We could be inventing new wheels with weird new physics

20.22, Wednesday 2 Mar 2022

There’s that old line that bumblebees shouldn’t be able to fly. Well, no they can’t if you model them as fixed-wing aircraft, and even flapping like a bird (which we’ve all seen) doesn’t seem like it would be enough to keep the bee aloft. The line resonates because the flight of a bumblebee is counterintuitive somehow.

It turns out that bees fly by flapping those wings (a) really, really fast, and (b) in a figure-of-eight wing motion to create low-pressure vortices to pull them up (source).

I don’t need to take air vortices into account on my morning commute, and nor does anyone else (excepting cyclists and pilots). So that’s the source of the intuition circuit break: the bee’s mechanism operates at a scale (time, size) that we don’t have access to, so we can’t access it in the imagination either.


ANYWAY. DeepMind has created an AI to wrangle plasma flux in experimental fusion reactions.

Fusion is the energy source of the future – but it requires holding in place a super hot, high energy plasma (a kind of electronic gas) inside a donut-shaped container called a tokamak. At sufficiently high temperatures, fuel can be injected such that fusion occurs, generating energy and also igniting the plasma (allowing for more fusion). The reaction becomes self-sustaining and that is the holy grail. But… plasma cools too quickly if it’s not held in place away from the walls of the tokamak. So rapidly adapting magnetic fields are used to cage it – which is hard. Plasma in the tokamak is wild and powerful; strange eddies, loops and currents braid it, making it writhe and snap with huge force and speed, an unwilling captured sun. Our magnets can’t react quick enough to ride it for long.

AND SO:

DeepMind researchers have trained a reinforcement learning agent to shape the distribution of plasma in a Tokamak fusion reactor. This requires training an agent that “can manipulate the magnetic field through a precise control of several coils that are magnetically coupled to the plasma to achieve the desired plasma current, position, and shape”. If that sounds complicated, that’s because it’s extremely complicated. The task is akin to being an octopus and needing to precisely shape a tube of clay that’s rotating at speeds faster than you can comprehend, and to never tear or destabilize the clay.

(Grateful hat tip to Matt Jones who posted this on Petafloptimism.)

They’re not just controlling the existing plasma but using their AI to explore new plasma configurations – and I am intrigued about what novel bumblebee physics they might find. Unexpected figure-of-eight plasma orbits that exploit unpredicted vortex physics to make fusion radically more efficient, perhaps. Who knows.


How fish swim!

Hey get this (there’s a connection I promise):

Fish swim by using the surface physics of their skin to create and exploit low pressure micro vortices in the water.

fish somehow exploit the vortices to reduce the amount of energy they need to combat the momentum of the flowing water.

And, in particular:

It appears that if there are no useful vortices already formed in the water (e.g. by rocks etc) – the trout can actually make some itself - and then extract energy from them.

Whoa.

The skin is vital: The scales create a ‘one way surface’.

Vortices in the water are generated by the skin, and the side-to-side movement of a trout is the fish slipping between the vortices, pinballing between them, propelled on them like a boat on wind. (Shown, says the article, by the fact a dead trout on a line in moving water will still exhibit the characteristic swimming action.)

All of which leads to this REMARKABLE line:

Fish don’t swim, they’re swum.

ARGH. Too good. Am dead now.


LOOK: wheels are great because they make movement easier – but it turns out there are other mechanisms (surface physics, rapidly evolving vortices, one-way skin) which similarly lower the energy for motion.

But they are hard for us humans to imagine. And hard to discover! And hard to do! These new kinds of wheels operate at scales which are outside the human everyday. We don’t derive them as simple solutions to equations. We stumble around in the dark and find them in the corners.

Until now.

It strikes me that what DeepMind’s AI is doing, in a general sense, is operating with enough attention to detail and enough complexity and enough resolution at a high enough speed to discover and exploit novel physics. That’s what explore new plasma configurations means.

So… we could put the DeepMind plasma wrangling AI to work in inventing new wheels? New efficient forms of motion and propulsion?


I was sorta kinda getting at this when I talked about faster-than-real-time computer simulation back in 2020:

So, with powerful simulation, could you figure out how to hit a mass of water with puffs of air so that it rises up and moves around the room, washing the windows; or robots with reed-thin jointed limbs that should never be able to hold themselves up, but with motors at each joint running at just the right vibration to keep the thing moving?

And WHY NOT, right?

It’s just high-hertz plasma wrangling, right? It’s just vortex pinball.

The question is not “can it be done” (yes it can be done, thanks DeepMind) but will it be done?


How can the tools for inventing new wheels end up in the hands of the people with the right imaginations?

What do we need?

  • Incredibly high-resolution real-world physics simulations, enough to accurately model vortices in turbulent air
  • 3D software that can model, at the least, the motion and surface effects of bumblebees wings could be replicates
  • A DeepMind AI that, given a goal and a starting point, could run a giant search to optimise the physics and controls of… whatever new objects the designer speculates

All wired together. Handed out to designers and mechanical engineering students.

And, given this package, perhaps the future will look very different from our science fiction.

Pinhead drones dragging copper wires behind them, darting through the home bouncing on air currents, generating electricity and power by dragging their tails through ambient magnetic fields.

Directional packaging that is can’t slip out of your hands (but dislodges easily when you move your hands the other way).

Cars with fine filament-bristles covering on the base, shaping and sweeping the air at nanometer resolution to ride on a silent and almost friction-free air cushion of vortex turbulence.

All mechanical objects with halos of filaments, magnets, mist, so fine that the eye can’t identify clean edges, no hard plastics or iron but all our artefacts in soft focus, encased as they will be in a gentle haze of turbulent air sculpted by alien intelligence.

More posts tagged:
Follow-up posts:

If you enjoyed this post, please consider sharing it by email or on social media. Here’s the link. Thanks, —Matt.