Image-identification AI is a vastly different beast than ChatGPT and these “art”-“making” bots. Beyond the variations in amount of energy used, AI is great for image identification, when properly trained. AI is not great for coming up with “new” stuff.
5 Likes
The popular claim is that ChatGPT uses half a litre of water for every 20-50 queries it responds to. In comparison, there’s a similarly famous estimate that it takes about 4 litres to grow a single almond.
That’s not entirely fair as my understanding is that almonds are a particularly water-hungry crop, but my point is that we probably underestimate the amount of water and energy that a lot of our regular activities require. I don’t think this article has a helpful title but the stats and graphs there are helpful to put things in context: https://andymasley.substack.com/p/individual-ai-use-is-not-bad-for
7 Likes
I think I would. I would still go out in nature the same amount, but – as I used to do before I found iNat – I would mainly be taking field notes for a Grinnell journal. Since I still do that, my use of iNat is additive, not substitutive.
As to the hours I spend identifying, before I got sucked into social media, I never had a problem filling my evenings – with reading, writing, or sewing/mending. I could go through four or five library books from one library visit without having to renew them. I used to listen to the radio, too – I had favorite programs on NPR, like “Democracy Now” and “A Prairie Home Companion,” which can only be enjoyed by active listening, not as background noise. iNat has replaced my other social media use (which had gotten excessive), but it still bothers me that I can’t seem to get back to those days before cyberspace.
6 Likes
Is that it? This feels like almost nothing.
Don’t get me wrong, I’m in the top 1% of AI haters. If I could snap my fingers and eliminate the abomination that is generative AI images/videos from the planet forever I would do it without hesitation, since that technology for specifically images has literally no benefit to society and tremendous downsides.
But I also feel the specifically environmental impact is overblown compared to the usual suspects of mining, parking lots, development, agriculture, etc. Maybe I’m just being contrarian?
Also if genAI led to the increased use of nuclear reactors, that would be by far the most useful thing to come out of the technology…
5 Likes
Self driving cars are more consuming than just an image recognation..which could be done with 300pixel photos as far as i know
Blockquote Neural Networks for Autonomous Vehicles: The development of autonomous vehicles involves training neural networks to interpret sensory data and make decisions in real-time. This process requires processing vast datasets of road images, videos, and sensor readings to cover a wide range of driving conditions and scenarios. The computational effort to train, test, and refine these models is enormous, resulting in significant energy usage .
But is known how much each image recognation request cost in carbon, C02, watt and energy ?
LPIRC started in 2015 with the aim to identify the technologies for computer vision using energy efficiently. IEEE Rebooting Computing has been the primary financial sponsor and provides administrative support. In 2019, sponsors are Google, Mediatek, Xilinx, and IEEE Circuits and Systems Society. Past sponsors include Facebook, Nvidia, IEEE GreenICT, IEEE Council of Design Automation, IEEE Council on Superconductivity. The 2019 LPIRC is organized by Purdue University and Duke University. Google provides a unified latency metric and an evaluation platform for the online track (also called On-Device Visual Intelligence Challenge).
In 2020, LPIRC is renamed to Low-Power Computer Vision Challenge (LPCVC). A new track is added for processing video captured by UAV (unmanned aerial vehicle). The 2020 LPCVC sponsors are Facebook, Xilinx, ELAN Microelectronics, Google, IEEE Rebooting Computing, IEEE Council of Electronic Design Automation, IEEE Circuits and Systems Societies. A workshop will be held on June 15, 2020 in Seattle (co-located with CVPR). More information is available at [lpcv.ai](https://lpcv.ai/).
``` https://andymasley.substack.com/p/individual-ai-use-is-not-bad-for
1 Like
alex said on the linked thread:
Our current computer vision training workstation, at full train, seems to pull 450-500 watts (per nvidia-smi and the rated TDP of our CPU).
At 120 training days, and 24 hours in a day, I believe that works out to ~1,400 kilowatt hours. However, I’m stupid with electrical stuff (and most stuff, honestly), so I’m happy to be corrected.
From what I understand, CV is predictive AI not generative AI. LLM like ChatGPT consume a lot more power being generative. The whole term AI is fairly stupid tgough, it leads to people thinking it can be or do things it cannot. LLMs can use a lot of power, and this should be a big concern.
This does bring up some good talking points, however.
It is best to try to use efficient computers in general. For example, my Raspberry Pi 4B can consume very little power but you need to account for the monitor as well. You can train a computer vision model, load it onto a Raspberry Pi AI camera, and once loaded the setup can run off a 15W 3.0A USB power supply, 0.05W under no load, so it depends on what USB, HDMI, or cameras you attach to it. The point is you can do desktop computing for lower power and lower cost as long as you don’t need do do anything like gaming or video editing.
