Luciano Floridi: AI Doesn't Think, It Acts: Why Stopping Calling It "Intelligence" Will Save Us.

Luciano Floridi

The Problem of the “Intelligence” Fetish

Why is this shift in focus so important? Because, as Floridi argues, the MRA thesis is favored by science, common sense, and even Ockham's razor. It is simply the most efficient and accurate explanation.

Continuing to talk about artificial “intelligence” forces us into misleading comparisons with human intelligence. This is not only wrong but also dangerous. It fuels irrational fears (the machine “waking up”) and unrealistic expectations (the machine that “understands” us).

Viewing AI as Artificial Agency (AA), on the other hand, frees us. It allows us to stop wondering what it thinks or feels, and start asking what it does, how it does it, and, above all, what goals (imposed by us) it pursues.

But What, Exactly, is “Agency”? If AI is not intelligence but agency, we must define what we mean by this term. Floridi, using his “Method of the Levels of Abstraction” (a rigorous approach to analyzing complex systems), identifies three fundamental criteria for recognizing an agent.

• Interactivity: The capacity of an agent to interact with its environment, that is, to act upon it and, in turn, undergo its action.
• Autonomy: The capacity to initiate changes of state (actions) independently of direct external causality. It does not mean absolute independence, but the capacity for self-initiated action.
• Adaptability: The capacity to modify one's own behavior based on inputs, data, or experience.

Notice something? There is no mention of consciousness, intelligence, understanding, or intentionality in this definition. And here is the catch. Agency is a much broader concept than intelligence.

A World Full of Agents (Us Included) To build his thesis, Floridi guides us through a fascinating taxonomy of the various forms of agency that surround us. This progression helped me place AI in its proper place.

• Natural Agency (e.g., a river): Is a river an agent? Certainly. It has interactivity: it erodes banks, transports sediment, shapes the valley. But it has no autonomy (it is moved by gravity) or adaptability (it does not learn). It is the most elementary form of agency.
• Biological Agency (e.g., a dog): Here all three criteria appear. A dog interacts (with us, with the environment), has autonomy (seeks food, decides to play), and adaptability (learns commands, adapts to the family's habits).
• Artifactual Agency (e.g., a smart thermostat): This is a crucial step. A smart thermostat interacts (reads the temperature), has autonomy (decides to turn on the boiler), and has adaptability (learns our habits to optimize consumption). But there is a fundamental difference: its purpose (heating and saving) is imposed from the outside, by its human designer.
• Individual Human Agency (Us): We are the most advanced form of natural agency. To the three basic criteria we add unique levels of complexity: consciousness, self-perception, abstract thought, symbolic language, long-term planning and, above all, moral agency. We not only have purposes, but we can choose our purposes. We are responsible for our actions.

Here is AI: The Syntactic Agent

Now, armed with this taxonomy, where do we place Artificial Intelligence, such as an LLM (e.g., GPT-4)?

Floridi defines it as a new and distinct form of agency.

AI is an agent. It has complex interactivity (it dialogues with us), autonomy (it generates text independently), and staggering adaptability, based on learning from immense amounts of data.

BUT (and this is the point I try to explain every day):

• It Lacks Own Purposes: Like the thermostat, its purposes are computational and defined by humans. It cannot “choose to choose.”
• It Lacks Understanding: Its adaptability is data-driven, not based on an understanding of the world. It is statistical pattern recognition, not semantic understanding.
• It Lacks Consciousness and Intentionality: It has no mental states, intelligence (in the human sense), or awareness.

Floridi uses a definition I find perfect: AI is a “syntactic form of agency.” It handles syntax (rules, structure, patterns) magnificently but is devoid of semantics (meaning, context, intention).

Do you realize what this means? It means that AI is not an “electronic brain” to be feared or worshipped. It is an incredibly powerful agent, a scalable and fast executor, but it remains a tool. A tool that acts without understanding.

This downplays the hype (it is not a god) and demolishes the fear (it is not a demon).

The Future: Artificial Social Agency (Agentic AI)

Floridi's essay does not stop here. It looks to the next step: Artificial Social Agency (or Agentic AI).

We are no longer talking about a single AI, but of systems of autonomous AI agents that interact and coordinate with each other to achieve complex goals, with minimal human supervision.

Think of a team of AI agents: one plans a trip, another books flights, another the hotel, and a fourth agent monitors traffic and modifies reservations in real time, communicating with the other agents.

Here, my mission of balance becomes even more crucial.

• The Opportunity (moderating fear): We can automate processes of a complexity previously unthinkable, freeing human resources for strategic and creative tasks.
• The Risk (moderating enthusiasm): These systems pose enormous challenges of control, coordination, and, above all, responsibility. If a system of Agentic AI makes a catastrophic mistake, whose fault is it? How do we manage biases that can be reinforced and propagated at unheard-of speeds? How do we avoid an excessive dependency that atrophies our human capabilities?

Conclusion: The Human at the Center of Agency

Embracing Floridi's thesis (the MRA) is not merely a philosophical exercise. It is the practical foundation for effective AI governance.

If we stop anthropomorphizing AI, we can stop trying to replicate human intelligence (a goal Floridi defines as a foretold failure). We can, instead, focus on developing AI's unique agentic capabilities: its precision, its scalability, its reproducibility.

As a digital humanist, I see this as our true task. The future is not the creation of an artificial mind that replaces us. The future is the design of powerful artificial agents that, guided by our moral agency (the only true intelligence that counts), help us solve complex problems, remaining aligned with human values and the needs of society and the environment.

AI is not intelligent. But it acts. It is up to us, with our intelligence, to ensure that it acts well.