Some climate-tech startups want you to believe AI tools can save the planet — but it's not that simple

Artificial-intelligence-powered robots that can separate a water bottle from a diaper may bring the world one step closer to a cooler planet — or at least that's what Matanya Horowitz hopes.

Horowitz, an engineer, started AMP Robotics after wrapping up his doctorate in robotics at Caltech. Founded in 2014, the Colorado company designs AI-powered robots to identify recyclables like office papers, aluminum cans, and plastic bottles in heaps of waste. The goal, he said, is to automate a recycling process that often relies on manual labor.

The benefits are twofold: more reusable material to sell to processors and fewer recyclable items entering landfills.

"At the core, we're trying to make recycling a more lucrative business with the goal of catalyzing much wider adoption of recycling around the world," Horowitz told Insider.

AMP Robotics is just one of more than 44,500 climate tech startups that have emerged since 2010. Last year investors poured $70.1 billion into climate tech, an 89% rise compared with 2021, according to HolonIQ Global Impact Intelligence. And since AI has become the darling of the tech world, it has been incorporated into some of these startups.

While AI tools like Horowitz's waste-sorting robots have been around for years, the advent of generative-AI tools like OpenAI's ChatGPT has reignited conversations around deploying AI to address societal issues.

The 2023 report from the UN's Intergovernmental Panel on Climate Change says cutting-edge tech like AI can "increase energy efficiency" and "promote the adoption" of renewable energy that, in theory, could lower carbon emissions while stimulating economic growth.

But some machine-learning researchers and climate activists told Insider that using AI to address the climate crisis could cause more harm than good if tools weren't carefully deployed. AI tools are complex, contain flaws, and may require significant amounts of natural resources to run — factors that could delay the goal of reaching net-zero emissions by 2050, which aims to limit global average temperatures to the crucial threshold of 1.5 degrees Celsius.

AMP Robotics had raised more than $176 million as of early May, per PitchBook, and Horowitz says it has deployed "hundreds of robots" around the world. He said the robots had helped his clients sort through a higher volume of waste faster. Carling Spelhaug, an AMP Robotics representative, told Insider the robots had saved "an estimated 1.6 million metric tons of greenhouse-gas emissions" from the end of 2021 to May 2023, the equivalent of "removing 326,000 cars from the road" during that same time frame.

Nevertheless, Horowitz acknowledges his AI robots aren't perfect.

"We expect there to be a learning process for the next 10 years," Horowitz said.

AI is best used to collect quality climate data

Climate change has already "caused substantial damages" to certain ecosystems, with "increasingly irreversible losses," the IPCC said in a summary of its 2023 report. It has exposed millions of people to food and water insecurity, disproportionately hurting low-income households, Indigenous folks, and vulnerable communities in places such as Africa and South America, the UN agency said.

Sasha Luccioni, a research scientist and climate lead at the AI platform Hugging Face, told Insider that AI could be deployed to understand the extent of who's most affected.

"AI is a really impactful tool in crunching that data and helping society be like, 'OK, well, let's focus on this particular aspect as a short-term solution,'" said Luccioni, who is also a founding member of the nonprofit Climate Change AI.

Luccioni pointed to Rainforest Connection, a bioacoustic monitoring tool that deploys AI-powered sensors in rainforests to flag noises like the sound of a chainsaw. Park rangers can use this data to investigate the noise, which the organization says can prevent illegal deforestation — a contributor to climate change.

David Rolnick, a computer-science professor at McGill University who studies AI applications to the climate crisis, agreed. Research he contributed to finds that AI can "distill data into useful information" to calculate the financial impacts of environmental damage, identify carbon-intensive parts of supply chains, and predict a firm's energy needs.

"Leveraging that data, even in relatively simple ways, can scale up action in ways that the manual actions can sometimes not," Rolnick, a cofounder and chair of Climate Change AI, told Insider.

Banks, manufacturers, and healthcare firms have turned to Persefoni, a carbon-management platform that uses AI to measure its clients' carbon footprints.

Demand for Persefoni's services could be a result of the government, investors, and shareholders putting pressure on businesses to reduce their carbon footprint, James Newsome, the company's chief data officer, told Insider.

Newsome said many businesses didn't have the tools or skills needed to manually identify emission sources in their operations — a process he said could be complicated and time-consuming.

Applying AI to the process, he said, can help clients detect outliers in their data and "suggest changes" on which parts of the business need emission cuts. An airline, for instance, could use Persefoni's software to compute how much carbon is emitted when burning jet fuel during flights and then use these insights to adjust its energy habits.

In turn, companies might develop climate action plans cheaper and faster.

"Tech has really democratized and helped the people in the companies who may not have had the expertise," Newsome said.

But researchers warn AI may do more harm than good

Researchers, activists, and climate tech execs agree that AI can't single-handedly cool the planet.

Rolnick says the hype around the technology may drive businesses to adopt advanced AI tools when "low-tech" options like building sensors that detect energy levels or greener insulation to reduce overall carbon emissions could be used instead.

"There may be situations where AI is overkill," Rolnick said. "It's very easy for people to jump to the fancier technology rather than the unflashy low-tech option, even if the low-tech option is more impactful."

Certain climate activists agree.

Bill McKibben, the founder of 350.org, a global environmental nonprofit, pointed to existing tools like solar panels and wind turbines as capable of reducing carbon emissions. Calculating emissions is "not so complicated that you need an AI," McKibben told Insider.

At the same time, focusing on AI solely as a climate solution may obscure the bigger picture — that climate change is linked to global economic growth, Helena Norberg-Hodge, a climate activist critical of market-based solutions to the climate crisis, told Insider.

"Before we look at the benefits of using AI for measurement, we need to be looking at what we're measuring and why," Norberg-Hodge said.

Researchers acknowledge that AI can be used in ways detrimental to the climate. Oil and gas companies, for instance, deploy AI and advanced analytics to "accelerate exploration and extraction," which by one estimate may generate up to $425 billion in added value for the industry by 2025, according to a Greenpeace report.

Deploying AI tools is also energy intensive. Luccioni said large language models like GPT, the model behind OpenAI's ChatGPT, were "super compute intensive" and less sustainable than simpler technologies, requiring significant amounts of water to train.

Even AI climate tech executives see limits to their impact

Even climate tech executives believe there are limitations to AI.

Horowitz, the CEO of AMP Robotics, said his robots didn't always sort through recyclables accurately, and Newsome, the data officer at Persefoni, said the quality of environmental data could vary by client, making it harder to accurately understand the full extent of their carbon footprint.

Despite these limitations, Horowitz, Newsome, and researchers agree that AI tools are useful as part of a larger, more comprehensive climate solution that involves many stakeholders, including the government, local communities, and climate scientists across a range of disciplines.

Without engaging a variety of groups, those who build the AI may "work in a silo," running the risk of the models being misunderstood, misused, and inaccessible, especially to those most vulnerable to the climate crisis, Luccioni said.

"You need people in the room who understand the AI, people in the room who understand the area of application, and often, people in the room who will be affected by the technology," Rolnick said.

This article is part of "The Great Transition," a series covering the big changes across industries that are leading to a more sustainable future. For more climate-action news, visit Insider's One Planet hub.