How does artificial intelligence (AI) generate texts?
It starts from a very simple idea: probabilities. It’s pure statistics.
In what sense?
Let’s see: say “The cat hunts mice.”
The cat hunts mice.
Now say “The cat hunts…”.
The cat hunts…
Well, after that “hunt” it is about guessing, as you do when you hear the phrase, the next word: it is not “water”, it is not “meat”… But if you have billions of phrases in your memory in which “the cat hunts” is followed by “mice”…
…Well, mice; It’s that simple?
And, for this reason, AI in texts is much smarter when they are short. When they are long, he starts to hallucinate and makes mistakes. It is less reliable, because the statistics for guessing and squaring content become complicated and require enormous computing power. And growing.
Well, seeing ChatGPT and the like work is chilling.
I wouldn’t call it “artificial intelligence.” We should start looking for another name for him.
Because?
Because it raises too many expectations for what it really is.
It’s artificial, but… is it intelligence?
We don’t even know what human intelligence is…
We know when it is not.
Any bacteria knows how to survive – feed, reproduce, proliferate – as well or better than us… And is it intelligent for that reason? As for AI, they are algorithms that have been known since Babylon.
But it is giving results.
The resulting texts, at least for now, are superficial; but no more than most of what most of us generate daily.
But isn’t AI learning?
Of course: the more data you accumulate at your disposal in your memory and the more computing capacity, the better your results will be in texts and images.
Where will we notice the impact of AI most: in the office, free time…?
What AI already does better than us is recognize patterns: distinguish when they are repeated and when they are not… In very small details that we do not appreciate.
How is that ability most effective?
Where AI is changing the world is in war. Right now, he is already the most effective soldier on the fronts in Ukraine and in Gaza.
How does AI fight in war?
Recognizes image patterns with enormous reliability. If you take a drone or satellite image of the battlefield, the AI ??will distinguish where each soldier is; each battle tank; every artillery battery… every anti-tank mine…
How do you know it’s a gun?
Because most of the attacks that are in the news now have already been executed with the help of AI. We are witnessing a gigantic experiment in real time of the effectiveness of this new technology in winning battles.
An ophthalmologist here said that only AI distinguished between male and female eyes.
That is the capacity that is now used to analyze images with unusual precision.
To improve AI, do we have to use copyrighted images and texts?
And the big platforms are now appropriating those rights to train their AI models without compensating the authors for that work. In the EU, we are already reacting and the fines are coming.
Isn’t so much regulation and European law harming innovation in AI?
On the contrary, what occurs is the Brussels effect: EU countries are so interconnected with the rest of the world that they adopt our regulations or adapt to them to continue trading.
It is easier to regulate than to manufacture and innovate, but aren’t we left behind like this?
From Brussels we can regulate and legislate more on artificial intelligence than the United States precisely because we do not manufacture and manufacturers cannot exercise their lobbying power over European regulators.
It is a consolation, but a modest one.
The first antitrust regulations of the 1920s took more than 30 years to become effective when the trusts of four billionaires controlled all the oil, all the railroads…
Aren’t Google, Microsoft, Apple… the big AI monopolies today?
These types of antitrust measures to prevent four or five companies from taking over an entire sector and subjecting us to their designs take decades to take effect…
We will see.
But if Trump wins, forget about them.
