The world is too loud. Read what matters.

The Skeptics' Guide to the Universe

Change One Law of Physics and Every Wish Backfires

A live thought experiment from Dragon Con: you may change exactly one law of physics to make science fiction real, and every wish gets taken apart for the consequences nobody counted — antigravity means free energy, hyperspace travel means handing the technology to every alien civilization.

Physics lawsSpace telescopesRobot learningScience fictionThought experiments
The first half is live-room riffing; the second half — the Roman telescope, robot babies, the falling-atom experiment on the Chinese space station and the Verne details — is denser and worth skipping ahead for.

The argument · tap a timestamp to hear it

2:07

Changing a law of physics changes it for the whole universe

The rule is hard: pick exactly one, and it must be practical and feasible within the laws of physics — no perpetual-motion-style having it both ways. Steven starts by handing most science-fiction premises a death sentence — warp drive, shields, phasers, artificial gravity, antigravity — either they violate physics or they are non-trivial: theoretically maybe possible, but never going to happen. The real fun of the whole discussion isn't in what gets picked, it's in finding the consequences inside each wish that nobody counted.

— Steven Novella
8:08

Antigravity is free energy thrown in for free

Jay wants antigravity, on the grounds that transport, construction and materials handling would all change. Bob points out it's a twofer: if you can feasibly turn gravity off, then turn it on while the water wheel goes down and off while you haul it back up, and you've got a perpetual motion machine. Steven, as buzzkill, patches it by force: the energy consumed switching gravity off must equal the energy you get out of it, no free lunch. The argument also incidentally kills off science-fiction movies — single-stage-to-orbit is simply impossible in a 1g gravity well, and every shot of a ship reaching orbit in a minute is lying, because it assumes antigravity exists.

— Jay Novella
10:08

Hyperspace travel hands the technology to every alien civilization

Bob picks entering subspace/hyperspace: you travel under ordinary engines over there and come out having crossed a far greater distance, essentially practical FTL. He argues there's no downside risk, because hyperspace is separate from real space and doesn't affect our universe. Steven and Jay immediately push back: changing a law of physics changes it for the whole universe, not just for you, and hostile civilizations could use it to colonize Earth just as well. Bob's answer is to hide — human radio signals dilute into noise within a few light years, so in principle nobody can work backward from them to the fact that Earth has life. Jay offers a compromise, jump gates: both ends have to be built, so expansion can only be slow, and not every alien comes knocking.

— Bob Novella
23:16

Zero-point energy is too thin, and forcing it triggers a big rip

Steven's second choice is free energy, specifically zero-point energy. Bob gives the objection: zero-point energy is observationally extremely dilute, and lighting a single bulb might require several cubic light years of space; and if you really injected that much energy into space, cosmic expansion would run away into a big rip that destroys all life. Steven's response is that this is exactly why these aren't laws of physics — every wish, once it actually lands, brings catastrophic consequences, so the fun of the thought experiment is finding the loophole that dismantles the wish. Julia adds her astrophysics-major perspective: if gravity had a dial, 90% of the observations that infer stellar phenomena from gravity would all be wrecked.

— Bob Novella
36:19

A bucket of water can compute like a neural network

The core of physical reservoir computing: any nonlinear system — a bucket of swirling water, time-varying mayonnaise — as long as you measure the right spot, its nonlinear interactions can be mapped onto an artificial neural network and made to "think" for you. Someone gave a talk saying, "we got mayonnaise to think very slowly." Once that analogy holds, scale becomes a variable: if there were a nonlinear system the size of a nebula, you feed it input and measure output at the right point, and it does your computation for you. The speaker himself admits this is a "wild idea," but stresses that shrinking physical constants down to something tiny throws up a pile of strange corollaries.

