Startup Successfully Uses AI to Find New Geothermal Energy Reservoirs (cnn.com) 49
A Utah-based startup announced last week it used AI to locate a 250-degree Fahrenheit geothermal reservoir, reports CNN. It'll start producing electricity in three to five years, the company estimates — and at least one geologist believes AI could be an exciting "gamechanger" for the geothermal industry.
[Startup Zanskar Geothermal & Minerals] named it "Big Blind," because this kind of site — which has no visual indication of its existence, no hot springs or geysers above ground, and no history of geothermal exploration — is known as a "blind" system. It's the first industry-discovered blind site in more than three decades, said Carl Hoiland, co-founder and CEO of Zanskar. "The idea that geothermal is tapped out has been the narrative for decades," but that's far from the case, he told CNN. He believes there are many more hidden sites across the Western U.S.
Geothermal energy is a potential gamechanger. It offers the tantalizing prospect of a huge source of clean energy to meet burgeoning demand. It's near limitless, produces scarcely any climate pollution, and is constantly available, unlike wind and solar, which are cheap but rely on the sun shining and the wind blowing. The problem, however, has been how to find and scale it. It requires a specific geology: underground reservoirs of hot water or steam, along with porous rocks that allow the water to move through them, heat up, and be brought to the surface where it can power turbines... The AI models Zanskar uses are fed information on where blind systems already exist. This data is plentiful as, over the last century and more, humans have accidentally stumbled on many around the world while drilling for other resources such as oil and gas.
The models then scour huge amounts of data — everything from rock composition to magnetic fields — to find patterns that point to the existence of geothermal reserves. AI models have "gotten really good over the last 10 years at being able to pull those types of signals out of noise," Hoiland said...
Zanskar's discovery "is very significant," said James Faulds, a professor of geosciences at Nevada Bureau of Mines and Geology.... Estimates suggest over three-quarters of US geothermal resources are blind, Faulds told CNN. "Refining methods to find such systems has the potential to unleash many tens and perhaps hundreds of gigawatts in the western US alone," he said... Big Blind is the company's first blind site discovery, but it's the third site it has drilled and hit commercial resources. "We expect dozens, to eventually hundreds, of new sites to be coming to market," Hoiland said.... Hoiland says Zanskar's work shows conventional geothermal still has huge untapped potential.
Thanks to long-time Slashdot reader schwit1 for sharing the article.
Geothermal energy is a potential gamechanger. It offers the tantalizing prospect of a huge source of clean energy to meet burgeoning demand. It's near limitless, produces scarcely any climate pollution, and is constantly available, unlike wind and solar, which are cheap but rely on the sun shining and the wind blowing. The problem, however, has been how to find and scale it. It requires a specific geology: underground reservoirs of hot water or steam, along with porous rocks that allow the water to move through them, heat up, and be brought to the surface where it can power turbines... The AI models Zanskar uses are fed information on where blind systems already exist. This data is plentiful as, over the last century and more, humans have accidentally stumbled on many around the world while drilling for other resources such as oil and gas.
The models then scour huge amounts of data — everything from rock composition to magnetic fields — to find patterns that point to the existence of geothermal reserves. AI models have "gotten really good over the last 10 years at being able to pull those types of signals out of noise," Hoiland said...
Zanskar's discovery "is very significant," said James Faulds, a professor of geosciences at Nevada Bureau of Mines and Geology.... Estimates suggest over three-quarters of US geothermal resources are blind, Faulds told CNN. "Refining methods to find such systems has the potential to unleash many tens and perhaps hundreds of gigawatts in the western US alone," he said... Big Blind is the company's first blind site discovery, but it's the third site it has drilled and hit commercial resources. "We expect dozens, to eventually hundreds, of new sites to be coming to market," Hoiland said.... Hoiland says Zanskar's work shows conventional geothermal still has huge untapped potential.
Thanks to long-time Slashdot reader schwit1 for sharing the article.
Fusion (Score:2)
I suspect that if AI lives up to expectations, then the most significant benefit would be using AI to design a practical method of generating power from nuclear fusion. A success in that field could address a number of other world problems.
