Proof of Humanity
Originally posted on LinkedIn in October 2024.
Working with AI models and, especially, agents calls for robust benchmarks to evaluate their capabilities. Many of you are likely familiar with the Abstraction and Reasoning Corpus (ARC) Challenge, which has become a popular way to benchmark AI models.
As this test is quite hard for SoTA models, I use it regularly to prove to myself that I'm still human.
The Bongard Problems
Recently, Arsenii Zemskov brought an old but fascinating test back into the spotlight: the Bongard Problems, collected by Harry Foundalis.
The Bongard Problems present two sets of shapes, and your goal is to describe what is characteristic of each set so you can discern the two groups. These problems predate the ARC Challenge, and it strikes me how visionary thinkers were, conceptualizing such challenges decades ahead of their time.
A five-year-old vs. problem 32
I printed out a portion of the problem set. After working through a couple of pages, I decided to show them to my five-year-old daughter. I was curious about how she would perceive these problems, given that she hasn't been exposed to IQ tests or other early developmental assessments at her age.
Below is problem number 32. Take a moment to look at it and consider the difference between the two groups.

As someone with mathematical training, one might say: "The figures on the left have sharp projecting angles, while the ones on the right do not."
However, my daughter's interpretation was beautifully different. Her solution:
"The ones on the left — you can see up in the sky. The ones on the right — you can see on the ground."
Fresh eyes
This perspective was a powerful reminder that unfettered human intelligence, especially in children, offers profound and intuitive insights we adults might overlook. Children approach problems without the constraints and biases that we adults often carry.
In the realm of AI and complex models, it's easy to get lost in algorithms and data. But moments like these remind me that human intelligence, in its purest form, is a true source of inspiration.
Perhaps, as humanity continues to develop advanced AI, it should strive to preserve and emulate that childlike wonder and unbounded creativity. After all, the most profound insights often come from seeing the world through fresh eyes.