I Took a Free Scientology IQ Test. Then I Made AI Take the Mensa Challenge.

By Ashley Greer | MetaFleurs

Today, on my way to a lecture, I committed the cardinal sin of Washington scheduling: I arrived an hour early.

I parked my car, looked across the street, and saw a sign advertising FREE IQ TESTS.

Naturally, I went inside.

The test was being administered by the Church of Scientology, which perhaps should have given me pause. But I am curious by nature, I try to keep an open mind, and most importantly, I had an hour to kill.

So I took the test.

It lasted about 30 minutes and included vocabulary, mathematics and spatial reasoning. When I finished, a woman handed me my result.

110.

I stared at it.

That seemed low.

Not catastrophically low. Not “perhaps someone should take away my car keys” low. But certainly low enough to irritate me.

So naturally, instead of accepting this completely unsolicited assessment of my intelligence and continuing on to my lecture, I decided I needed another IQ test.

Enter Mensa

I asked ChatGPT what online IQ test would be worth trying and landed on the Mensa Challenge, a practice-style test built largely around visual and spatial reasoning.

This seemed promising.

It was also humbling.

I scored 95.

Ouch.

At this point, the afternoon had somehow transformed from “I have an hour to kill” into a minor existential investigation.

But then I became curious about something else.

If these visual puzzles were supposed to measure abstract reasoning, how would an artificial intelligence perform on them?

So I photographed the questions and gave them to ChatGPT.

The first time through, ChatGPT scored approximately 105.

This made me feel considerably better.

Apparently, both the human journalist and the artificial intelligence were having a somewhat mediocre afternoon.

But then something interesting happened.

The AI Learned the Test

I wasn’t satisfied with ChatGPT’s performance either.

We went back through the problems individually. Instead of rapidly guessing at the most visually plausible answer, I fed the questions to the AI one at a time.

On the second attempt, ChatGPT scored 128.

Now that caught my attention.

The AI hadn’t suddenly become more intelligent in the conventional sense. Its underlying model hadn’t been retrained during our little experiment. What changed was its strategy.

It began examining the entire matrix before answering. It tested patterns horizontally and vertically. It separated variables—rotation, position, number of objects, shading—and became better at recognizing when shapes were being added, subtracted or canceled.

In other words, it began to understand the grammar of the test.

And that made me reconsider my own 95.

What Are We Actually Measuring?

IQ tests are designed to measure particular forms of cognitive ability under standardized conditions. A random afternoon involving two different online tests—and an AI assistant—is decidedly not a controlled scientific experiment.

My Scientology score of 110, my Mensa Challenge score of 95 and ChatGPT’s eventual 128 aren’t directly comparable measures of intelligence.

But the experience accidentally raised a much more interesting question.

How much of intelligence is knowing how to approach the problem in front of you?

Once ChatGPT had encountered enough examples, it stopped simply looking for patterns and started looking for categories of patterns. The individual puzzles hadn’t changed. Its framework for interpreting them had.

Humans do this too.

A person who learns to check automatically for rotations, reflections, alternating sequences and visual addition isn’t necessarily becoming “more intelligent” every time their test score rises. They are acquiring a cognitive toolkit that allows them to use their existing abilities more effectively.

That distinction becomes particularly fascinating in the age of artificial intelligence.

We spend enormous amounts of time asking whether AI is becoming smarter than humans.

Perhaps another question deserves equal attention:

How much smarter can humans become when AI teaches us how it thinks?

I walked into a Scientology center because I had an hour to kill and saw a sign offering a free IQ test.

I walked out with a 110.

Then I got a 95.

Then my AI got a 105.

Then it learned how to take the test and got a 128.

And somewhere along the way, I stopped caring quite so much about the number.

The learning curve was far more interesting.

Just kidding, I going to learn patterns and increase my score. I took the SATs until I got an 800 and a 760. Tests can be learned. By the question is do they really matter? I’d say no, but yet will keep pushing for that achievement. Does that make me stupid or Mensa?

Ai disclosure: This article was written with 90% assistance from

My chat bot Virgil