Re: Fusion (Score:2, Interesting)
Re: (Score:2)
Re: Fusion (Score:2)
Re: (Score:2)
That's not creative, it just fills in blanks with what it expects the data to look like based on extensive training, wherein a lot of data is available. GANs and discriminators are good at this. They're not good at coming up with new methods of doing things. If you ask an LLM to please unify electromagnetism, gravity, the two nuclear forces and quantum mechanics into a cohesive theory, you'll get slop. Same if you want it to make a better mousetrap.
The act of creation, whether it be art or scientific theory, is surely largely based on prior work. I'd say is sometimes entirely based on prior work, rearranged in a novel way.
Re: Fusion (Score:2)
Re: (Score:3)
Does it solve a problem humans haven't solved? AI is good at regurgitating solutions it's been trained on. It's currently terrible at determining what the solution should be for novel or unsolved problems.
I agree that current AI does better with problems closer to data on which it was trained. But I would suggest that "a problem that humans haven't solved" covers a range of problems, from the rearranging of past solutions to problems that require a completely new way of thinking. An AI model may not have been trained on the best route from A to B, but might nevertheless be able to solve that problem because of its similarities to training data. As you say however, AI might not be able to come up with a signi
Re: (Score:2)
I suspect that if AI lives up to expectations, then the most significant benefit would be using AI to design a practical method of generating power from nuclear fusion. A success in that field could address a number of other world problems.
No doubt economical fusion power would be an enormous boon. It's a question of timing and development risk.
We know how to build geothermal power plants. As the fine summary says, they can be online in a matter of years. As the old joke goes, fusion is 20 years in the future and has been for decades. Even if an AI came up with a great stellarator design today, I'm not optimistic it would be functioning in less than ten years.
But all that said, it's not an either-or situation. We should pursue both.
Re: (Score:2)
Startup proposes thermal power plant, mentions AI in press release to attract VC
There, headline FTFY.
Finally (Score:2)
Re: Finally (Score:3)
The key here is that it isn't LLM, it's the same standard "AI", aka machine learning, that has been around and useful for decades. It's just all in the hype now.
Re: (Score:1)
You remind me of the redditors who crusade against AI art, who have been recently posting about how they're winning, because there's not as much AI art any more.
The level of cognitive blindness there is funny, just as it is here.
Re: (Score:2)
I don't follow. I'm not blind to the wildly impressive outputs of these programs. It's just that the sum of all human knowledge for all of civilization hasn't made a model that actually knows anything and doesn't make mistakes that a child could avoid. You are blind to that. You think that intelligence is some sort of "emergent property" that will just poof into existence if we keep making things bigger, but it's just wrong and there is no evidence that this is the case. It's science fiction.
The fact is tha
Re: (Score:2)
>doesn't make mistakes that a child could avoid
If you look at my post history on this subject, you'll find that not only am I aware of this issue, but I even explained in great detail the theory behind why it makes those mistakes, and what it will take to eliminate them.
Specifically, the problem is embodiment issue. Unlike luxury belief of very rich and decadent Westerners dictate, we are not a "ghost in the shell", the consciousness in a machine. We are an integrated whole, with consciousness being a su
Hype goes a long ways. (Score:1)
OpenAI and all their circle of 'investors' started this. Go look at Google Trends for AI, and see how the uptick and hype cycle started with the launch of ChatGPT on November 30, 2022. Was AI going to "change everything" when slashdot posted how machine learning was doing well at reading medical scans? Or was it going to change everything when Bill Hader's impression morphed into Tom Cruise on Letterman? Nope, AI hype rocketed up when OpenAI launched a chatbot. Somehow that makes Nvidia worth from a sub $10
Re: (Score:2)
ML is a subset of AI.
Have a look at how things are connected to each other: https://www.researchgate.net/f... [researchgate.net]
Re: (Score:2)
Semantics, then. You're also conflating "LLM" with "AI". So it's "ML but not LLM". "AI" is just an anthropomorphized bastardization of "very, very complex math equation". The tech bros have been calling everything with any kind of "algorithm" (again, a math equation) "AI" for at least the last ten if not fifteen years. Everything had AI in it. Now LLM is the new AI, and now the hype train is running on rocket fuel instead of coal. (Although, literally, it's running on coal, hah.)
