Six Foundations of Math, Their Scientific Power and AI’s Role

Terence Tao, a professor of mathematics at the University of California, Los Angeles, discusses the six essential pillars of mathematics, its historical interaction with science, its role in anticipating scientific developments, and the future impact of AI on math and science.

The Six Essential Elements of Mathematics

Tao organizes his forthcoming book, "Six Math Essentials," around six fundamental concepts that have evolved from intuitive beginnings into sophisticated mathematical fields.

1. Numbers

Numbers are one of the oldest and most useful mathematical inventions, predating written language. They provide precision, allowing for the communication of concepts like quantity, size, and magnitude in a portable way. Without numbers, descriptions would be imprecise and prone to misinterpretation. While humans are not naturally wired for quantitative thinking, understanding numbers is crucial for making informed decisions, from basic agriculture and trade to modern finance and medicine.

Numbers take on a life of their own, leading to abstract study divorced from immediate application. This abstract exploration naturally extends number systems: - Negative numbers: Developed to allow subtraction in all cases (e.g., 3 - 4). - Fractions: Created to enable division. - Irrational numbers: Discovered as numbers that cannot be expressed as a ratio (e.g., square root of two), initially shocking but proving very useful. - Complex numbers: Invented to take square roots of negative numbers, later found to be essential for describing phenomena in electromagnetism and quantum mechanics.

These extensions, often developed for solving equations or practical problems, remarkably became the natural language for describing complex real-world phenomena.

2. Algebra

Algebra is the next layer of abstraction, moving beyond specific numbers to generic placeholders (variables like x and y) and studying the properties of operations themselves (addition, multiplication, etc.). - Commutativity: A property where the order of operation does not change the result (e.g., A + B = B + A). This property is found not only in arithmetic but also in other operations like rotations. - Non-commutative operations: Some operations, like putting on socks then shoes versus shoes then socks, do not commute.

Algebra allows for the transfer of intuition and proofs from numbers to more complex structures like matrices, which are crucial for modern technologies such as large language models. Manipulating equations purely symbolically, without focusing on specific values, is a powerful technique.

Historical Example: Kepler and Wine Barrels Johannes Kepler, an astronomer, observed wine sellers efficiently measuring barrel volumes with a stick. This puzzled him because barrels varied in shape. He used rudimentary calculus (a precursor to the formal theory developed by Newton and Leibniz) to derive formulas for barrel volumes, assuming merchants maximized profit. His calculations accurately matched the barrel shapes used in the market, providing a scientific explanation for an empirical practice.

3. Geometry

Geometry, meaning "measurement of the earth," has ancient roots in practical problems like navigation and distance measurement. It involves understanding relationships between distances and angles. - Similarity: A key geometric law where similar shapes have proportional sides, allowing for the estimation of inaccessible distances (e.g., mountain height, distance to the moon and sun).

The study of geometry led to the realization that there are multiple consistent geometries beyond Euclidean geometry: - Spherical geometry: No parallel lines; lines are great circles that always intersect. - Hyperbolic geometry: Multiple parallel lines can be drawn through a point not on a given line.

These non-Euclidean geometries, initially abstract curiosities, became crucial for Einstein's theory of general relativity, which describes gravity as the bending of space and time. Riemannian geometry, developed by Bernhard Riemann, provided the mathematical language for Einstein's equations, stating that mass and energy create curvature in spacetime.

4. Probability

Probability is the mathematical framework for encapsulating uncertainty in the real world, moving beyond sanitized, predictable problems. - Origins: Initially developed by gamblers seeking to optimize their strategies by calculating odds. - Applications: Now widely used in fields like finance, medicine, and any system too complex to model from first principles. - Universality laws: Despite the apparent randomness, certain common shapes (like the Gaussian or bell curve) emerge in the distributions of many random systems, which can be explained mathematically.

Probability works best for events with many trials. For extremely rare events, new mathematical approaches are still being developed.

