AI: The Big Picture

The below provides a simple way of looking at AI from six key perspectives. Whether you’re new to AI or communicating with cross-functional teams, these lenses provide clarity on what AI is, how it works, and where it’s going.

Photo by Ecliptic Graphic on Unsplash

1. 🧠 Capability: What can this AI do compared to a human?

Narrow AI: Specialized in a single task (e.g. face recognition, chatbots).

General AI: Hypothetical AI that can reason across domains like a human.

Super Intelligent AI: Theoretical AI far surpassing human intelligence.

2. ⚙️ Functional Behavior: How does the AI make decisions and learn?

The “Functional Behavior” lens helps us understand the decision-making maturity of an AI system — from simple reactive systems to context-aware learning systems, and, in theory, to more advanced forms of intelligence that do not yet exist.

Reactive Machines: It reacts, but doesn’t remember. It responds only to the current input. For example, chess engines determining the next move or rule-based robot reacting to a sensor.

Limited Memory: Uses recent context. May learn from historical data during training and use short-term context while operating. For example, Large language models using conversation context or Self-driving cars using recent sensor history

Theory of Mind (theoretical): Future AI that understands human emotions and beliefs. For example, an AI that truly understands deception, sarcasm, or intent at a human level.

Self-Aware AI (Hypothetical): AI with consciousness and self-awareness.

3. 🌐 Embodiment: Where does the AI live and act?

Virtual AI: Exists in software only (e.g. language models, recommendation engines).

Physical AI: Embedded in machines acting in the real world (e.g. robots, drones).

4. 🧱️ Learning Method: How does the AI learn and improve?

Supervised Learning: Learns from labeled examples (e.g. cat vs. dog images). Expensive labeling. Source of Feedback: Human

Self-Supervised Learning: Generates its own labels from raw data to learn useful representations. (common in LLMs). Massive scale enabling foundation models. Source of Feedback: Data Structure

Unsupervised Learning: Discover hidden patterns or structure in unlabeled data. Source of Feedback: Statistical Patterns

Reinforcement Learning: Learns via trial and error with rewards/penalties. Often layered on top of supervised/self-supervised systems. Source of Feedback: Environmental Rewards.

All modern AI systems are usually hybrids. For example:

• LLMs → Self-supervised pretraining + Reinforcement Learning from Human Feedback.
• Robotics → Supervised perception + Reinforcement learning control.
• Recommenders → Supervised prediction + Reinforcement learning optimization.

The categories are conceptual — real systems combine them.

5. 💼 Application Domain: Where is AI being applied?

Healthcare AI: Diagnostics, drug discovery, robotic surgery.

Finance AI: Fraud detection, trading algorithms, risk scoring.

Retail AI: Personalization, inventory, chatbots.

Manufacturing AI: Robotics, quality control, predictive maintenance.

6. 🏧 Societal Impact: How does AI affect humans and society?

Assistive AI: Supports human ability (e.g. exoskeletons, accessibility tools).

Autonomous AI: Replaces human decisions (e.g. self-driving cars, surveillance).

Ethical AI: Focuses on fairness, transparency, and human-centered design.

✅ Summary Framework: Key Lenses

Capability: How smart or general-purpose is the AI?

Functional Behavior: How does it make decisions and learn?

Embodiment: Is it software-only or does it operate physically?

Learning Method: How does it learn from data?

Application Domain: Where is it being used in the real world?

Societal Impact: What is the effect on people, ethics, and systems?

Photo by Steve Johnson on Unsplash