Search the archive.

Here only worldwide events are included. A widespread endemic disease that is stable in terms of how many people become sick from it is not a pandemic. 260 3.1.4.1 Expected impact Infectious diseases have been one of the greatest causes of mortality in history. Unlike many other global challenges pandemics have happened recently, as we can see where reasonably good data exist. Plotting historic epidemic fatalities on a log scale reveals that these tend to follow a power law with a small exponent: many plagues have been found to follow a power law with exponent 0.26.261 These kinds of power laws are heavy-tailed262 to a significant degree.263 In consequence most of the fatalities are accounted for by the top few events.264 If this law holds for future pandemics as well,265 then the majority of people who will die from epidemics will likely die from the single largest pandemic. Most epidemic fatalities follow a power law, with some extreme events – such as the Black Death and Spanish Flu – being even more deadly.267 There are other grounds for suspecting that such a high impact epidemic will have a greater probability than usually assumed. All the features of an extremely devastating disease already exist in nature: essentially incurable (Ebola268), nearly always fatal (rabies269), extremely infectious (common cold270), and long incubation periods (HIV271). If a pathogen were to emerge that somehow combined these features (and influenza has demonstrated antigenic shift, the ability to combine features from different viruses272), its death toll would be extreme. Many relevant features of the world have changed considerably, making past comparisons problematic. The modern world has better sanitation and medical research, as well as national and supra-national institutions dedicated to combating diseases. Private insurers are also interested in modelling pandemic risks.273 Set against this is the fact that modern transport and dense human population allow infections to spread much more rapidly274, and there is the potential for urban slums to serve as breeding grounds for disease.275 Unlike events such as nuclear wars, pandemics would not damage the world’s infrastructure, and initial survivors would likely be resistant to the infection. And there would probably be survivors, if only in isolated locations. Hence the risk of a civilisation collapse would come from the ripple effect of the fatalities and the policy responses. These would include political and agricultural disruption as well as economic dislocation and damage to the world’s trade network (including the food trade). Extinction risk is only possible if the aftermath of the epidemic fragments and diminishes human society to the extent that recovery becomes impossible277 before humanity succumbs to other risks (such as climate change or further pandemics).

The AI/ML models are now extensively used to expedite drug discovery; the FDA has taken proactive actions to ensure that machine-made decisions are at least as robust and reliable as human analysis; in certain instances, AI/ML is now used to enhance human decision-making ability. Dozens of regulations, guidelines, and White Papers have been introduced to ensure that machine-driven decisions are continuously challenged. Of the tremendous significance of AI/ML, the first is its ability to process extensive data that is impossible to achieve otherwise. The entry of quantum computers is waiting to make these analyses more efficient. Second, data analysis in the research and manufacturing to optimize the process and validate findings. An excellent example of how this aspect has been the mainstream is the 21CFR Part 11 compliance that ensures digital data storage remains accurate and unalterable. Third, AI/ML allows experimental design creation that can be the most creative application to avoid following traditional paths, despite their efficiency or inaccuracies. The design space includes statistical modeling that can help reduce the study size by calculation from large databases that would generally not be possible using human inquiry alone. The AI/ML applications will expand exponentially; while, in 2021, more than hundreds of FDA filings deployed AI/ML tools, I anticipate most filings to engage these tools somehow; thus, scientists must receive proper training in using these tools. The FDA and related agencies have published many guidelines, policies, and data handling suggestions that require personnel trained to identify redundancies in implementing AI/ML at every stage of development and manufacturing. The role of AI/ML in manufacturing brings the traditional models to a different stage—advanced manufacturing that has benefitted significantly from automation, machine learning, and decision-making in continuous process control. None of these improvements were possible by human interaction alone. While AI/ML enhance productivity, this will not impact the job market since humans will still be needed to create the algorithms, supervise their operation, and finally judge their conclusions. To introduce the many applications of AI/ML, the FDA is open to earlier meetings to discuss the nature and extent of reliance on AI/ML in drug development or manufacturing. Developers should capitalize on these opportunities. It should be noted that the FDA guidelines admit that they are learning too when these interactions arise; the goal is to make the discovery more efficient, and thus more affordable; and manufacturing more reliable and efficient. All these plans will help reduce the cost of drugs to patients, and that alone should be a sufficient reason to remove mindsets about AI/ML, perhaps using another AI/ML model. However, it all starts with assuring that the myths around AI/ML are curbed at the onset. The industry should trust the FDA’s efforts to make drug discovery more productive, manufacturing more compliant, and clinical testing more ethical.

