– So this lights gonna go on and that’s gonna light up. – The anticipation. – There you go. Triggering. – I pushed a button. – [Intercom] Triggering. (beeping) (loud metal popping) – So what is that sound we just heard? That’s the sound of $150
million plasma collider, blasting jets of super hot ionized gas at hundreds of miles per second. We’re at TAE Technologies,
a 20 year old start up, that hopes to finally
achieve nuclear fusion. It’s the same energy that fuels the stars but harnessing it here on earth has been a scientific dream for decades. Fusion power plants would
mean plenty of energy, no carbon admissions, little
long term radio active mess and zero chance of a
Fukushima-like meltdown. Governments, universities
and private companies have been pursuing fusion since the 1940s. We’re still nowhere near
adding a fusion power plant to the grid, but in February 2018, a panel
of experts published a list of game changers for fusion research and at the top of their list
were advanced algorithms like artificial intelligence
and machine learning. One of the private companies
chasing fusion is TAE, which began collaborating
with Google a few years ago. Together they developed
machine learning tools that TAE thinks could finally
bring fusion within reach. So we headed to a nondescript stretch of Foothill Ranch, California. We wanted to see just how
far along TAE Technologies is and where, if anywhere,
AI enters the picture. – So right now we’re looking
out towards the main vessel, the experiment. See all the blue cabinets around, they are our power supplies. Primarily for the magnets which help to control and shape the plasma that we are heating. – [Rachel] Our guide is Erik Trask, a lead scientist at TAE Technologies. He’s taking us on a tour of Norman, the company’s plasma collider. Inside magnetic fields propel and control the super hot spinning clouds
of ionized gas called plasma. While particle beams heat
it to temperatures hotter than the sun’s core. – The goal of this machine is to show that we can heat up our plasma to high enough temperatures to show that it is relevant to make a fusion power plant out of. – [Rachel] All attempts at fusion involve smacking lightweight particles into one another at
temperatures hot enough for the particles to fuse. That produces a new element plus energy. In the starts, fusion occurs
when hydrogen nuclei fuse to form helium. There are a few different
ways to try and replicate this process on earth. Some experiments control the
plasma with magnetic fields inside a massive metal
donut called a tokamak. TAE twirls the plasma inside
a linear machine named Norman. Sounds great but none of this
has produced enough energy to be useful for a power plant. (loud metal popping) A machine in the UK called
Jet did set the world record by generating 16 megawatts of fusion power but it took 24 megawatts
to heat up the plasma and even more electricity to contain it. So the whole thing was
a net loss of energy. TAE’s Norman machine produces
some fusion reactions but not many according to
the CEO Michl Binderbauer. – So things get hot enough
that yes occasionally a few elements actually
collide hard enough and you get fusion. And we actually use this
for diagnostic purposes. Now this isn’t bulk fusion yet, right. So this wouldn’t be economic. This makes a handful of events. But as we go up the scale
eventually we’ll get there. – [Rachel] That’s the goal for the future, a fusion reactor that
runs for months at a go. For now, the goal is to produce a plasma that’s 30 million degrees
Celsius for 30 milliseconds. To get there the team fires
Norman 50 times in a day for just milliseconds at a time. Each shot they may tweak
thousands of variables like the timing, the shape
of the magnetic field, the voltage of different power supplies or the amount of plasma. Since the machine came online, it’s fired over 11,000 shots, each one a bit of a blind
stab at the optimal variables. – And we have a fault on N1N. Do we know what happened there? Stopping shot. – So what happened? – So there’s potentially
a resistor that’s gone bad or a board that is
reaching an over voltage and it’s tripping something inadvertently. – [Rachel] It’s impossible for a person to keep all those variable in their head or to change them one at a time. That’s why the team is
collaborating with Google using a system called the
Optometrist Algorithm. – The advantage of one variable at a time is you know exactly what happened and why. Which makes prediction and
model building very easy. You can do that as long as the
number of variable you have is not very large. How do you do that in four dimensions? Or 20 dimensions? – [Rachel] The algorithm is
what lets them change scores of variables in each shot. If the team thinks the
new combination produced better results than the
last one they pick one path of pre-programmed shots. If it’s worse, they pick another, hence the name optometrist. Is shot one or shot two
closer to viable fusion? Three or four? 654 or 655? According to Erik, what’s happening here is really machine learning rather than AI. Right now, it’s still the
humans making the call about what’s better and what’s worse. – We’ve gone through countless classes in undergrad and graduate school, time in the lab, so our
knowledge is being used to produce data. At some point, an AI system could be used, could be trained on the
knowledge we’re gathering. We’re not quite at that stage yet. – [Rachel] We asked some
other fusion researchers what they thought of TAE’s
approach, Julian Kates-Harbeck, who worked with Princeton’s
physicals laboratory, and Christina Rea at MIT. They use machine learning
a little differently but they’re intrigued by
TAE and the optometrist. – This algorithm supports
human decision making but doesn’t take away the
human out of the loop. So it’s sort of this beautiful synergy between the human and the machine. – [Rachel] Julian and Christina have each developed algorithms to predict when the plasma might misbehave in order to keep it under control. The methods differ but Christina
says fusion experiments are just so data heavy that it’s natural for machine learning to help here and there within reason. – There are many low
hanging fruits in our field that can really benefit from
using this advanced statistics that lies under the hood of AI in general. Of course, we are not getting to the point where we are building
a self driving tokamak. – [Rachel] These teams
all see machine learning and AI the same way, as important tools just not as a panacea to all of fusion’s problems. Right now there is no such thing. The TE team is still
thinking in milliseconds and looking for one good day at a time. – If we can get there in one day, great. It probably will take more than one day. – [Rachel] Is that a good result, the plasma 20 million
degrees for 10 milliseconds? – That would be terrific, yeah. Champaign would be flowing by now. – Hey everyone, this
video is actually part of the Real World AI, a special
issue here at The Verge. This week we’re bringing
you all sorts of stories about how AI’s being
used in the world today. So if you want more go to
The Verge’s YouTube channel or go to theverge.com/AI.

