Unpacking the Suitcase: What Do We Really Mean When We Say "AI"?

If you ask ten different experts to define “Artificial Intelligence” in 2025, you will likely get ten different answers.

To a gamer, it’s the NPC (Non-Player Character) flanking them in Call of Duty. To a data scientist, it’s a stochastic gradient descent algorithm minimizing a loss function. To a Wall Street analyst, it represents a multi-trillion-dollar market cap shift. To a sci-fi fan, it’s a robot with a soul. To a regulator, it is a black box of liability.

Why is this one term so slippery? Why does it seem to mean everything and nothing at the same time?

The answer lies in a brilliant metaphor coined by cognitive scientist and AI pioneer Marvin Minsky in his book The Emotion Machine. He called terms like “intelligence,” “consciousness,” and “thinking” suitcase words.

Much like a travel suitcase, the term “AI” is just a handle. It is a convenient container that we use to carry around a jumbled collection of vastly different algorithms, mysteries, technologies, aspirations, and mathematics.

In this deep dive, we are going to pop the latches, open the lid, and inventory exactly what is inside the suitcase of AI—and importantly, what is missing.

1. The Origin of the Suitcase (1956)

The suitcase was packed for the first time in the summer of 1956. A group of scientists including John McCarthy, Marvin Minsky, and Claude Shannon gathered at Dartmouth College. They coined the term “Artificial Intelligence” based on a bold conjecture:

“Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”

They thought they would solve the problem in a summer. They were wrong. But they gave us the “handle”—the term AI—that we still lug around today, even though the contents of the bag have changed completely.

2. Unpacking the Inventory: The Three Layers

If we open the “AI” suitcase today, we find it packed with distinct tools that often have nothing to do with each other. We can categorize them into three historical layers:

Layer A: Symbolic AI (The “GOFAI” Era)

At the bottom of the suitcase are the oldest tools: Good Old-Fashioned AI (GOFAI).

• The Logic: This is rule-based. Humans explicitly program logic into the machine. If X happens, then do Y.
• The Strength: Perfect for defined systems like Chess or Tax Accounting.
• The Weakness: Brittle. If you change the rules slightly (e.g., put a cat on the chessboard), the system breaks.
• Item in the Suitcase: Expert Systems, Search Algorithms (A)*.

Layer B: Machine Learning (The Statistical Turn)

In the 1980s and 90s, we realized we couldn’t program every rule for the real world. So, we stopped writing rules and started feeding data.

• The Logic: Instead of programming the computer, we program the computer to learn from data. It finds the statistical correlation between “Input A” and “Output B.”
• The Shift: From “Logic” to “Probability.”
• Item in the Suitcase: Random Forests, Support Vector Machines, Recommendation Engines (Netflix/Spotify).

Layer C: Deep Learning & Generative AI (The Neural Revolution)

This is the top layer, the one currently bursting out of the suitcase. Inspired by the biological brain, these are vast networks of artificial “neurons.”

• The Logic: Representation Learning. The machine learns to identify features (edges, textures, shapes, concepts) without being told what they are.
• The Transformer: The specific engine inside modern AI (like GPT). It utilizes a mechanism called “Self-Attention” to understand the relationship between every part of the data simultaneously.
• Item in the Suitcase: LLMs (Large Language Models), Diffusion Models (Midjourney), Computer Vision.

3. The “AI Effect”: The Moving Goalpost

Here is the strangest phenomenon regarding this suitcase: As soon as we understand how a tool works, we take it out of the suitcase.

This is known as the AI Effect or Tesler’s Theorem, named after computer scientist Larry Tesler. It states:

“AI is whatever hasn’t been done yet.”

Consider the historical timeline of what we called “AI”:

• 1950s: A computer playing checkers was considered miraculous AI.
• 1980s: Optical Character Recognition (OCR)—reading text from paper—was the cutting edge of AI.
• 1997: Deep Blue beating Garry Kasparov at chess was the peak of AI.
• 2011: Siri recognizing voice commands.
• Today: We don’t call OCR, Chess apps, or spellcheck “AI” anymore. We just call them “software,” “algorithms,” or “features.”

Once the “magic” wears off and we see the math behind the trick, we stop calling it intelligence. We reserve the label “AI” for the things that still seem mysterious (like reasoning or creativity).

4. The Scale of the Suitcase: Data & Compute

While the definition changes, the sheer weight of the suitcase is growing exponentially. We can measure this “weight” in Parameters (the adjustable internal variables in a model—roughly equivalent to synapses in a brain) and Compute (processing power).

The Explosion of Parameters

The size of these models has followed a logarithmic scale that is hard to comprehend.

The Energy Cost

Carrying this suitcase is expensive. The “training” of these models requires massive data centers.

• Training a model like GPT-3 consumed roughly 1,287 MWh of electricity.
• That is equivalent to the emissions of roughly 550 gas-powered cars driving for a year.
• A single query to ChatGPT costs roughly 10x to 15x the energy of a standard Google Search.

This is the hidden item in the suitcase: Physical Infrastructure. AI is not just code; it is copper, lithium, water for cooling, and electricity.

5. The Paradoxes: What is Missing?

When people look at the AI suitcase, they often assume it contains things that aren’t there yet. This confusion arises from two famous paradoxes.

Moravec’s Paradox

Discovered by Hans Moravec in the 1980s, this paradox observes a counter-intuitive truth:

“It is comparatively easy to make computers exhibit adult-level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.”

• Easy for AI: Calculus, Chess, Stock Trading, Medical Diagnosis.
• Hard for AI: Folding laundry, walking over uneven rocks, emptying a dishwasher.

We have packed the suitcase with “High-Level Intelligence” but forgot to pack “Sensorimotor Skills.”

Polanyi’s Paradox

Named after Michael Polanyi, who famously said: “We know more than we can tell.” We cannot explain the rules of how to ride a bicycle or how to recognize a face—we just do it. Because we cannot explain the rules, we cannot program them into Classic AI. We had to wait for Deep Learning to solve this by letting the machine learn the rules we couldn’t articulate.

6. The Dangerous Items: The Black Box

There is one item in the suitcase that worries experts the most: The Black Box.

In traditional software (Layer A), if a program failed, a human could look at the code and see exactly line-by-line why it happened. In Modern AI (Layer C), we don’t write the code. We write the architecture, and the AI writes its own internal representation using billions of floating-point numbers.

If an LLM hallucinates a legal precedent or denies a loan application, we often cannot mathematically trace exactly why it made that decision. We are carrying a suitcase, but we don’t have the key to open the inner-most compartment. This is the Interpretability Problem.

Conclusion: ANI vs. AGI

The final distinction to make when carrying this suitcase is the destination.

Currently, the suitcase is full of ANI (Artificial Narrow Intelligence). These are tools that are superhuman at one specific task (folding proteins, playing Go, writing copy) but useless at others.

The dream—and the fear—is AGI (Artificial General Intelligence). This would be a system that can transfer knowledge from one domain to another, reason about the world, and improve itself.

Right now, AGI is not in the suitcase. We have a bag full of incredibly sophisticated pattern-matching calculators. They are powerful, they are changing the world, and they require immense responsibility to wield.

But the next time you hear a headline screaming “AI achieves consciousness,” remember Minsky. Remember the suitcase. Ask yourself: “Did they actually invent a mind? or did they just pack a really, really efficient auto-complete into the bag?”

The suitcase is heavy enough as it is. We don’t need to fill it with imaginary magic.