5. Analysis

Analysis deals with inaccuracy in measurements (error bars) and infinities. - Error bars: Understanding the uncertainty around numerical values is crucial, as perfect precision is often impossible. Analysis helps quantify how errors shrink with more precise measurements, even if reaching zero error might require infinite precision. - Infinities: While algebra works well with finite operations, infinite operations introduce paradoxes (e.g., rearranging an infinite sum can change its value). Analysis provides tools to reason with infinities carefully, avoiding these paradoxes. - The Infinite Monkey Theorem: Illustrates the unintuitive nature of infinity. An infinite number of monkeys typing randomly will eventually produce any given text (like Shakespeare's works), but the time required grows exponentially with the text's length, making it practically impossible in a finite universe. - Idealized assumptions: In mathematics, it's common to first solve problems with unrealistic assumptions (e.g., infinite resources, zero friction) to understand the ideal case, then use analysis to bridge back to the finite, real-world scenario.

Mathematics allows for cheap failure, unlike other disciplines. This freedom to experiment with incorrect assumptions is vital for problem-solving.

6. Dynamics

Dynamics is the mathematics of change over time, studying how states evolve incrementally and how simple rules can generate complex emergent behaviors. - Examples: - Evolution: Simple rules of reproduction and survival lead to massive diversity and complex predator-prey relationships. - Traffic flow: Individual cars optimizing their movement create emergent phenomena like traffic waves, which can persist even after the initial cause (e.g., an accident) is removed. - Equilibria: Systems can have stable equilibria (return to state after perturbation) or unstable equilibria (move away from state after perturbation). Understanding these is crucial for predicting system behavior (e.g., climate change, where humanity is moving away from a stable equilibrium). - Weather prediction: A triumph of dynamical systems theory, allowing accurate forecasts for several days. - Chaos: Many systems exhibit chaos, meaning small initial differences can lead to vastly different long-term outcomes. Newton's laws could solve the two-body problem (e.g., Earth and Sun), but the three-body problem proved intractable, showing that even deterministic systems can be unpredictable over time. The solar system, while seemingly stable, has long-term instabilities.

In advanced dynamics, even deterministic systems are often modeled using probability due to their chaotic nature, leading to predictions that become increasingly blurry over time.

How Math Solves the Problems of Science

Mathematics is part of a larger STEM ecosystem, starting with curiosity-driven basic research. - Curiosity-driven research: Mathematicians explore patterns in numbers and shapes out of curiosity. - Pipeline to application: These abstract discoveries can later be connected by other scientists to real-world phenomena, leading to useful technologies and commercial applications. - Unreasonable effectiveness of mathematics: Mathematical concepts, often developed for abstract reasons, frequently turn out to be exactly what is needed to understand new scientific phenomena decades later.

Historical Example: The Parallel Postulate and General Relativity Euclid's five axioms of geometry included the controversial parallel postulate. Mathematicians later explored geometries where this postulate did not hold, leading to the discovery of non-Euclidean geometries (spherical and hyperbolic). These abstract curved spaces, initially curiosities, provided the perfect language for Einstein's theory of general relativity, which describes gravity as the curvature of spacetime.

Historical Example: Sphere Packing and Wireless Communication A British sailor's curiosity about efficiently packing cannonballs led Johannes Kepler to propose the hexagonal closed packing arrangement (like oranges in a supermarket). This "Kepler conjecture" remained unproven for centuries. Its eventual proof, requiring computer assistance, opened the door to studying sphere packing in higher dimensions. This seemingly abstract problem found a crucial application in digital wireless communication: efficiently packing bitstrings in high-dimensional discrete spaces is mathematically equivalent to separating signals to avoid interference, allowing for the design of efficient communication codes and determining theoretical limits of data transmission.

Modern Example: Compressed Sensing Terence Tao was involved in the development of compressed sensing. Collaborating with a statistician and an electrical engineer, he investigated a new technique for MRI scans that produced perfect resolution from very few measurements, far exceeding the capabilities of standard methods. Initially skeptical, Tao discovered the mathematical explanation for this phenomenon. It turned out that seismologists and astronomers had independently found similar "ad hoc" fixes for their respective problems (fault line detection, star location from sparse data). Compressed sensing unified these disparate techniques into a general mathematical framework, now widely used in MRI, wireless broadband, and sensor networks.