8 Likes
Here to chip in on the whole “nuclear good or bad” thing.
Obviously it’s a nuanced issue. Nuclear waste, when disposed of properly, is not actually as damaging to the environment as some would make it seem. Thermal pollution can be an issue, and obviously any large construction product will use lots of energy and resources. Not to mention that the whole mining process is pretty bad in general. But that’s in a vacuum. One of the most difficult things about environmental science is that basically everything is bad. The idea is to do the most good by using the least bad tools at our disposal, and hopefully ensure a future where people won’t have to rely so much on the bad stuff. Right now, nuclear is one of the best options we have.
As for AI, as others have said, AI is kind of a stupid and nebulous term that makes no sense. At the end of the day, (in my opinion), the good being done here far outweighs the energy usage of the site, especially in comparison to the evil corporate conglomerates who employ generative AI without thought for the consequences. Just my two cents though.
10 Likes
Oh in case it wasn’t clear, I love the iNat CV. I was talking about generative images, which I pretty much just outright despise. What iNat is doing is fundamentally not generative AI and using “AI” for everything even vaguely involving machine learning is a rather annoying quirk of the last five years for me
6 Likes
As others have said, iNat’s use of CV uses much less electricity than the generative/LLM stuff that grabs the headlines, I don’t think they’re really comparable. So I don’t think iNat’s AI is a major contributor to energy use/carbon emissions.
For what it’s worth the training machines (basically a few small computers in room, not a big server farm) are hooked up to electricity that claims to be 100% sourced through renewable resources, meaning that the money we pay for electricity goes to purchase renewable energy.
13 Likes
Much is made about radioactive nuclear waste, for good reason, but there’s also a lot of misinformation or misunderstanding about it and radioactive waste in general.
A big one is the idea that nuclear power plants are the main source of radioactive waste… they’re not, coal plants are. Uranium and thorium in the fly ash produced by the coal plants is the primary source of this, and even after capture the little bit that escapes the capture results in massive amounts of radioactive pollution released.
Among the surprising conclusions: the waste produced by coal plants is actually more radioactive than that generated by their nuclear counterparts. In fact, the fly ash emitted by a power plant—a by-product from burning coal for electricity—carries into the surrounding environment 100 times more radiation than a nuclear power plant producing the same amount of energy.
Another misunderstanding is the amount of nuclear waste of the type that requires long-term storage, that is spent nuclear fuel. It’s immensely less than people realize, totaling to around 400,000 tons in total as of 2022-2023 (Edit for clarity: this amount is the total amount produced globally over the the entire time nuclear power has been on use on the planet).
That sounds like a lot, but let’s run some numbers to get a more realistic assessment of just what that means. 1 cubic meter of plutonium is a bit less than 20 tons, and for uranium it’s 19 tons per cubic meter. That means that the 400,000 tons waste is around 20,000 cubic meters volume. That’s a block 50 meters long, 20 meters wide, and 20 meters thick. Put another way that’s roughly 260 shipping containers (the larger size). For comparison, the average cargo ship can carry 15,000 shipping containers, so all the the seriously hazardous nuclear waste produced takes up less that 2% of the volume of the average cargo ship capacity (volume, not mass).
In addition, this type of waste can be reprocessed and iself used as a fuel source, greatly reducing the total amount of waste, since the waste becomes a resource. 90-96% of the spent fuel can be repurposed this way, and spent fuel still contains around 90% of its potential energy, so there is a lot of power still available from it. Unfortunately, in the US the anti-nuclear sentiment has stymied work in this and the US does not repurpose spent fuel, but France does.
It’s estimated that well managed fast nuclear reactors reprocessing and using spent fuel would extract 50-100x more energy than we currently get from the single use system currently in use, which would mean both a massive reduction in the amount of fuel initially needed, as well as vastly less overall final waste (which would also be less radioactive). In addition, it would mean being able to make use of the existing waste and reduce that as well.
None of this is to say that nuclear waste is not a concern, nor is it to ignore the secondary short-lived waste products (although those are short lived and become effectively inert in short order), nor is it to present a false dichotomy between coal and nuclear as the only options. What it is meant to do is highlight the fact that many people who involve themselves in the discussion over nuclear power and issues of waste and safety don’t have a good grasp of the actual numbers involved, what the larger sources of radioactive pollution are and wind up overhyping dangers volumes as as result.
8 Likes
Coal versus ‘protected’ baobabs in Limpopo (South Africa)
The more uncomfortable choice is to change our lifestyle and use less power. Our geyser is solar, but on this grey autumn day the timer is set for 1600 to bring the temperature up (in summer that is not needed).