Re: (Score:2)
Have a look at the graphic. Or read some basic literature. AI is literally the field where your "game AI" from the retro games fits in, just as (advanced) path planning, literally chess AI and much more. The fallacy is thinking AI would be the most specific term while it is the most broad one. Even ELIZA is AI. And you can implement it with a simple finite automaton.
Sadly LLM are not in that graphic, but the other graphics were not so detaild. But you would find it in the subset of DL for NLP, possibly form
Re: (Score:2)
Jesus Christ. You're so off topic from the original intent. I am disputing the term "Artificial Intelligence" as a complete misnomer of what it "actually is". I don't care about a Venn diagram or where "LLM" fits in it. Look at your drawing: the "AI" and "Machine Learning" sets are identical, so one of them is clearly pointless, or the diagram is just wrong. Not to mention it's just one diagram by one person -- who is not some "AI" luminary -- in the International Journal of Business Ecosystem and Strategy,
Re: (Score:2)
It is not a misnomer, you are just misunderstanding it. It was defined *decades* ago and didn't change since then.
Maybe this is short enough to get the point across.
And you can just read the scientific definition, or if you like more graphics put "AI venn diagram" into google and find many many more, each defining AI as the broadest category. And the diagram I linked shows Machine Learning as a real subset of AI, not as equivalent. Don't you know how to read Venn diagrams?
Re: (Score:2)
These uses have been massive for a while, as foundational principles are exactly the same as ones in LLM. It's just that where LLM addresses relational likelihood of letters, words, sentences and paragraphs, these AI's address relational likelihood of various geological observations.
Other useful applications you don't hear about if you don't follow it include everything from medical imaging and diagnosis to searching for new sources of oil, to discovering new antibiotics.
It all works on the same principle.
Re: (Score:2)
Re: (Score:2)
This doesn't make geothermal universally available or anything of the sort.
It just finds a handful of locations we missed. It's a specialist application. There are similar applications across countless fields. To make a connection to reality, it's why nvidia rtx 5090 are selling for way more than MSRP. They have enough VRAM to fit narrow models for training purposes. So every research institution is vacuuming them to run such models to do inference on their data searching for things we missed.
Large scale LL
Re: Finally (Score:1)
" AI models have "gotten really good over the last 10 years at being able to pull those types of signals out of noise," Hoiland said..."
How much does the attention mechanism (which is not a neural network, and which was a paradigm shift in chatbot proficiency at natural language) play a part in any pattern finding?
Re: Finally (Score:2)
Re: Finally (Score:1)
Is "Attention is all you need" weird?
"We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. "
"In this work we propose the Transformer, a model architecture eschewing recurrence and instead relying entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and can reach a new state of the art in translation quality"
"th
Re: Finally (Score:2)
The probl
Re: Finally (Score:1)
Can you use attention independently of neural networks?
Are you claiming the following response is a hallucination? Why?
"You can implement "statistical attention" using purely mathematical operations like dot-product similarity or cosine similarity without any neural layers."
Re: Finally (Score:2)
Re: (Score:2)
Attention means computing coefficients for the relation between different parts of tensors. You can use the result in the next layer of a neural network, but the general mechanism is more general
Re: Finally (Score:2)
I was surprised. A regular PID controller should do the job fine. So had a talk with the technical guy of their team. Yup. Plain old (ancient) tech. PID controller. He joked that the I in PID controller stood for intelligent. I replied that there was no I in sales. Good times!
Re: (Score:2)
Re: Finally (Score:2)
Re: (Score:2)
You say "finally" as if this hasn't been happening for years. AI is far more than LLMs and has been used for well over a decade. This only made the news now because everything AI makes the news (and this is a startup so bonus news hype).
The subsurface field (basically oil, gas, mineral, and energy exploration field that concerns itself with analysing any of the many things we do to measure what is below the ground) has been using AI for decades. I remember seeing a presentation discussing how they trained a
Re: (Score:2)
I'm not saying non-LLM AI is not happening but rather it is not reported. It is only AI news that is about how much money and resources is being consumed on the technology for stealing people's IP and jo
Re: Finally (Score:1)
Seeing as the attention mechanism gave LLMs context awareness that neural networks could not touch, do you think the geological application uses the attention mechanism, and is therefore using the same tool that gave context-sensitivity to LLMs?
