and got a different response.

Expand full comment
author

Hi Angus. Thank you. You are correct that a visual logic tree cannot be copied into the model as such, but if you make it entirely text based you can do this, though it is not very practical. This is why I prefer to take the cause-effect statements one by one. The key thing when using those models is how you phrase your questions. In this case, for example, I would state the premises and the conclusion and then ask if the conclusion unavoidably leads from the two premises or if some additional premises may be needed for it to hold true. This means you prompt the model to consider that question. If you just plug in the statement and ask if it is valid, the model may just consider the logical validity. Often you may have to ask a number of times and tweak your prompts based on the answers you get.

Expand full comment