But we also don’t need the ahem power shower for my hair.
2 Likes
Still, I’m just saying it’s not perfect.
interesting thread here (if you suffer x.com) - x.com/_KarenHao/status/1763536620752437602
I don’t think this is in any way likely.
Answering the title question (Is the iNaturalist use of artificial intelligence damaging this planet) somewhat indirectly here, we could ask if iNaturalist use of AI is healing the planet? I can think of a few ways:
-
iNaturalist has in my opinion led a revolution in habitat and species mapping, while being invaluable to hundreds (if not more) of ecological studies, including those to understand the effects of 120 years of climate change on flowering phenology of Joshua trees so that we can better predict their future distribution and aid decision-making with regard to assisted migration to preserve the species’ genetic integrity. Without the big data provided by iNaturalist, studies like these would simply not be possible with the current levels of funding and vocational scientists.
-
Additionally, speaking as someone who has performed biological assessments and understands the current “state of the art”, biologists are often undertrained, not formal botanists/herpetologists/ornithologists (etc.), and/or are asked to assess species in an area they’re unfamiliar with; and thus, iNaturalist becomes an invaluable resource to help narrow down which species are in an area and potentially detect and safeguard protected species.
-
Further, iNaturalist has (and continues) to help discover new species (and rediscover hundreds of “lost” species , some of which are considered endangered at discovery. There can’t be species protection for species science doesn’t know exist, so in this way iNaturalist also helps protect the planet, converting a relatively small carbon footprint in exchange for vitally-needed understanding of the planet and its ecology as we reach new ecological tipping points.
-
The role of iNat in environmental education and increasing citizen engagement with nature is difficult to measure, but shouldn’t be ignored. People who engage more with nature through the app learn to recognize and appreciate their local biodiversity, and are more likely to engage with protecting it.
So to me at least, the large benefits significantly outweigh any environmental impacts, even if we bring up the more controversial sources of energy (coal, nuclear, etc.)
P.S. these are just some of the big reasons why I donate to iNat! :)
10 Likes
That’s great if you are making that personal decision to use less power, but the reality is most people (and corporations) will not do this. Wind and solar cannot support the full demand of the power grid without massive power farms that cover vast areas (and perhaps not even then), posing hazards to wildlife and damaging habitats in the process. The realistic options are either sticking with fossil fuels or going nuclear
Nothing is perfect. There is no power generation system that doesn’t come with negative environmental consequences.
6 Likes
Sorry, what?
Please provide references for your claims about the amount of space required for renewable power sources and the environmental destruction connected with this, as well as a comparison with the environmental destruction caused by fossil fuel power plants.
There are numerous states/countries in the Global North that are already able to meet 100% of their electricity demands using renewables for at least a few days per year, and this number continues to increase. As far as I know, they are doing this without blanketing “vast areas” with power farms. I am quite aware that renewable energy sources are not without environmental and social costs (particularly e.g. hydroelectric dams), but at least in Europe, the regulations for building large-scale power farms are fairly strict and include environmental impact assessments.
It has always been clear that phasing out fossil fuels will not just require replacing it with renewable sources, but also learning to be more sparing in our energy use, particularly those of us in power-hungry countries.
Edit: Since I asked for sources, here are some estimates that suggest renewable energy can meet demands without creating major land-use conflicts or habitat destruction even though it requires more space per kw produced than fossil fuels:
https://blog.ucs.org/steve-clemmer/how-much-land-would-it-require-to-get-most-of-our-electricity-from-wind-and-solar/
https://eeb.org/ample-land-for-sustainable-renewables-expansion-in-europe-new-study-reveals/
It also isn’t necessarily an either-or situation – solar can be installed on buildings or “intercropped” on agricultural land (agrivoltaics), wind turbines are often installed among fields, etc.
5 Likes
We can place solar panels on roofs and over parking lots, which would not damage farmland or wild lands. Trouble is, that’s more expensive than putting the panels in a field. Therefore, subsidies are necessary, or requirements that new construction include the panels. It could be done.
7 Likes
I bet iNat AI is not among the most worst energy demanding ones. Maybe you should rather be more concerned about other AIs, also for the possibility they will substitute human beings for certain jobs.
3 Likes
It has been said, “AI is not going to take your job, but people who use AI might.”
An example is my experience as an editor. I had one long-term gig where I essentially worked in tandem with an AI – the AI did the low-value “slog work” of correcting spelling and grammar and sometimes adjusting the phrasing; then I came along behind it and did the higher-value edits of ensuring that the author’s intended nuances of meaning were preserved, correcting where the AI had made a wrong assumption, and optimising phrasing. Far from taking my job, the AI improved my time efficiency.
2 Likes