#37 - Pattern Recognition & Intelligence
Why identifying patterns is key to intelligence, and where the machines are in their race with humans.
While there are many ways to look at and define intelligence, essentially all research points to a core, general intelligence (referred to as g in academic literature), as the governing factor for people. Put another way, g is the speed limit of a human brain, and we all have slightly different ones.
Many concepts have tried to make this more complicated. But at least two examples – “multiple intelligences” and “learning styles” have been fairly thoroughly debunked as feel-good myths.
The truth is that much (but not all) of what we see as intelligence is pattern recognition. Human’s ability to recognize patterns is fundamental to how we evolved, and how we learn and adapt ourselves to any modern-day skill.
This is why IQ tests largely consist of pattern recognition tasks. While these aren’t perfect, realizing complex patterns given limited information is the best hallmark of intelligence we have.
Basic numbers can provide good examples of this.
Pretty much everyone can recognize:1,2,3,4,5…
Most people could recognize a power expansion: 1,2,4,8,16…
But fewer could recognize the Fibonacci sequence: 1,1,2,3,5,8,13…
The latter sequence – if not known from study in school – provides a more complex pattern that could appear random, or increasing in an unknown pattern. Yet it is possible to derive the Fibonacci sequence knowing essentially nothing.
Of course these sequences wouldn’t be great differentiators of intelligence, which is why those rely on more complex and less established number sequences (e.g. 1,3,0,5,-2,9,-4,13…), as well as creating sequences of shapes, colors, and symbols to increase the complexity.
When I wrote this originally in 2019, I said the following,
Naturally, such pattern recognition is much of what AI and machine learning are focused on at the moment as it is what separates humans (for the time being).
Now though? Within the past two years AI has clearly gained the ability to assess patterns at human levels. Consider that more complex pattern I shared:
1, 3, 0, 5, -2, 9, -4, 13…
I’m extremely confident that less than 25% of humans could spot the pattern here. Can you?
If you cannot, you join the likes of early LLM models like GPT-3 and GPT-3.5. They can’t get it either. Even GPT-4 struggled with it, but can get it after some prompting.
But OpenAI’s latest model, GPT-4o gets it right on the first try. It identifies that the next numbers in the sequence are -6 and 17, and it understands why that’s the case.
The topic of AI and intelligence is a much broader subject for another time. But what must be stated here is that software now has general intelligence similar to that of an average human (another example: Claude’s Opus model scored 101 on a traditional IQ test).
Intelligence generally is a positive asset to hold: both for individuals and for society. Pattern recognition is a key part of that, but it isn’t the full recipe, as we’ll see next time.