What Are the 4 Types of Artificial Intelligence?
There are four types of artificial intelligence: reactive AI, limited memory AI, theory of mind, and self-aware AI. Today, businesses almost exclusively use the first two. The last two do not yet exist in operational form. Here is what each type means concretely for an executive.
Type 1: Reactive AI
This is the most basic type. It reacts to an input and produces an output. No memory. No learning. No context.
The classic example: Deep Blue, IBM’s program that defeated Kasparov at chess in 1997. It analyzed the position on the board and chose the best move. It remembered nothing from previous games.
In your business today: spam filtering systems, basic recommendation engines, some first-level conversational agents. They do one thing, well, without context.
Operational limit: as soon as the situation falls outside the predefined scope, the system fails. Zero adaptability.
Type 2: Limited Memory AI
This is where everything happens in 2026. This type of AI uses historical data to improve future decisions. It learns, within a defined time window.
Concrete examples: autonomous vehicles analyzing the behavior of surrounding drivers, language models like GPT-4 or Gemini, bank fraud detection systems, candidate pre-screening tools in recruitment.
In Morocco, Orange Maroc stated, according to FNH.ma, that AI is a lever for internal transformation and value creation for its clients. The Concentrix-Ipsos Observatory separately reveals massive adoption of these tools in Moroccan companies, with conversational agents and personalization systems among the most widespread uses.
This is also the type of AI powering the tools your teams are already using, often without any formal framework. Kaspersky has flagged massive and poorly governed use in Morocco: in many organizations observed, this is precisely limited memory AI operating without any established AI governance policy.
As I analyzed in my article on AI projects in Morocco in 2026, the majority of local deployments rely on this second type. This is the real playing field for executives today.
If you want to structure your approach and assess where you stand across each type, download the Board Pack AI 2026. It is a diagnostic framework designed for executives, not technical teams.
Type 3: Theory of Mind
This type does not yet exist in production. It refers to an AI capable of understanding the emotions, intentions, and beliefs of a human interlocutor, and adapting its behavior accordingly.
Not a conversational agent that simulates empathy. An AI that genuinely understands why you are angry, what you actually want, and what you left unsaid.
Researchers are working on it. Some prototypes show partial capabilities. But no system deployed at scale meets this definition today.
For a CHRO or CEO: this type of AI would radically change the manager-employee relationship, recruitment, and negotiation. But it is not coming tomorrow morning.
Type 4: Self-Aware AI
This is the ultimate level. An AI that is aware of its own existence, its internal states, its limitations. It does not merely process data: it understands itself.
This type does not exist. The philosophical debates around its very definition remain unresolved, and it still belongs to the realm of fundamental research.
Why mention it then? Because many executives confuse this level with what their current tools actually do. ChatGPT predicts the next word. That is powerful. It is not the same thing as self-awareness.
Understanding this distinction prevents you from overestimating your tools and underestimating real risks, particularly around AI governance and regulatory compliance.
The challenge for Moroccan businesses is not whether self-aware AI will arrive someday. It is mastering type 2 before unstructured AI creates incidents you never saw coming. As I explained in my analysis on which jobs will survive AI, the question is not technological. It is organizational.
If you are a CEO or CHRO and want to know exactly which types of AI are circulating in your organization, request a free diagnostic. We look together at what is deployed, what is governed, and what is not.
FAQ
What is the difference between weak AI and strong AI?
Weak AI (or narrow AI) is designed for a specific task: playing chess, recognizing an image, translating text. Every system deployed today is weak AI. Strong AI refers to an AI capable of reasoning about any problem the way a human would. It does not yet exist.
Do the 4 types of AI correspond to weak AI, strong AI, and super AI?
Not exactly. The four-type classification (reactive, limited memory, theory of mind, self-awareness) describes the level of cognitive sophistication. The weak/strong/super AI distinction describes the scope of capabilities. The two classifications overlap but are not identical.
Which type of AI does my company use today?
In most organizations observed: type 2, limited memory AI. This powers text generation tools, recommendation systems, conversational agents, and data analysis tools your teams use daily.
Is self-aware AI a risk to anticipate now?
No. What needs to be anticipated now is the governance of the limited memory AI already embedded in your processes. That is where the real risks lie: data leakage, biased decisions, regulatory compliance. The alerts from Kaspersky and the findings of the Concentrix-Ipsos Observatory address precisely these current uses, not hypothetical scenarios.