What Are the 4 Types of Artificial Intelligence?
There are four types of artificial intelligence recognized by computer science research: reactive AI, limited memory AI, theory of mind, and self-aware AI. The first two exist and are deployed in businesses today. The last two remain theoretical. Here is what each one means in concrete terms.
Type 1: Reactive AI
This is the simplest form. It stores nothing. It predicts nothing about the future. It reacts to an input and produces an output.
The classic example: Deep Blue, IBM’s chess program that defeated Garry Kasparov in 1997. It analyzed the position on the board and chose the best move. Nothing more. It had no memory of previous games.
Spam filtering systems and basic recommendation engines operate on this principle. Useful, limited, predictable.
For an executive, this is the easiest type of AI to audit and control. No surprises. No unintended learning.
Type 2: Limited Memory AI
This is the dominant type in virtually all enterprise deployments at present.
This type of AI uses historical data to improve its decisions. It learns, but within a defined time window. It does not build permanent memory the way a human does.
Concrete examples: large language models like GPT-4, bank fraud detection systems, recruitment tools that analyze CVs, autonomous vehicles that adapt to traffic conditions.
Two recent initiatives in Morocco illustrate this type. Orange Morocco deployed “Live Intelligence”, presented as a sovereign generative AI solution for businesses. ABA Technology launched an AI platform designed and built in Morocco. Both projects fall, in my editorial interpretation, under the limited memory AI category, which encompasses all generative and machine learning systems available commercially.
This is the type of AI on which you are making investment decisions today. It automates processes, analyzes data, generates content, and is beginning to restructure entire functions within your organizations.
If you want to understand which tools of this type are actually being used in business, I analyzed the 5 most used AI tools in business in 2026.
I have built a diagnostic framework to assess your organization’s AI maturity across these first two types, which are the only ones that concern you operationally. Download the Board Pack AI 2026.
Type 3: Theory of Mind
This type does not exist yet. It refers to an AI capable of understanding the emotions, intentions, and mental states of the humans it interacts with.
Not just detecting that a customer is frustrated based on keywords. Understanding why, anticipating their reaction, adapting its behavior accordingly, the way an experienced manager would.
Advanced research is underway in this area, particularly around multimodal models that combine text, voice, and image. But no commercial system meets this definition.
For a CHRO or CEO, this type of AI raises fundamental questions: if a machine understands human emotions, who is responsible and accountable for the decisions it influences? This is an AI governance question that boards will need to address well before the technology matures.
Type 4: Self-Aware AI
This is superintelligence. An AI with its own consciousness, desires, and understanding of its own existence.
It does not exist. It may never exist in this form. But it shapes global AI regulation debates. The GPAI (Global Partnership on Artificial Intelligence) is a concrete example: at the AI Impact Summit 2026, Senegal delivered a formal declaration on these governance issues.
For an executive, this debate is not abstract. The laws being built around this hypothesis will progressively govern your limited memory AI deployments. Prospective regulation on advanced AI risks is already translating into concrete obligations for the systems you have in production.
What This Classification Changes for Your Organization
Most executives I meet conflate all four types in their strategic discussions. They talk about superintelligence risks when they should be talking about AI governance for limited memory systems. They underestimate what they have already deployed.
This classification is not an academic exercise. It helps you ask the right questions.
Types 1 and 2: what systems do we have in production? Who is accountable? What guardrails have we put in place?
Types 3 and 4: what regulations are coming? What impact on our contracts, our hiring, our operating model?
The momentum is accelerating. AI adoption in business remains uneven across sectors, but it is progressing. Companies that structure their approach now will have an advantage over those waiting for the market to force their hand. I analyzed the impact of AI on jobs in 2026 if you want to go deeper on the HR implications.
If you are a CHRO or CEO and want to structure your AI approach on solid foundations, request a free diagnostic.
FAQ
What is the difference between weak AI and strong AI?
Weak AI refers to all current systems: they are designed for a specific task or a defined set of tasks. Strong AI, also called artificial general intelligence (AGI), refers to an AI capable of reasoning about any problem the way a human can. It does not exist yet.
Is generative AI a separate type of AI?
No. Generative AI is a technique, not a type in the sense of this classification. Generative models like GPT-4 or the solutions deployed by Orange Morocco fall under type 2: limited memory AI. They learn from historical data and generate outputs based on that foundation.
When will theory of mind AI be available for business use?
No reliable timeline exists at this point. Researchers are working on components of this type of AI, particularly understanding emotions and intentions. But a complete commercial system qualifying as type 3 is not expected in the near term.
Should I worry about superintelligence for my decisions today?
Not directly. What concerns you is the regulation being built in anticipation of these risks. Bodies like the GPAI, where Senegal delivered a formal declaration at the AI Impact Summit 2026, show that this regulation is taking shape internationally. It will eventually govern your current deployments. That is the real strategic issue.