Deepmind worked on reinforcement learning for plasma control back in 2022 and this also led to a paper in nature. I don't really understand the differences between their earlier work and this paper but deepmind don't seem to be involved in this one: <a href="https://deepmind.google/discover/blog/accelerating-fusion-science-through-learned-plasma-control/" rel="nofollow">https://deepmind.google/discover/blog/accelerating-fusion-sc...</a>
I remember having this idea when I was studying machine learning in college. I'm <i>really</i> happy to see that it occurred to someone else in a position to actually look into it, because it "felt like something might be there" to me.<p>The basic idea I had was that fusion plasma containment involves containing a turbulent, dynamical system, so it might require some kind of actual intelligence learning or co-evolving with the system.<p>I wondered if this might be the only way to achieve over-unity fusion outside gravitational confinement (stars, black hole accretion disks, etc.). This would mean there are two fusion mechanisms in nature: gravitational confinement and cognitive confinement. The latter can only be a product of a living system.<p>When a living system achieves this, its biosphere "ignites" and becomes something I termed a "biostar." Biostars could be potential SETI targets -- biospheres that have harnessed fusion and so emit anomalous amounts of optical and infrared radiation on their night side. This moment of ignition would be an event in a biosphere comparable to the evolution of photosynthesis-- a fundamental change in the energetic dynamics of life.<p>In the far future life the that achieved fusion could settle things like rogue planets in deep space, so that would be another potential SETI target. Find objects emitting anomalous infrared in the interstellar void. The advantage would be being far from destructive events like solar storms.
Another team from Japan and the US published a paper on this recently too:<p><a href="https://www.nature.com/articles/s41598-023-49432-3" rel="nofollow">https://www.nature.com/articles/s41598-023-49432-3</a><p>Sabine has a good review of it:<p><a href="https://www.youtube.com/watch?v=4VD_DLPQJBU" rel="nofollow">https://www.youtube.com/watch?v=4VD_DLPQJBU</a>
I’m not sure I’m ready to trust an ml system that will control a fusion power plant. The potential of the the mistake making a ( bizarre, ml-like ) mistake seems very high to me.
Tokamak is dumb. I'm sick of hearing about Tokamak.<p>The plasma in the Safire reactor has self-containing magnetic fields and doesn't need the $20billion+ super-magnet infrastructure. A Safire reactor costs under $20-million to build, and probably much less these days.<p>The Safire reactor can keep the plasma lit and going for hours if not days without interruption. The Safire reactor has been around for over seven years now.
Tokamaks are 1960's technology. The future of economical fusion appears much more likely to be based on the field-reversed configuration (FRC). Helion expects to produce net positive energy production from a reactor designed primarily for He3 production in the next few years: <a href="https://www.helionenergy.com/" rel="nofollow">https://www.helionenergy.com/</a>