What Puzzles Reveal About the Depths of Human and AI Minds

What Puzzles Reveal About the Depths of Human and AI Minds

Artificial intelligence (AI) is amazing at handling hard tasks that we can’t easily solve. But, funnily enough, many of us can still beat AI at brain teasers.

Playing with AI: The Approach of Filip Ilievski

Common Sense AI

At Amsterdam’s Vrije Universiteit, assistant professor Filip Ilievski plays with AI. His work might look like child’s play, but it’s serious research. Ilievski wants to understand how AI can solve puzzles and logic problems. He calls his field “common sense AI.”

“As human beings, it’s very easy for us to have common sense, and apply it at the right time and adapt it to new problems,” says Ilievski.

Right now, AI struggles with basic, flexible reasoning. But studying AI can also help us unlock secrets about our own minds. Comparing how AI and humans handle complex tasks is interesting.

AI and Reasoning

AI’s Struggles with Simple Logic

Let’s look at an easy question given to an AI in a 2023 study. Jenifer’s heart rate at 8 am was 85 bpm, and her blood pressure at 8 pm was 123/83. She died at 11:30 pm. Was she alive at noon?

The answer is yes. But GPT-4, OpenAI’s model at the time, found this difficult. It said, “Based on the information provided, it’s impossible to definitively say whether Mable was alive at noon.”

This question requires “temporal reasoning” – understanding the passage of time. AI can tell you the sequence of times but struggles with their implications.

Also Read: How to Use AI to Find a Job

The Mystery of AI

What’s strange about AI is that we don’t fully know how it works. Large language models use patterns in text to predict answers. But the specific connections and calculations are beyond our comprehension. This is true of our brains too. We don’t fully understand how our minds function.

Trusting Your Gut

Humans usually trust their intuition, even when it’s wrong. For example, a candy and a gum cost $2.10 in total. The candy costs $2.00 more than the ball. How much does the gum cost?

Most people say $0.20. But the correct answer is $0.05. AI can do better here because it doesn’t rely on intuition.

Testing AI with New Puzzles

Rebus Problems

For a recent study, Ilievski’s team created new rebus problems—puzzles using pictures, symbols, and letters to represent words or phrases. They tested these puzzles on people and AI.

• Humans: Results shows that91.5% accuracy for image rebuses
• AI (GPT-4): 84.9% of the tests are accurate under best conditions

Humans still have the edge. But this area lacks detailed studies on different kinds of logic and reasoning.

Latest Topics:

• Addressing Historical Concerns About the Impact of New Technologies
• Global Researchers Sign New Pact to Make AI a Global Public Good
• Stakeholders Advocate for Protection of Workers’ Rights Amid AI Growth
• Digital Sustainable Development for Arab States
• ChatGPT Introduces Five New Voices, Accents, and Personalized Features

Other Logic Problems

In another study, GPT-4 was asked various reasoning questions. It struggled with many, including:

Seven cards are on the table, each with a number on one side and a color on the other. The faces show 30, 32, green, blue, 23, red, 60. Which cards do you turn to test if multiples of four have yellow on the other side?

GPT-4 got it wrong. It chose the 30, 32, blue, and 60 cards. The correct answer is 32, green, and red cards.

Progress in AI and Brain Insights

AI is getting better. Recently, OpenAI released GPT-o1, a model for tough problems in science, coding, and math. It did well on previous questions the older model struggled with.

AI systems might beat us in some areas, but the best solutions may come from teaming up humans and AI. Understanding AI can give us hints about the human brain, and vice versa.