44:29

Roman is a thousand times faster than Hubble, but cast as the supporting act

The Nancy Grace Roman Space Telescope is a wide-field infrared telescope with a 3.4-meter primary mirror, the same size as Hubble, but its design goal isn't "better than existing telescopes" — it's complementary: James Webb excels at deep space, Hubble at staring at specific objects, Roman at taking in a huge swath of sky at once. It can survey a thousand times faster than Hubble, and over its five-year mission it will sweep an area of sky equivalent to a thousand years of Hubble. Once it finds something interesting, Webb and Hubble can be turned toward it for a closer look. It will also build a three-dimensional map of the universe and carries a coronagraph to block starlight glare and directly image exoplanets — it can find smaller planets farther from their host stars, offsetting the transit method's and radial velocity method's bias toward "hot Jupiters."

55:41

A robot baby kicks its legs and kicks up explore versus exploit

The University of Hamburg and Czech Technical University built a "robot in a crib": they put the infant robot iCub in a crib, hung a mobile with QR codes above it, tied one leg to it with a string, and ran their own deep learning algorithm to predict the mobile's motion. The result was that the algorithm spontaneously produced the explore versus exploit trade-off — being curious and exploring, but also exploiting the causality already learned. The researchers treat it as a platform: you can't change a baby's learning algorithm, but you can swap algorithms in a robot, so you can test different versions of contingent learning theory. By contrast, the mainstream of current robot control is deep reinforcement learning: put thousands of copies of the robot in simulation, train with a reward function, then transfer to the real machine in one shot — that's "do this and something good happens," not "this causes that."

1:04:44

China's space station redid the Leaning Tower experiment with rubidium atoms

On China's space station, researchers redid Galileo's falling-body experiment with two clouds of rubidium atoms of different isotopes, one heavier than the other. The free-fall time was about 10 to 12 seconds, far longer than in ground experiments, and averaging over many repeats, the two clouds fell at rates agreeing to five parts in ten million. What this verifies is that gravitational mass and inertial mass are exactly identical — in Newtonian mechanics mass shows up in two ways, and during a fall the two effects cancel precisely. What Einstein asked was precisely: why must they be the same? There is no reason they must be. From that comes his thought experiment: in free fall gravity seems to vanish and you are in an inertial reference frame; and in an accelerating elevator light bends, so if gravity and acceleration are equivalent, gravity should bend light too.

In their own words · checked verbatim

Well, Jay gets a good twofer on this one because what you got the ability to turn off gravity in a feasible way, you got like free energy, too

Steven Novella8:08

The best we could hope for is a fraction of the speed of light. 60 70 80% potentially which would be nice.

Bob Novella10:08

There's no way I would you would want to live in a universe where you could travel to the past and actually change your present.

Steven Novella16:10

the end result would be universe expansion so nasty that it could cause the big rip, which we know is scare scary and it just basically destroy all of life

Bob Novella24:16

so in a weird way you can get a bucket of water to think compute

that anything could include consciousness if it's big enough and might maybe impossible to train with current technology but those things could happen and it could theoretically be mayonnaise

matter and energy tell spacetime how to curve. Curved spaceime tells matter and energy how to move

AI doesn't have to become conscious to change civilization. It only has to become competent.

Figures

Quantum tunneling waiting timeseptillion to the septillion years31:17
Space needed for zero-point energy to light a bulbseveral cubic light years23:16
Share of stellar observations inferred from gravity90%25:16
Years the Roman telescope's fuel can hold its position10 years50:35
Precision of agreement between rubidium clouds' fall ratesfive parts in ten million1:04:44
Year a crewed capsule is launched in a Verne novel18651:16:52
Year Mark Twain wrote the teleoscope short story18981:18:54
Year Edward Bellamy's novel Looking Backward was published18881:19:54

Glossary

physical reservoir computing
Treating any nonlinear physical system as a neural network and reading off its computational power.
island of stability
A predicted region in nuclear physics where superheavy elements might be stable.
coronagraph
An instrument that blocks a star's glare to directly image the planets around it.
contingent learning
The way infants learn by kicking a leg to establish a causal link with a moving object.
big rip
The hypothesis that dark energy makes cosmic expansion run away and eventually tears apart all structure.

How to listen

Who it's for

For listeners who like science-fiction premise-building plus space telescopes and robot learning progress; founders can skip the first-half riffing and go straight to the technical details in the back half.

Skip

The change-a-law-of-physics riffing from 2:07–23:16, unless you only want the jokes.