AI Isn’t Making You Dumber, It’s Breaking Your Feedback Loops

AI Makes Smart People Feel Dumber? Cognitive Hygiene in the Age of Intelligent Tools

Dunning-Kruger effect, or the upper-skill inversion, took an ugly turn with the AI revolution. More precisely, that is what makes intelligent, talented individuals distrust themselves, much more than individuals of lower intelligence and proficiency. People with lower abilities tend to overestimate, while people with higher abilities tend to underestimate their competence.

Take a closer look at what the AI does. Clideo AI helps people edit videos into top-notch content. Decktopus AI helps create vibrant presentations. ChatGPT does research. Basically, the latter gives immediate answers to complex questions, few-second responses even after deep researching a complicated topic.

Now individuals who deal much with such models, who have acquired some experience with them, know they are programmed to return an AN answer, but they have not guaranteed you the RIGHT answer. And more so with the problems which can be influenced by bias in politics or research, or, quite honestly, bias of any sort, of the human variety. This brings us to an important question.

Automation bias. Smart humans over-trust confident automated outputs

It is one thing to know and another thing to do. Remember, you had that little voice saying you shouldn’t date that person? That they’re unreliable? Yes, that is the one which we hear now, having faith in AI. As we discussed, smart people are often prone to self-doubt. That flaw is exacerbated with artificial intelligence even further.

This has become known as automation bias. Humans tend to trust AI as it has a vast research net and gives immediate response. However, the pattern results in flaws of another curious phenomenon that smart people rely on heavily. Metacognitive disruption.

Metacognitive disruption

Humans have an interesting skill of metacognition. The ability to experience our cognition. To put it in straightforward words, it makes us see what we know and what we do not know.

When we study, learn something new, our brain is capable of creating a sort of a ‘knowledge loop’ and confirming that we have actually UNDERSTOOD something. That’s when a notion becomes our own, we internalize it and create a neural pathway that helps us to return to that notion later. Kind of like breadcrumbs.

When we trust AI to think on our behalf, metacognitive disruption takes place. In other words, we might recognize a certain notion later, but we can’t really tell if we understand it to a satisfactory level. Do I truly get it? Can I explain it to someone else? Not really. Or at least you think you can’t.

Immediacy kills healthy doubt

Smart people often love to marinate. No, not their brains in a jar. It’s just that immediate solutions are not often the best. This comes in apparent conflict to the automation bias, making the process even more tricky. On the one hand, we don’t trust notions that haven’t stood the time test in our heads. On the other hand, we trust AI that made a conclusion three seconds ago.

AI is quick and assured in responding even when the logic behind it is thin. And even intelligent beings confuse fluency with understanding. Mismatch of result, is it evident? Resulting mismatch is apparent, isn’t it?

The output seems complete, while the in-brain process feels flawed. External confidence overrides internal doubt. But here’s the rub. Doubt is what makes us researchers in the first place. CONSTANT doubt of every ‘established’ notion is what drives science, adventure, and, quite frankly, what makes us human.

Cognitive hygiene necessary, not optional

Cognitive hygiene, or as I like to call it, brain-flossing, is the modern superpower. It’s a recovery mechanism that will let us continue using AI, but without the pesky side-effect of feeling dumb, or, God forbid, turning dumb in the process. It’s easier than one might think, by the way. Try riding a scooter instead of walking for years, and watch your muscles atrophy.

Closing the loops manually is something we should do while interacting with AI. Reintroduce feedback by doing several things, the first one of which is NT accepting quick answers. When your kid comes into the room and claims that they brushed their teeth, do you just accept the premise, or do you check the result? I hope for your sake that it’s the latter. Acceptance comes with confirmation ONLY. Treat your AI tools as a smart kid.  

- Ask AI to expose assumptions.

You will be amazed at how many things it will blurt out as truths, while it can be a faulty assumption. Such as “The majority of people use AI these days.” Look at that expression. It is faulty, becausea hefty portion of global population isn’t even online. (roughly 2.7 billion out of 8.1 overall). See what I mean?

- Force alternative explanations.

Confirmation bias is a b*tch. AI is a mirror, my friend. Ask AI explicitly to confirm the opposite of your assumptions. Think long and hard to prevent flaws in the process. Your brain will be richer for it.

- Explain outputs back in your own words.

Brain function and speech are interconnected, that’s a fact (oh, no you don’t! Don’t take it as a fact, and go check!) Take this as a metacognition exercise and explain the output back to yourself. Do you see flaws? Did you take assumptions as truths? Can you argue the opposite? It will help establish whether the output was cohesive. If it was, you could then internalize it better and close the cognitive loop.

- Defining what a wrong answer would look like.

This is a fun one, as we can’t always tell if some notion is flawed by simply looking at it. It might SOUND right, but not BE right. What does the wrong look like? Same answer. Who knows? Wrongs are so diverse and vibrant, and tempting. Try to establish what wrong looks like before you ask AI, so you will recognize it when you see it.

This is painful, my friend, but necessary. So are dentist appointments. Hence, the importance of hygiene. Practice, and thou shalt be

‍