Help! I have been working my way through a series of puzzles designed to evaluate artificial intelligence systems by setting tasks that are supposed to be easy for humans to solve but hard for current AI engines to crack. All was going well until I came across one particular puzzle. I simply can’t work out why my answer is wrong so I am hoping that one of my genius students will read this and put me right.
The puzzle can be accessed at https://arcprize.org/play?task=8e5c0c38 and my wrong answer is shown below. I tried the alternative task for this level and I got that right so I really don’t know where I have gone wrong with this one. Please comment or contact me via my out-of-school email address to tell me what I am missing. Thank you.
To provide a bit more context, this puzzle is part of the ARC Prize initiative, which aims to accelerate the development of Artificial General Intelligence (AGI) going beyond just scaling up existing AI models. The tests require the ability for AI to adapt to new problems much as humans do, rather than simply re-applying the same rules that have worked previously.
I should add that the “prize” part of ARC Prize refers to AI engines that can successfully solve tasks like these. There is no “prize” for being a human solver as humans are supposed to find such tasks quite straightforward, which makes my “failure” all the more humiliating
If you want to find out more about ARC Prize, either to attempt the tasks yourself (just for fun) or to test an AI engine that you have been developing (to win prizes) then visit https://arcprize.org/. It’s all free and it’s also enormous addictive – when you don’t get stuck!
UPDATE: Problem solved. Many thanks to JB, who pointed out that the examples suggested trying to preserve the squareness of the shapes (whereas I was trying to move the overall centre line as little as possible). JB’s correct solution is shown below.
Mr Tarrant's physbang 'blog 