Scientists Just Found Something Unbelievably Grim About Pollution Generated by AI
Tech companies are hellbent on pushing out ever more advanced artificial intelligence models — but there appears to be a grim cost to that progress.
In a new study in the science journal Frontiers in Communication, German researchers found that large language models (LLM) that provide more accurate answers use exponentially more energy — and hence produce more carbon — than their simpler and lower-performing peers. In other words, the findings are a grim sign of things to come for the environmental impacts of the AI industry: the more accurate a model is, the higher its toll on the climate.
"Everyone knows that as you increase model size, typically models become more capable, use more electricity and have more emissions," Allen Institute for AI researcher Jesse Dodge, who didn't work on the German research but has conducted similar analysis of his own, told the New York Times.
The team examined 14 open source LLMs — they were unable to access the inner workings of commercial offerings like OpenAI's ChatGPT or Anthropic's Claude — of various sizes and fed them 500 multiple choice questions plus 500 "free-response questions."
Crunching the numbers, the researchers found that big, more accurate models such as DeepSeek produce the most carbon compared to chatbots with smaller digital brains. So-called "reasoning" chatbots, which break problems down into steps in their attempts to solve them, also produced markedly more emissions than their simpler brethren.
There were occasional LLMs that bucked the trend — Cogito 70B achieved slightly higher accuracy than DeepSeek, but with a modestly smaller carbon footprint, for instance — but the overall pattern was stark: the more reliable an AI's outputs, the greater its environmental harm.
"We don't always need the biggest, most heavily trained model, to answer simple questions," Maximilian Dauner, a German doctoral student and lead author of the paper, told the NYT. "Smaller models are also capable of doing specific things well. The goal should be to pick the right model for the right task."
That brings up an interesting point: do we really need AI in everything? When you go on Google, those annoying AI summaries pop up, no doubt generating pollution for a result that you never asked for in the first place.
Each individual query might not count for much, but when you add them all up, the effects on the climate could be immense. OpenAI CEO Sam Altman, for example, recently enthused that a "significant fraction" of the Earth's total power production should eventually go to AI.
More on AI: CEOs Using AI to Terrorize Their Employees
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
The Verge
15 minutes ago
- The Verge
Tech billionaire's sunken superyacht raised from the sea.
Posted Jun 22, 2025 at 5:22 PM UTC Tech billionaire's sunken superyacht raised from the sea. The AP, CNN, and Reuters report that Italian authorities are continuing to investigate last year's sinking of Autonomy founder Mike Lynch's superyacht, Bayesian, during a storm last year that killed seven people, including Lynch. A salvage operation lifted the 184-foot from the seafloor this weekend, after first removing its 236-foot mast.
Forbes
an hour ago
- Forbes
What The June 28 European Accessibility Act Deadline Means For Brands
This photograph taken on March 19, 2025 shows European flags outside the EU headquarters in ... More Brussels. (Photo by Nicolas TUCAT / AFP) (Photo by NICOLAS TUCAT/AFP via Getty Images) In less than a week from now, on June 28, 2025, the requirements of the European Accessibility Act come into force. The intention behind the EAA is to provide fair and equitable access to digital products to the more than 100 million consumers with disabilities living across the European Union. One of the key things to keep in mind about the EEA is that it doesn't just affect European companies. It applies to any organization from around the world selling into the EU market. It goes beyond standard web accessibility and e-commerce, incorporating areas such as computers and operating systems, smartphones, digital TV services, ATMS, ticketing machines and other digital kiosks within its remit. The legislation requires companies to do more than simply make digital products and content accessible. Instead, it seeks to embed accessibility commitments and best practices at a more fundamental organizational level through requirements like official accessibility statements and providing specialized training to staff. Non-compliance with the EAA can lead to fines and ultimately the termination of EU market access. The proximity of the deadline may for some be a cause for panic but cutting through the noise, here are what some digital accessibility specialists are urging organizations to keep in mind with less than a week to go: It's still not too late to get started 'It's not too late to start addressing accessibility. Starting now is better than not doing anything about EAA compliance,' says Jon Avila, Chief Accessibility Officer at Level Access, one of the leading providers of digital accessibility services in the United States. Avila continues, 'For organizations just getting started, the most important steps to take now to mitigate risk are demonstrating that you're taking action to make sustained progress, and beginning to put processes in place aligned with EAA requirements. One of the biggest gaps we observe is third-party content. Companies often assume that embedded tools or plugins are outside their responsibility, but under the EAA, they're very much in scope. We recommend that organizations make accessibility a requirement during the procurement process and request documented proof of accessibility (e.g., a completed VPAT report) from vendors prior to making a purchase.' Eric Portis, a Developer Evangelist at Cloudinary, a software company that helps organizations manage their visual content online, also agrees that it's still not too late to get started with EAA compliance. 