From the paper's method section, a bit more about which type of ML algo was used:
An RF machine-learning model was developed to predict lithium concentrations in Smackover Formation brines throughout southern Arkansas. The model was developed by (i) assigning explanatory variables to brine samples collected at wells, (ii) tuning the RF model to make predictions at wells and assess model performance, (iii) mapping spatially continuous predictions of lithium concentrations across the Reynolds oolite unit of the Smackover Formation in southern Arkansas, and (iv) inspecting the model for explanatory variable importance and influence. Initial model tuning used the tidymodels framework (52) in R (53) to test XGBoost, K-nearest neighbors, and RF algorithms; RF models consistently had higher accuracy and lower bias, so they were used to train the final model and predict lithium.
Explanatory variables used to tune the RF model included geologic, geochemical, and temperature information for Jurassic and Cretaceous units. The geologic framework of the model domain is expected to influence brine chemistry both spatially and with depth. Explanatory variables used to train the RF model must be mapped across the model domain to create spatially continuous predictions of lithium. Thus, spatially continuous subsurface geologic information is key, although these digital resources are often difficult to acquire.
Interesting to me that RF performed better the XGBoost, would have expected at least a similar outcome if tuned correctly.
I would have guessed better results in the 1am to 2am time slot, but 3am is not totally out of line. I bet the fraction of drivers at 3am that are drunk is much higher than at, say, 3pm.
People downvoted you to the point that your comment is grayed out and about to be hidden but there is hardly metric by which Arkansas is not in the bottom ten on a list of states.
Infant mortality rate? 3rd most deadly for babies.
Poverty rate? 7th poorest.
Homicide rate? 7th most dangerous.
Obesity rate? 3rd fattest.
Practically any map of any measurable statistic where states are colored red for "bad" and green for "good" Arkansas will be a deep, blood, red.
People vote in good faith, I presume. Sometimes a comment’s factual basis matters less than its overall contribution to a productive and open discussion. Downvotes in this case are an example of HN’s surprisingly effective system for self-moderation working as it should. It isn’t vile enough to censor, but it also isn’t what a lot of readers come here for. It didn’t personally offend me (I didn’t vote either way), but I take occasional downvoting that I don’t fully agree with in stride, as the overall system seems to work better than most.
My guess is that the presence of lithium in the groundwater is in trace amounts if at all, while the dosing of lithium is in the domain of ~300mg. A casual search for the quantity of lithium in brine from a mine shows a max of 1400ppm for a rich mine in Chile[1] so drinking straight brine wouldn't get you anywhere near the therapeutic dose. Good question!
I am not a health researcher or anyting, but a quick googling seems to suggest its possible that it lowers risks of suicide[0] and other affective disorders, which by extension it would lower the rates of issues that can contribute to these issues I'd think.
That said, I honestly am unsure. It also is a requisite that it must be in the water in sufficient but low amounts
The formation is 7000 feet below the surface, if I understand correctly, so I don't think there would be any communication of its brine with potable groundwater.
I would like to think that if there were any interaction between theses putative deposits to the groundwater that we wouldn't have needed an ML model to find these deposits in the first place!
When the tide goes out on the AI hype there’s going to be a lot of companies currently using expensive API calls for simple classification tasks that will be quietly revamped to use a simple CNN.
ML is a toolbox of methods. Not every problem needs a transformer.
The USGS predictive model provides the first estimate of total lithium present in Smackover Formation brines in southern Arkansas, using machine learning, which is a type of artificial intelligence.
I was disappointed in that line. They could’ve mentioned it used a random forest, which is much more informative. “ML is a type of AI” isn’t even a cocktail party understanding of the topic.
Isn’t it a critical component of everything currently sporting anything remotely close to a legit “AI” label? I wouldn’t call cows “one part of a broader beef ecosystem” for example. They’re fundamental to it.
From the paper's method section, a bit more about which type of ML algo was used:
An RF machine-learning model was developed to predict lithium concentrations in Smackover Formation brines throughout southern Arkansas. The model was developed by (i) assigning explanatory variables to brine samples collected at wells, (ii) tuning the RF model to make predictions at wells and assess model performance, (iii) mapping spatially continuous predictions of lithium concentrations across the Reynolds oolite unit of the Smackover Formation in southern Arkansas, and (iv) inspecting the model for explanatory variable importance and influence. Initial model tuning used the tidymodels framework (52) in R (53) to test XGBoost, K-nearest neighbors, and RF algorithms; RF models consistently had higher accuracy and lower bias, so they were used to train the final model and predict lithium.
Explanatory variables used to tune the RF model included geologic, geochemical, and temperature information for Jurassic and Cretaceous units. The geologic framework of the model domain is expected to influence brine chemistry both spatially and with depth. Explanatory variables used to train the RF model must be mapped across the model domain to create spatially continuous predictions of lithium. Thus, spatially continuous subsurface geologic information is key, although these digital resources are often difficult to acquire.