100 thoughts on “Can AI help crack the code of fusion power?”

  1. Huh, Foothill Ranch! A sleepy little town just up the road from me. Definitely not a place I would have predicted daily nuclear fusion experiments to be going on.

  2. Is Net positive Fusion achievable? I assume so cause of the stars, but I can't understand what exactly is the problem of maintaining the process

  3. I hope I will be proven wrong by science but I don't think a feasible fusion power will be achieved within my lifetime… Let's hope this will be another one of the "proven wrong" quotations.

  4. "No device can generate energy in excess of the total energy put into constructing it. Energy always and only comes flowing from the past into the future."(The Fifth Law).

  5. I believe man should not have access to this power, if it malfunctioned life as you know it for 46% of world population would be gone, also there’s a completely different type of radiation that comes from this, that we don’t fully understand yet

  6. AI? Why are buzzwords used randomly only because they are cool? Maybe to attract money of ignorant investors.

  7. polywell team is already doing iterative design in simulations to nail down the final design parameters for their demo equivalent device.

  8. Who are these "So" people. What is different about these "So" people.
    Why do they respond with a sentence that always begins with the word, "So" ???
    If you are against discrimination of any kind, then you are in for a big shock.

  9. Fusion is a diversion from real free energy technologies. It’s also expired tech. 1951 Inventions Secrecy 5k more exotic tech suppressed/secretised.

  10. There is only one problem, here on Earth you have the effects of Gravity, make a little test model for the ISS and I bet it'll work much better, all they have to do is Develop, Artificial Gravity or an AG Feild capable of simulating it.

  11. You mean to tell me a 20yr old "start up" has a $150mil piece of equipment, has failed to ever produce a viable product, and is optimistic about it's future? I chose the wrong field of work.

  12. The problem is the process has too many variables to control, it needs to be extremely simplified. Simplifying the model is the key.
    The gravity of the sun is the simplification key for the sun, on a small scale, you have to find it's equivalent.

  13. Just as long as they are using machine learning techniques and AI to help find the best possible solution as opposed to allowing the AI to manage security procedures, and anything else that might control safety features and physical systems that run the fusion process – otherwise, well just use your imagination…

  14. Fusion Engines were successfully made and worked years ago , so why this crackpot vid about nothing new ?!! See; David Adair at Area 51 for this old ' news " ,     -George

  15. The amount of data needed to train an AI to do that via machine learning would be huge. How fast do they crank out those answers? Can they realistically produce that much data?

