The Deceptive Power of Maps (with Paulina Rowinska)
What this covers
Mathematician Paulina Rowinska and economist Russ Roberts examine the hidden mathematics of maps and how distortion shapes human decision-making. The conversation moves from the elementary fact—a sphere cannot flatten to a plane without loss—to its consequences for navigation, logistics, and policy. Rowinska's central claim is that all maps are mathematically-constrained simplifications whose usefulness lies not in accuracy but in fitting the problem at hand. The discussion treats maps as windows into how humans and systems actually function when faced with complexity too large to compute perfectly.
The range spans historical examples, contemporary logistics, and design philosophy. Henry Beck's London Tube map appears early as a victory of topological thinking over geographic fidelity—riders need connection knowledge, not true distances. The Coastline Paradox illustrates why measured length depends entirely on measurement scale and yields no single true value. The Mercator Projection receives careful treatment: its distortion of polar landmass sizes follows from mathematical law, not intent, and no projection escapes trade-offs. In logistics, a UPS driver's routes number more than stars in the universe, so companies deploy secret algorithms producing a "good enough" solution in seconds rather than optimal ones in geological time. The conversation turns on a repeated pattern: when humans attempt theoretical perfection—full route optimization without local knowledge, fifteen-variable triage systems requiring calculators, elaborate trash-sorting bins—usability collapses. Good Enough Over Optimal emerges not as laziness but as engineering realism. Rowinska reflects on self-driving cars' reliance on hyper-accurate maps and how AI will function as an advanced calculator rather than replace human judgment about which problems matter.
Rowinska argues that all maps are mathematically-constrained simplifications of reality whose value depends on understanding what they distort and omit for a given purpose, and that 'good enough' approximations serving human needs often beat theoretically optimal solutions.
- A sphere cannot be flattened to a plane without unavoidable distortion (a mathematical theorem), so every projection trades off distances vs angles vs area
- Coastline length is scale-dependent and has no single true value, yet comparisons remain meaningful if methods are agreed
- Topological maps (subway maps) and 'good enough' delivery algorithms succeed by discarding accuracy in service of usefulness
Every map is a simplification made for a specific purpose; knowing what's included, omitted, and useful determines whether it serves.
- Every map is made for a purpose and is necessarily a simplification that leaves things out; a map with no simplifications would just be the real world, too big to be useful, so the key skill is knowing what a map is useful for, what it isn't, and crucially what is missing.
“every map maker makes a map for some purpose. So, we need to know what this map is useful for, what it's not useful for, and... what's missing?”
- The most powerful infographics and data visualizations—like airplane safety leaflets—contain exactly the amount of information the reader needs and no more, because humans can only process limited information, so the design challenge is selecting what to include and omit to make people act.
“The most powerful are the ones that have the exact amount of information that the reader needs.”
- Colorado is assumed to be a rectangle, but because surveyors marking the border centuries ago made small errors and had to navigate around physical features like ponds and forests, the actual border is not perfectly straight—it has 697 sides, making it a hexahectaenneacontakaiheptagon.
“the line drawn on the map was straight. But then, you have people... actually going out there in the land and marking the border... people make errors: it's impossible to do it in a straight line.”
- A coastline's measured length depends entirely on the length of the measuring stick (or map scale) used: shorter sticks capture more irregularities and yield ever-larger numbers that never converge to a single true value, unlike a smooth shape; therefore a coastline is 'as long as you want it to be,' though comparisons remain meaningful if methods and scale are agreed upon.
“The problem is that if the stick is very long, we are missing many of the irregularities... And, the shorter the stick is, the more we are capturing... So, for a normal, regular shape, it should converge to the actual number. There is one. Here, there's none.”
All map projections mathematically must distort either distances or angles because spheres cannot flatten without loss.
- Because the earth is a sphere and maps are flat, a mathematical theorem proves that projecting a sphere onto a flat surface must distort something—either distances or angles—and this cannot be avoided; recognizing this lets us make and read maps more carefully.
“when we try to take something like the earth--so a sphere--onto a flat piece of paper, there will be problems. Something will be not working, something will be distorted, be it distances or angles, and we cannot get around it.”
- The Mercator projection makes landmasses near the poles look far larger than they are (e.g. Greenland appearing similar in size to Africa) as a mathematical consequence of the projection; while critics call it racist for enlarging richer northern countries, the distortion stems from its design for navigation, and no projection is inherently better—each must be chosen for its purpose.
