Does AI "think"? That's the sort of topic that gets people debating for hours about the nature of "thinking" and all of that. But here's another, possibly more important thing to understand: something can sound certain, that doesn't mean it knows.

What ever happened to, "Don't believe everything you read"?

Throughout our schooling we were taught something basic about research: verify sources. Along came the internet and we knew:

• Chatrooms were full of liars. (We didn't need the term "catfishing" to know some folks were full of BS.)

• Wikipedia could be wrong. (Of course it could. Anybody could edit it.)

• Reddit could be wrong. (Obviously. That was literally some stranger named TacoDad472 giving you his opinion.)

• Facebook could be wrong. (Half the posts were memes, somebody’s aunt had “done her own research,” and there was a decent chance you were arguing with a bot.)

We knew this.

We taught our kids this.

Check the source. Consider who wrote it. Look for corroborating evidence. Think for yourself.

And then artificial intelligence came along and apparently all it had to do was speak in complete sentences.

Suddenly:

“My AI said…”

“ChatGPT says…”

“I asked AI and it told me…”

As though a large language model had just descended from a mountaintop carrying stone tablets. Even though these models may have trained on some of the same sources we know to be piles of heaping dung.

Confidence Is a Writing Style

Here's the thing any writer knows: confidence is a writing style. Mostly people are taught the opposite, to avoid declaratives or absolutes in their writing.

Then came AI.

One of the strangest things about generative AI is how authoritative it sounds.

Ask a question and you don’t usually get:

“Well, BeerLover916 mentioned this thing, but you should probably double-check.”

You get polished prose.

Organized paragraphs.

Maybe even bullet points.

It sounds like someone who knows what they’re talking about.

And it isn't just quoting BeerLover916 and TacoDad472 and so we forget that it might be quoting folks who are quoting other folks and something gets repeated a bunch of times but for all we know the "original" idea might have been one voiced by the beer lover.

But that is partly because producing convincing language is exactly what a language model is designed to do.

Fluency with "pretty words" does not equal knowledge. Confidence is not the same thing as accuracy.

And as much as our English class education might have emphasized grammar as Holy-above-all-else, excellent grammar has never been proof that something is right.

Human beings should probably know this already.

We have, after all, met lawyers, a profession where fluency with pretty words just might pay the bills.

AI Changes the Skill Set. It Doesn't Eliminate One.

There is a persistent idea that AI means young people won't need to learn to write.

Why bother teaching essays when AI can write them? Why learn research when AI can summarize everything? Why struggle through complicated ideas when a machine can explain them?

Well, at the very least the odds are that the person operating an AI in the future will still need words with which to do that.

The person operating an AI still needs to know if the words make any sense.

Words are still how we make sense of the ideas of others, in writing or verbally.

The humans of the future will still need to think, be curious, solve problems, and interact with one another (we hope!).

Let AI get better at building spreadsheets or writing code than a human.

That doesn't make it better at thinking or human judgment.

If anything, it makes it more important that we continue to develop those skills for ourselves.