AI Is Not a Genie: It Is a Test of Power

Between immediate risks and existential threats, artificial intelligence forces us to ask who controls technology and for what purpose.

We already know that any new technology, especially one with the enormous impact generated by the emergence of AI, presents new challenges.
The issue, then, is how to face those challenges.
From the point of view of the ordinary user, the first thing is to know that they exist and to take them into account.
Not to believe that AI is a kind of genie like Aladdin’s, except that it does not come out of a lamp, but from another magical object such as the cell phone, always ready to solve all our doubts.
The other point of view is that of the creator of AI.
Analysts see two categories of risks.
On the one hand, the risks they call “tangible and present.”
On the other, the risks known as “existential.”
Let us begin with the tangible and present ones.
Disinformation and the deterioration of trust
That capacity to produce texts, audios and videos with the appearance of reality, and at almost no cost, which we see every day, is not always understood by the public.
Sometimes the meme-like quality of the product is obvious. At other times, it is not.
And in those cases deception works as such.
To the usual disinformation produced daily in immeasurable quantities, a technology is now added that amplifies it extraordinarily.
Does this not undermine trust in the media and in institutions?
Not to mention the very frequent intention of users to obtain medical, historical or legal data and take it as true, even though AI may invent it without blushing.
Socioeconomic and Bias Risks
The Industrial Revolution had a particular impact on jobs in which physical strength prevailed.
AI strikes another kind of work, such as programming, writing, translation, financial analysis, design…
Why hire a specialist to create a book cover if AI can produce it with extraordinary speed and without charging a cent?
Unemployment, moreover, is increased by the enormous speed of the transition, which may leave millions of professionals without time to retrain.
As for biased information, we must take into account how models are trained.
The data they collect from the Internet reflect prejudices, racism, and often slanted interpretations of history, as well as the guidelines and inclinations of their creators.
It is enough to search the Internet for information about the so-called “recent history” in Uruguay and observe the burden carried by the Internet to verify it.
Suppose an AI is used to qualify someone for a job.
The AI will produce results based on the loaded data and present them as the product of a neutrality that does not exist.
And we are referring to the smaller systems.
What happens with advanced AI tools?
Technical and Security Risks
Those superior versions can be used to discover weaknesses in computer systems at supersonic speed, or to design different types of malware.
Let us not think about infiltrating the Pentagon or the Chinese services. Let us think more humbly of someone introducing a virus into the modest software of our notebook and then selling us the solution.
Geopolitical Risks and Concentration of Power
The development of cutting-edge AI requires investments of billions of dollars.
This implies the concentration of global power and the consequent control of information in the hands of a handful of corporations and state superpowers.
Two visions
Faced with this panorama, the industry is divided into two approaches:
The preventive approach, supported by Anthropic, holds that models must be developed in highly controlled and regulated environments before they reach the market, and that source code should not be revealed because it could fall into the wrong hands.
Because if an AI system becomes extremely intelligent and autonomous, but its internal objectives are not perfectly aligned with human values, it could cause catastrophic harm.
And it is not a matter of fearing a Hollywood creation like Skynet that hates humans, but rather the use made of it by humans.
Let us imagine what the totalitarian massacres of the twentieth century would have been like with an instrument such as this.
That is why Olah understands that building gigantic models without understanding how they work internally is a technical irresponsibility.
On the other side are the “accelerationists,” who propose that it is better to release AI codes and let public scrutiny find their flaws while, in the meantime, they rapidly earn billions, of course.
And speaking of bias, I could not get either Gemini or ChatGPT to produce an image showing the tension between the two currents that was not biased toward accelerationism. So I asked AI to represent the triumph of spirit over the lust for profit, and that is what the composition illustrating this note is meant to signify.

Axes of Analysis
AI and power
Present risks

Continue reading in Global Order and Geopolitics