Interesting to me that RF performed better the XGBoost, would have expected at least a similar outcome if tuned correctly.
Serious question:
Given the mood alerting properties of lithium, are people living here chiller than would be expected (controlling for instance for poverty / SES) ?
Potentially. See "Lithium in drinking water linked with lower suicide rates" [1].
[1] https://www.kcl.ac.uk/news/lithium-in-drinking-water-linked-...
I would assume any positive effects are balanced out by living in Arkansas.
My only experience with Arkansas was waking up to a speeding ticket at 3 in the morning. Who puts out a speed trap at 3 in the fucking morning?
But if it’s anything like Oklahoma…
I would have guessed better results in the 1am to 2am time slot, but 3am is not totally out of line. I bet the fraction of drivers at 3am that are drunk is much higher than at, say, 3pm.
Um, why were you waking up while driving at 3 in the morning?
Some cars have seats for up to seven people, including the driver.
Happened to me on Ambien.
People downvoted you to the point that your comment is grayed out and about to be hidden but there is hardly metric by which Arkansas is not in the bottom ten on a list of states.
Infant mortality rate? 3rd most deadly for babies.
Poverty rate? 7th poorest.
Homicide rate? 7th most dangerous.
Obesity rate? 3rd fattest.
Practically any map of any measurable statistic where states are colored red for "bad" and green for "good" Arkansas will be a deep, blood, red.
But it is rude to point that out.
People vote in good faith, I presume. Sometimes a comment’s factual basis matters less than its overall contribution to a productive and open discussion. Downvotes in this case are an example of HN’s surprisingly effective system for self-moderation working as it should. It isn’t vile enough to censor, but it also isn’t what a lot of readers come here for. It didn’t personally offend me (I didn’t vote either way), but I take occasional downvoting that I don’t fully agree with in stride, as the overall system seems to work better than most.
> But it is rude to point that out.
No, that is not rude at all. Making a flippant derogatory remark gets downvotes, people like to see numbers. Like the ones you just gave...
My guess is that the presence of lithium in the groundwater is in trace amounts if at all, while the dosing of lithium is in the domain of ~300mg. A casual search for the quantity of lithium in brine from a mine shows a max of 1400ppm for a rich mine in Chile[1] so drinking straight brine wouldn't get you anywhere near the therapeutic dose. Good question!
[1]: https://www.sciencedirect.com/science/article/abs/pii/S01691...
1400 ppm is one part in 700, so you'd get your dose from one cup (250 ml) of that brine.
I agree it's not likely you'd get a measurable effect from the local groundwater.
I am not a health researcher or anyting, but a quick googling seems to suggest its possible that it lowers risks of suicide[0] and other affective disorders, which by extension it would lower the rates of issues that can contribute to these issues I'd think.
That said, I honestly am unsure. It also is a requisite that it must be in the water in sufficient but low amounts
[0]: https://pubmed.ncbi.nlm.nih.gov/32716281/
It also shrinks your white matter I think, and has other gigantic bad effects.
Source: am bipolar and take 600mg daily.
The formation is 7000 feet below the surface, if I understand correctly, so I don't think there would be any communication of its brine with potable groundwater.
I would like to think that if there were any interaction between theses putative deposits to the groundwater that we wouldn't have needed an ML model to find these deposits in the first place!
Only when Mercury is in retrograde
Me thinks we might switch batteries to sodium in just a few years.
Love to see a project that uses bog standard ML techniques and doesn't call them AI. Respect.
When the tide goes out on the AI hype there’s going to be a lot of companies currently using expensive API calls for simple classification tasks that will be quietly revamped to use a simple CNN.
ML is a toolbox of methods. Not every problem needs a transformer.
Quoth the article:
I was disappointed in that line. They could’ve mentioned it used a random forest, which is much more informative. “ML is a type of AI” isn’t even a cocktail party understanding of the topic.
For a layperson, this is an accessible and directionally correct definition.
For the HN audience, of course this is 'technically incorrect'.
The article was written for the (larger) general public.
I am also glad they didn't squeeze in a word salad of LLMs and quantum technology and instead stuck to 'it's just standard ML'.
The only informational dividable from the statement is "we used a computer to analyze data".
ML is one particular field in the overall area of AI.
Isn’t it a critical component of everything currently sporting anything remotely close to a legit “AI” label? I wouldn’t call cows “one part of a broader beef ecosystem” for example. They’re fundamental to it.
It's the new Hacking vs Cracking. Or calling any computer a PC.
In this case it's Fracking
Nothing bog standard about contemporary ML. If anything calling it AI is underselling it.
This is what it was called back in the day. https://link.springer.com/article/10.1007/BF02478259