Confronting Deniers: AI is Just a Tool™

A series debunking classic AI risk denier arguments, with the occasional well-deserved rant.

I wanted the left side to read “TOOL - JUST A TOOL” but I am not that patient with Dall-E.

This article is the first in a series where I will examine some of the most ridiculous arguments put forth by AI risk deniers.
We will start with the classical and sinister claim: "AI is just a tool". How should one respond when confronted with this statement? Thankfully, this is just a play on the meaning of the word "tool", and its total lack of substance makes it quite simple to refute.

AI is not a tool. Whoever is using this argument is either:

• Incompetent in deep neural networks, or

• A liar.

This needs to be made very clear. AI is not a tool, and here is why:

• We do not control deep neural networks, we do not understand deep neural networks. They are artificial brains that we teach and hope will behave as intended. Sometimes they do, sometimes they don't, and you get problems, scandals, and harm. The more advanced these brains become, the less we understand them, the less we control them. Even at a philosophical level, the problem of control is unsolved, possibly unsolvable. Ponder on this question for a moment: who decides what the brains do, the constructor or the user?

• Advanced AI systems are more and more autonomous. Autonomy lies on a quantifiable scale (time or degree of autonomy for example). Fully autonomous drones are, right now, making life-or-death decisions in Ukraine, choosing to kill or spare lives based on loose indications.

• AI displays learning and adaptation. Does a hammer display learning and adaptation?

• AIs can persuade humans. Article. Study EPFL. Study Anthropic. Study MIT. Let that sink in. This is a "tool" that can persuade you to do things for it.

• AIs engage in creative problem-solving, finding novel solutions that humans have never considered. As demonstrated by AlphaFold, AlphaGo, AI-assisted mathematical proofs (Deepmind, Erdos discrepancy problem).

• AIs are integrated into decision making processes - more and more taking decisions without human assistance.

Let's return to my initial claim. The person you hear parroting the "AI is just a tool" argument is likely not very knowledgeable in AI and propagating a narrative that has been strategically implanted by much more sinister actors. Because, make no mistake: this argument is a propaganda weapon wielded by profit seeking individuals and entities.

It is an argument that you will find in the toolkit of industrials, politicians, and lobbyists like Yann LeCun (Head Evangelist for Meta), Arthur Mensch (CEO of Mistral, an AI company that mocks the mere idea of risks), and Cedric O (French lobbyist who conspired to sabotage the EU's AI Act) who have a history of manipulating the truth for their own ends.

Just last week, Arthur Mensch was recorded straight out lying to the French Senate, making such outrageous claims as (translated):

When you write this kind of software, you always control what will happen, all the outputs of the software.

We are talking about software, nothing has changed, this is just a programming language, nobody can be controlled by their programming language.

Given his background, Mensch must be knowledgeable in Deep Learning, hence fully aware that these words are lies and manipulations. Endangering the world's entire population for his own profits, happy to just roll the dice for the rest of us.

Another example of such criminal¹ behaviour comes from Martin Casado, partner at the venture capital firm Andreessen Horowitz, and a famous accelerationist (a movement dogmatically focused on racing ahead towards AGI regardless of risks).

Casado wrote to the US Senate and UK House of Lords:

While proponents of AI safety guidelines often point to the “blackbox” nature of AI models – i.e. that the reasoning behind their conclusions are not “explainable” – recent advances by the AI industry have now solved this problem.

Oh, really? Interpretability has been solved? The last time I checked, it sure seemed like an open problem with no sign of resolution in the next 20 years.

This same statement was also signed by the usual suspects:

• Marc Andreessen, Andreessen Horowitz

• Ben Horowitz, Andreessen Horowitz

• Yann LeCun, Evangelist at Meta

• Arthur Mensch, Mistral

How long will we allow these people to get away with murder?

1. Yes, I mark my words, lying to the US Senate is a crime.