AI-powered job automation is a pressing concern as the technology is adopted in industries like marketing, manufacturing and healthcare. By 2030, tasks that account for up to 30 percent of hours currently being worked in the U.S. economy could be automated — with Black and Hispanic employees left especially vulnerable to the change — according to McKinsey. Goldman Sachs even states 300 million full-time jobs could be lost to AI automation.

“The reason we have a low unemployment rate, which doesn’t actually capture people that aren’t looking for work, is largely that lower-wage service sector jobs have been pretty robustly created by this economy,” futurist Martin Ford told Built In. With AI on the rise, though, “I don’t think that’s going to continue.”

Schmidhuber’s position contrasts with a number of his contemporaries, including Dr Geoffrey Hinton, who spectacularly quit Google this week after a decade with the company in order to speak more freely on AI.

Hinton, who is referred to as the godfather of AI, won the Turing award in 2018 for his work on “deep learning”, which is the foundation for much of the AI in use today.

In this section, we describe three sets of threats associated with the misuse of AI, whether it be deliberate, negligent, accidental or because of a failure to anticipate and prepare to adapt to the transformational impacts of AI on society. The first set of threats comes from the ability of AI to rapidly clean, organise and analyse massive data sets consisting of personal data, including images collected by the increasingly ubiquitous presence of cameras, and to develop highly personalised and targeted marketing and information campaigns as well as greatly expanded systems of surveillance. This ability of AI can be put to good use, for example, improving our access to information or countering acts of terrorism. But it can also be misused with grave consequences. The use of this power to generate commercial revenue for social media platforms, for example, has contributed to the rise in polarisation and extremist views observed in many parts of the world.21 It has also been harnessed by other commercial actors to create a vast and powerful personalised marketing infrastructure capable of manipulating consumer behaviour. Experimental evidence has shown how AI used at scale on social media platforms provides a potent tool for political candidates to manipulate their way into power.22 23 and it has indeed been used to manipulate political opinion and voter behaviour.24–26 Cases of AI-driven subversion of elections include the 2013 and 2017 Kenyan elections,27 the 2016 US presidential election and the 2017 French presidential election.28 29 When combined with the rapidly improving ability to distort or misrepresent reality with deepfakes, AI-driven information systems may further undermine democracy by causing a general breakdown in trust or by driving social division and conflict,26–28 with ensuing public health impacts.

Several countries are currently using AI surveillance technologies. These tools are transforming international relations between states as they are transforming international security but also relations inside the state between the governor and governed. These advanced surveillance technologies are facilitating the transition from democratic to authoritarian states. AI and how it’s used by different actors in international society and societies leads us to the AI surveillance as a new indicator enabling to measure democracy and to identify authoritarian and democratic countries. In a sense, “The index compiles empirical data on AI surveillance use for 176 countries around the world” (Carnegie Endowment for International Peace 2021). These AI surveillance tools offer several advantages to states to ensure security through social control and the surveillance of citizens and non-nationals including diplomats and representatives of other states in the state’s territory. According to the AI Global Surveillance (AIGS) Index, “At least seventy-five out of 176 countries globally are actively using AI technologies for surveillance purposes. China is a major driver of AI surveillance worldwide exporting these advanced technologies and its authority’s system to sixty-three countries” (Kane 2017). US companies are also active in this space in 32 countries. A growing number of states are deploying advanced AI surveillance tools from East and the West such as France, Germany, and Israel. Japan is also playing important roles in proliferating this technology. That means that in AI era, democracy needs to be revised considering the new evolution in AI surveillance and how they are used by democratic and authoritarian countries and how these tools are creating a new tyranny, a technological dictatorship. In a sense, the AIGS Index shows that 51% of advanced democracies deploy AI surveillance systems, and 37% of closed autocratic states, 41% of electoral autocratic/competitive autocratic states, and 41% of electoral democracies/illiberal democracies deploy AI surveillance technology (Feldstein 2019) (Fig. 1).