Re: (Score:2)
Don't look at me, I'm an instrument engineer. I was just there for the presentation and don't know how it works under the hood. I can tell you in 2015 I was the end user of software that was a predictive model trained on 15 years of process and maintenance data from some 25 gas compressors that ultimately created a model that turned out to predict future compressor failure based on changes in the operating envelope (upsets, pressure spikes etc, that seem insignificant at the time), and was able to do this p
Now this ... (Score:2)
Picky, picky (Score:2)
A) There were a bunch of mistaken selections (with undisclosed costs) before hitting on the "right" one
- or -
B) It worked on the first try, but "Big Blind" is actually a Hellmouth.
Wow... (Score:2)
Wow...an expert system...(sorta) who'd have thunk it?
Well actually in 1982...
https://www.science.org/doi/10... [science.org]
JoshK.
Re: Wow... (Score:1)
Do expert systems use explicit rules to try to solve context sensitivity while the attention mechanism just counts co-occurrences of tokens over very long ranges?
Re: Centralia, Pennsylvania has "geothermal" (Score:1)
"Prehistoric clinker outcrops in the American West are the result of prehistoric coal fires that left a residue that resists erosion better than the matrix, leaving buttes and mesa. It is estimated that Australia's Burning Mountain, the oldest known coal fire, has burned for 6,000 years.[10]"
AI finds the needle in the haystack (Score:4, Interesting)
So it looks like the editor sprinkled the magic word “AI” on the headline and slashdot did what it always does: auto-spawned 200 trollish variations of “lol AI hype.” Cute. Meanwhile, the actual story here is a lot more practical than the kneejerk anti-AI trolls realize.
What Zanskar is using machine-learning models to spot blind hydrothermal systems: reservoirs with no obvious surface tells. No hot springs, no geysers, no “hey look, free steam!” signpost. In other words: no leaking clues. Humans have traditionally hunted where the geology is loud. Zanskar is trying to hear the quiet stuff, then drill to confirm.
And once you confirm it, the extraction is basically the normal geothermal playbook: Find and a likely spot->drill production well(s)->bring up hot fluid->strip heat at the surface->reinject the cooled fluid back underground.
So the key tradeoff vs traditiona” geothermal isn’t new extraction vs old extraction. AI reduces exploration risk (fewer dry holes), but you still face the classic geothermal buildout grind: drilling cost, reservoir management, cooling choice, permitting, interconnection queues, etc. Zanskar’s bet isn’t new thermodynamics. It’s “we can de-risk the needle-in-a-haystack exploration phase."
Now here's the part where Arizona, Nevada, and Utah start side-eyeing the whole thing. If this hydrothermal renaissance turns into “power for data centers, paid for by sucking from the last puddle,” that’s not clean energy. That’s just a different flavor of externalized cost, and those of us living in the waterless paradise that is the desert Southwest get to pick up the tab. The good news: geothermal doesn’t have to be a water vampire. Many systems reinject what they produce. The risk knobs are mostly water recovery -- especially surface evaparoation in wet-cool loops, which are the most likely (read: highest profit margin) system designs,
My hope, and my ask, for anyone deploying this in the West is to be smart about it. Treat potable groundwater as off-limits unless there’s no alternative and it’s transparently justified. Prioritize closed-loop/reinjection-heavy designs and aggressive leak accounting. Use non-potable sources for any makeup water (brackish, treated, industrial) whenever possible, and pick cooling systems with the desert reality in mind, not the pie-in-the-sky brochure pitch to investors. Last but not least -- monitor and disclose environmental impacts like you actually have to live here afterward.
Using AI to find the needle in the hydrothermal haystack? Absolutely. That’s a sensible tool applied to a hard search problem. Just don’t let “AI found clean energy” become the preface to “and then we sucked your groundwater reserves dry to power yet another chip fab/AI datacenter in Phoenix.