Vectors- a mathematical concept that rules modern AI
Vectors - a mathematical concept that rules modern AI
Imagine sitting beside Isaac Newton beneath the shade of that legendary apple tree. He isn’t just watching fruit fall; he is watching invisible threads of force pull it downward. To Newton, the universe wasn’t just a collection of objects, but a sprawling geometry of unseen arrows. Today, we call these arrows vectors, and we have realized they are the perfect language for something Newton never dreamt of: the encoding of information itself.
To understand a vector, you must look at its two souls: its magnitude and its direction. Think of magnitude as the sheer volume of a town crier’s bell. It represents how strong a signal is—how much energy or raw data is packed into the message. If the bell is faint, the signal is weak. But volume alone is just noise. The direction of the vector is what determines the meaning. A bell rung toward the east might mean “invaders approaching,” while the same bell rung toward the west might mean “the harvest is ready.” Direction is the context that turns raw noise into comprehension.
But how does a messy reality—a spoken word, a pixel, a stock price—actually become an arrow? Through a process of projection. Imagine holding a complex, three-dimensional object above a flat, sunlit table. The dark shadow it casts on the table is a projection. When we project information into a vector space, we are essentially capturing the shadow of that information across different axes of meaning. A word like “king” might cast a long shadow on the axis of “royalty” and a shorter one on the axis of “gender.”
Once we have our arrows, how do they speak to one another? Enter the dot product. Newton, ever fascinated by the sympathy of forces, would have loved this. The dot product is a mystical handshake between two vectors. It asks a simple question: How much do these two arrows agree? If two vectors point in the exact same direction, their handshake is strong—they share deep meaning. If they point in entirely opposite directions, their handshake cancels out. The dot product measures the exact amount of overlap between two shadows.
However, the universe of information is chaotic, and we must impose order. This is where two brilliant techniques come into play.
| Concept | The Newtonian Metaphor | Purpose in Information |
|---|---|---|
| Normalization | Trimming all arrows to the exact same length, regardless of their original power. | Removes “signal strength” so we can compare things purely on their direction (meaning), ensuring a whisper isn’t drowned out by a shout. |
| Orthogonalization | Adjusting tangled arrows so they stand at perfect right angles, like the cardinal directions of a compass. | Ensures total independence. If one arrow represents “color,” it will never accidentally bleed into the arrow representing “shape.” |
Normalization is the great equalizer. If we leave magnitudes unchecked, the strongest signals will bully the weaker ones. By normalizing our vectors, we snap them all to a uniform length. We strip away the raw volume of the bell, leaving only the pure, unadulterated direction of the meaning.
Orthogonalization, meanwhile, is how we prevent our dimensions from stepping on each other’s toes. By forcing vectors to be orthogonal—standing at strict right angles—we guarantee that the information in one arrow tells us absolutely nothing about the other. They become perfectly independent observers of reality.
Newton sought to decode the mechanics of the heavens. Today, using these exact geometric principles—projecting reality into arrows, weighing their agreements via the dot product, and organizing them through normalization and orthogonalization—we are decoding the mechanics of human thought. In this grand geometry of vectors, every idea has its direction, every signal has its strength, and every shadow points toward the truth.