Why the "Unreasonable Effectiveness"? Tao theorizes that the effectiveness stems from mathematicians' drive to find the most concise and elegant explanations for phenomena. Just as scientists refine theories to be simpler and more accurate (e.g., Einstein's relativity simplifying the concept of time), mathematicians seek the most elegant expression of mathematical ideas. Often, the most concise mathematical description also happens to be the most concise physical description.

The Process of Mathematical Discovery Mathematical discovery is rarely a "eureka" moment. It involves: 1. Trial and error: Trying approaches that fail, identifying obstacles. 2. Simplification: Breaking down complex problems into simpler sub-problems. 3. Exploring negative space: Understanding what doesn't work helps clarify what might. 4. Internalization: Through repeated effort, the solution becomes clear and feels natural, rather than a sudden insight.

This process highlights the importance of making mistakes. An expert is someone who has made all possible mistakes in a narrow field. While mathematics demands high standards for outcomes, the process of discovery thrives on experimentation and "failure."

How AI is Changing Math and Science Forever

Traditionally, science relied on theory and experiment. Later, simulation and big data emerged. Now, AI is transforming all these modes. - Automation: AI can automate experiments, theoretical calculations, simulations, and data analysis, performing tasks at scale much faster than humans. - Loss of insight: While efficient, this automation risks reducing human insight. Scientists who manually perform tasks often gain deeper understanding, discover new phenomena, and make connections that AI might miss. AI can achieve goals efficiently (like a helicopter flying directly to a waterfall), but the journey of discovery (the hike) provides valuable learning and unexpected findings.

AI and Machine Learning Modern AIs are powered by machine learning, which predicts patterns in data. - Regression: A simple example where inputs and outputs are mapped to find relationships (e.g., food intake vs. weight). - Large Language Models (LLMs): Like ChatGPT, LLMs predict the next word in a sentence based on vast amounts of training data. This process, when scaled up, leads to coherent and seemingly intelligent outputs, even though it's fundamentally pattern recognition rather than deep understanding. - Mathematical capabilities: LLMs can answer math questions and even generate proofs, though these proofs may be verbose or lack human-like elegance. They can be guided to check their work and proceed step-by-step, improving accuracy.

Complementary Strengths of Humans and AI - Human experts: Excel in depth, focusing on a few difficult problems, gaining insights, and sharing knowledge. - AI: Excel in breadth, capable of tackling thousands of problems, even if they only solve a percentage. They can find solutions that human experts might miss due to biases or lack of time to explore all possibilities. AI can act as an independent pair of eyes, sometimes revealing surprisingly simple solutions to problems thought to be difficult.

Challenges and Concerns - Overfitting: AI models might create overly complex models that fit existing data perfectly but fail to generalize or extrapolate. - Proof indigestion: AI accelerates the generation and verification of proofs, leading to an overwhelming volume of new results that humans struggle to digest and integrate into existing knowledge. - Nurturing future scientists: AI can solve problems typically assigned to graduate students, potentially hindering the training and development of the next generation of human scientists. - Loss of "slow science": The emphasis on speed and efficiency from AI might devalue the slower, curiosity-driven research that leads to truly original ideas. - Human-AI collaboration: Current AI tools are not yet fluid conversational partners, making deep collaborative work challenging. They are better suited for secondary tasks like literature searches, code checking, or proofreading.

The Future of Science The scientific enterprise faces a critical juncture. While AI offers powerful tools, there's a risk of optimizing for the wrong metrics (e.g., sheer output volume) at the expense of deeper understanding and human development. It's crucial to foster open discussions about the value of basic, curiosity-driven research and the importance of a human community in exploring and sharing knowledge. Science provides clarity and reduces anxiety in a complex world, values that need continued support.