'The best time to consider the accessibility of a project was before you shipped it, but the second-best time is today. The Web Content Accessibility Guidelines are just a big checklist. And while checking every item off the list before June is not a realistic goal for many teams, everyone can start tackling an item or two and begin building momentum towards full compliance. All progress is good progress,' Portis says. The AI landscape has changed Back in 2020, when AI website accessibility overlay widgets hit the market, they were met with a degree of skepticism. That's because the marketing claims were bold – buy a software subscription and let AI do its magic in both spotting and fixing accessibility fails, saving you time and money. It quickly became widely accepted that the technology was, in fact, nowhere near ready to entirely replace human oversight for accessibility remediation. Fast forward five years, and the same remains true but the technology has undoubtedly gotten a lot better meaning that, though AI won't solve all digital accessibility issues, it can now do a lot more of the heavy lifting which genuinely allows teams to focus more precisely on the more nuanced human elements related to user experience. What this ultimately means for the market was recently brought into sharp relief in a report authored by AAAnow, leading digital compliance benchmarking specialists based in London. AAAnow's report looked at the websites of organizations comprising The Valuable 500, a collective of some of the most profitable companies on the planet with long-standing pledges to accessibility and inclusion, including the likes of Apple, Google and BP. According to the V500 scorecard, the average member organization manages around 178 websites, each comprising around 211 pages. To remediate accessibility on the average of 85% of pages where it is required would take 3,480 workdays, or 13 full-time staff working for a year (per org). In terms of cost, this equates to $767,000 per company just for initial compliance, let alone maintenance. Scaled across the full Valuable 500, that's over $380 million and 9 years of full-time work at current capacity. 'We argue that only a risk-managed, AI-powered approach can address the scale, cost, and operational complexity. It's not a 'nice-to-have' - it's fundamental,' commented AAAnow CEO Lawrence Shaw in an earlier interview. Regulation is not something to fear Instinctively, when any type of new regulation comes into force, particularly one that potentially involves fines and other punitive measures, fear and trepidation tend to be the most common reactions off the bat. Yet, it can be argued that brands should welcome the new regulations from the EU as they provide clarity and a roadmap into areas of technical compliance that have previously been murky. In a recent company press release, Amit Borsok, CEO of Italy-based Accessiway, which takes an end-to-end approach to accessibility remediation involving both automation and human testers, said, 'European regulations have shifted the narrative around accessibility from a compliance issue to a strategic investment. Until now, regulatory fragmentation across member states has prevented companies from fully capitalizing on this potential, increasing compliance costs and limiting competitiveness. 'Now, instead of being a hurdle, the new framework provides clarity that encourages companies to invest in accessible technologies and services to expand their market reach. For many organizations, accessibility has evolved from a nice-to-have to a competitive necessity. With regulation now offering predictability and standardization, uncertainty is reduced for investors, and new opportunities for innovation are unlocked.' Though the word deadline usually implies finality, whether, as an organization, you are right at the start of your digital accessibility journey and anxiously playing catch-up, or significantly further down the line, June 28 will surely be the opening of a new chapter for all involved. The story about to unfold will be one of whether human intention layered onto ever more advanced technology can ultimately help make access to the digital realm an equitable experience for everyone.
Yahoo
an hour ago
- Yahoo
Google AI is worse at Pokemon than I was when I was 5 – taking 800 hours to beat the Elite 4 and having a breakdown when its HP got low
When you buy through links on our articles, Future and its syndication partners may earn a commission. If you're someone who thinks AI is almost ready to take over the world, I have some good or bad (depending on your stance on things) news for you: Google's Gemini 2.5 Pro took over 800 hours to beat the 29-year-old children's game Pokemon Blue. There's a Twitch account called Gemini_Plays_Pokemon, a pale imitation of the incredible Twitch Plays Pokemon account that started this trend. First things first: how long did it take the AI to actually complete the game? Well, it was a staggering 813 hours. I feel like you could hit buttons randomly and beat the game faster than that. After some tweaks by the creator of this Twitch channel, the AI managed to halve its time to a still outrageous 406.5 hours. That is actually dead on half the time, which is interesting mathematically but still far too long to beat a game you can win with an overleveled Venusaur. Additionally, as spotted by our friends at PC Gamer, Google DeepMind reported on the Twitch account, and something unusual happens whenever its Pokemon get low on health or power points (PP). Whenever one or both of these conditions are met, "model performance appears to correlate with a qualitatively observable degradation in the model's reasoning capability – for instance, completely forgetting to use the pathfinder tool in stretches of gameplay while this condition persists." This, combined with the AI mistakenly thinking it was playing FireRed and LeafGreen and would need to find the Tea to progress, are part of the reasons it took so long to finish. Honestly, AI just isn't very good at playing Pokemon. Someone else made Claude Plays Pokemon, and that AI spent hours trying to get out of Cerulean city because it kept jumping down a ledge to talk to an NPC it had already spoken to dozens of times. So, these AIs aren't able to beat a game that we could when we barely knew our times tables. Let's not worry about them taking our jobs any time soon. In the meantime, check out the best Pokemon games of all time.