“the closer to the poles we get, the bigger the things look--the bigger the continents look. So, we have this massive Greenland, and then we have, like, tiny Africa.”
- Although Greenland is the largest island in the world, on the Mercator map it appears roughly the size of Africa while in reality it is around 14 times smaller than Africa.
“while it looks like the size of Africa on Mercator's map--I don't know the number. I wrote it in the book, I think it's like 14 times smaller.”
Combining mathematical optimization with human local knowledge and judgment produces better real-world outcomes than either alone.
- When UPS attempted to fully automate route optimization it was a costly failure because the algorithms ignored local knowledge (like avoiding driving past a school at 8-9 AM); combining drivers' human experience and area knowledge with mathematical optimization produces the best results.
“UPS had... they decided: We are going to make everything automatic... they paid a lot of money... And, this was a big failure. It just didn't work.”
- The number of possible routes for a delivery driver grows factorially with stops—five stops yield 120 possible routes, but 135 stops (a typical UPS day) produce more possible routes than there are stars in the universe—so even the most powerful computer cannot compute the optimal route each morning; companies instead use secret algorithms that produce a 'good enough' route quickly, perhaps 10% worse than optimal.
“with five stops... I have 120 ways to do that. But then, we grow very, very quickly to numbers that we cannot even imagine.”
- AI will not kill mathematics as a field but will change it the way calculators did—handling mechanical, semi-creative proof steps—while humans remain essential for a long time to figure out which steps are needed, how they fit together, which problems are relevant, and what questions to ask; AI is best understood as a very advanced calculator letting mathematicians spend time on thinking rather than computing.
“I don't think mathematics will die out as a field, but it's going to change. And it's changing already. So, kind of in the same way that calculators changed the field.”
- Self-driving cars are becoming safer than human drivers in part because they are more risk averse—stopping or slowing when unsure—and because they don't get angry, tired, or emotional, whereas humans often proceed despite uncertainty because they are rushing.
“They don't have feelings. They don't have emotions, they don't get angry, they don't get tired.”
- Maps combined with mathematics that looks like 'common sense' but is actually deep math—such as geographic profiling—can locate the source of an epidemic or narrow the search for serial criminals, demonstrating that maps and mathematics together can be life-saving tools.
“we can find things, or people, or the source of an epidemic by kind of using something that looks like common sense, but it's actually deep mathematics.”
- Germany's earlier system of around 20 different trash bins confused people so much that they would dump everything into mixed waste that wouldn't be recycled; reducing the number of bins increases actual recycling efficiency, illustrating that serving humans matters more than maximizing theoretical sorting precision.
“if you are so confused, you're just going to put everything into mixed recycling or mixed bag, which won't get recycled in the end.”
- Mercator originally made globes by printing wedge-shaped pieces and pasting them onto a sphere, and confronting this opposite problem—translating flat shapes onto a curved surface—likely inspired his thinking about projecting the curved earth onto a flat map.
“he started from the opposite problem. He was making globes... they were making these wedges, printing... drawing these wedges and pasting them onto a globe.”
Topological and navigational maps prioritize connection and route logic over geographic accuracy because users need different information than spatial realism.
- Henry Beck reinvented the London Tube map by borrowing the logic of electrical circuit diagrams—discarding geography and showing only stations and their connections with straight horizontal, vertical, and diagonal lines and roughly equal spacing—because riders only need to know which line to take and where to change and get off, making a topological (connection-based) map far more useful than a geographically accurate one.
“Henry Beck... was, like, 'What if we did with the Tube map something that engineers do with electrical circuits?' So, a map that forgets about the geography but only focuses on the stations and the connections between them.”
- Because topological subway maps space stations roughly equally regardless of real distance, riders mistakenly assume adjacent-looking stations are equally close, when in reality some are a three-minute walk apart and others a 40-minute walk apart.
“Some tube stations are just, like, three-minute walk away, and some are, like, 40-minute walk away. But they look the same.”
- Self-driving cars need far more accurate maps than human navigation aids—down to where the curb is—because pre-knowing the road frees the car's processing to handle the unexpected (like a child running out); this is why current self-driving cars mostly operate in specific well-mapped cities, though improving AI may eventually let them handle novel places like human drivers.
“it's very helpful if the machine at least knows the road very well and also knows what to expect so it can focus on the unexpected.”