  16. Twenty years from now you will believe you've made progress. If you then look closer, you will realize nothing has changed. Your paradigm is over 100yrs wrong. But, its at least a paycheck.

  17. 20ms is not a success either… It's a fluke… Scientists get stuck on one idea way too easily… Obviously it needs a new approach if its not progressing anywhere

  18. On an day the AI will solve our problems in just a matter of seconds, am convinced that ai can ansure us lots of questions to solve problems.

  19. you will always need more energy in because unless there is a chain reaction like fission how can it run itself? fusion does not work like fission. I'm sure if it was possible it would have been conquered years ago. when they make the first perpetual motion machine then they will conquer the fusion problem….Never

  20. nopes, its another can of worms just for a couple of people to get paid.. nuclear power is the cleanest and best power ever.. if you know what your doing..

  21. Nuclear fusion as you called it throws you off in your approach? It's really cold fusion as our sun is driven? Imagine the elements in our periodic table that becomes harmonically hot when put under certain pressure(s) as you know they exist in our Sun? It's been proven and possible as proven? I hope I've given you another approach rather that AI as humans have done this and know it's possible. You can do it without AI. So it has been said so I wish it will be done.

  22. One answer to the fermi hypothesis is that there is a perfectly reasonable field in science all intelligent civilizations will research in that is 100% likely to result in total extinction. This is a possible reason why we see no evidence of intelligence out there. Is it just me or do you also get a nervous feeling when you see people bringing the power of the sun to earth? There is no way of telling which field of research (if it exists) will bring extinction. It’s probably one we would least expect. I guess as nothing can be done it’s pointless worrying about it. Just a thought tho

  23. It will always take more power to run then it gives back, they're trying to create perpetual motion and it won't happen. Just put some solar panels up and a Tesla storage battery= done.

  24. Hopefully, I'll be working on advanced AI systems after I graduate. I would love to help physicists on such a project.

  25. Now that it is proven that Google would like to plunge the US into socialism I am a weeeeee bit worried about Google meddling in fusion.

  26. Isn't it impossible to get more energy out than put in ? i thought the only way it would be viable is if we can create a self sustaining fusion reaction like a mini sun .

  27. I cannot see how fusion can compete on economic, fuel, waste disposal or scalability parameters with 4th generation molten salt reactors. Maybe we are going in the wrong direction ?

  28. we already have a fusion reactor. perfectly safe, doesnt require fuel or emit hazardous waste materials for us to deal with afterwards. it is our sun. throwing billions for fusion reactor tech while improving solar cell efficiency would have almost immediate effect with far less investment is ignoring the climate change crisis the world is finding itself in.
    we have the solution now – it is time to stop playing the piper

  29. Is the inside of the machines void of any particles other then what is being used to create the fusion? Could the problem with fusion be that there are other particles or atoms that are present within the machine which are interfering with the collisions?

  30. I bet you can enter a VC office with a blank white paper with "AI" written on it and come out with a million dollar investment contract.

  31. So fusion tech is not hard to create it’s just organization that are making money off of other fuel supplies are holding us back.

  32. Fusion power can only be achieved in zero gravity. Consider the gravity of this planet in your calculations.
    For more answers to your questions please contact me.
    I am in this planet for a limited time.

  33. its not laid out even in theory yet

    we know lightning could power a small town, yet we cant store it

    same apply to fusion, we cant just drain the energy directly to batteries or to a powergrid
    heating water and spinning generators is too many steps, yet way less material invested like in a coal based energy plant


  35. Did someone try a volcano like big and creating a tornado like inside adjustable speed and hole above making the fusion to balance overheating just like volcano ?☢😇

  36. The missing component is and always will be the enormous gravity required to bring the two nuclei close together enough to fuse. Gravity is something that the sun and stars have in abundance, but not us tiny humans.

  37. It is easy to succumb to the notion that fusion energy is 30 years away and always will be after watching bit like this, and I imagine that will be the case until it isn't.

  38. A a lot of times in this video she calls it nuclear fusion it is not it is 2 atoms fusing to form a heavier atom

  39. What if the future of getting free energy are letting the world live in a world of microwave system so just like Tesla always wanted that is why he created a electricity that cannot harm a person? 👽🤓🧐

Leave a Reply

Your email address will not be published. Required fields are marked *