However, as with all technologies, AI can be applied in ways that are detrimental. The risks associated with medicine and healthcare include the potential for AI errors to cause patient harm,11 12 issues with data privacy and security13–15 and the use of AI in ways that will worsen social and health inequalities by either incorporating existing human biases and patterns of discrimination into automated algorithms or by deploying AI in ways that reinforce social inequalities in access to healthcare.16 One example of harm accentuated by incomplete or biased data was the development of an AI-driven pulse oximeter that overestimated blood oxygen levels in patients with darker skin, resulting in the undertreatment of their hypoxia.17 Facial recognition systems have also been shown to be more likely to misclassify gender in subjects who are darker-skinned.18 It has also been shown that populations who are subject to discrimination are under-represented in datasets underlying AI solutions and may thus be denied the full benefits of AI in healthcare.16 19 20 Although there is some acknowledgement of the risks and potential harms associated with the application of AI in medicine and healthcare,11–16 20 there is still little discussion within the health community about the broader and more upstream social, political, economic and security-related threats posed by AI. With the exception of some voices,9 10 the existing health literature examining the risks posed by AI focuses on those associated with the narrow application of AI in the health sector.11–16 20 This paper seeks to help fill this gap. It describes three threats associated with the potential misuse of narrow AI, before examining the potential existential threat of self-improving general-purpose AI, or artificial general intelligence (AGI) (figure 1). It then calls on the medical and public health community to deepen its understanding about the emerging power and transformational potential of AI and to involve itself in current policy debates on how the risks and threats of AI can be mitigated without losing the potential rewards and benefits of AI.

For the past several years, the U.S. Department of Defense has been exploring how it could employ its own very robust cyberattack capabilities to compromise or destroy enemy missiles from such states as North Korea before they can be fired, a strategy sometimes called “left of launch.”3 Russia and China can assume, on this basis, that their own launch facilities are being probed for such vulnerabilities, presumably leading them to adopt escalatory policies such as those espoused in the 2018 NPR report. Wherever one looks, therefore, the links between cyberwar and nuclear war are growing.

The Nuclear-Cyber Connection

These types of legal uncertainties are compounded by factual uncertainties in the cyber realm. While the United States appears to have an increasingly impressive ability to determine attribution, many other States lack that sophistication and are thus more likely to make costly errors. State D may mistake a cyber operation launched by a rogue hacker or organized non-State group operating out of country Y as being perpetrated by State Y itself. And governmental or non-governmental cyber hackers in a third country may very purposefully make it look like State Y conducted a cyberattack. Another difficulty in the cyber realm is determining whether certain effects of a hostile cyber operation were intended by the attacker. For example, State D may detect an imminent threat of malware in its systems that appears would at least temporarily compromise its most sensitive military arsenal—but is that what the perpetrator intended (and how, as a legal issue, should that question of intentionality matter)?

In cyberspace, many of these actions and interactions will take place at greater speeds thanks, in part, to artificial intelligence. These conditions may shrink the window of time for political and military leaders to make decisions, and place pressure on them to pre-authorize or create automated responses. It is not hard to imagine tit-for-tat uses of force quickly ascending a ladder of escalation. And there is no legal reason the rungs of that ladder will remain confined to the cyber realm.

Cybercriminals today have many sophisticated AI-driven cyberattacks methods by which they can create problems for the government, organizations, businesses, and individuals. Existing cybersecurity tools are no longer viable against this advanced cyber-weaponry (Thanh and Zelinka Citation2019). The negative use of AI to compromise digital security is known as AI-driven cyberattack, in which cybercriminals can train robots to socially engineer targets at human or superhuman levels of performance (Brundage et al. Citation2018). AI-assisted attacks will be able to adapt to the environment it infects. Learning from contextual data to emulate trustworthy features of cyberspace or target weak points it discovers (DarkTrace, 2020). AI-driven cyberattacks are not a far-fetched future threat. The necessary tools and building blocks for launching an offensive AI-driven cyberattack already exist (Dixon and Eagan Citation2019). Recent advancements in AI have influenced the tremendous growth in automation and innovation. Although these AI systems have many benefits, they can also be used maliciously. AI-driven cyberattacks have been on the rise in recent years. Even as many organizations transit to more secure cyberspace, such as cloud storage services, their resources remain vulnerable to cyber criminals (Bocetta Citation2020). In general, AI-driven cyberattacks will only worsen, then it will be almost impossible for traditional cybersecurity tools to detect them. It is simply a question of machine efficiency vs. human effort. The complexity and size of this growing trend are submerging cybersecurity teams. In contrast, the advanced and qualified cybersecurity specialists necessary to counter this threat successfully are increasingly expensive and difficult to find. The consequences of these emerging AI-driven attack techniques could be life-threatening and highly destructive. These subtle attacks undermine trust in organizations by undermining data security and integrity, potentially resulting in systemic failures (Cabaj, Kotulski, Księżopolski & Mazurczyk, Citation2018; Hamadah & Aqel, Citation2020). An AI-driven cyberattack will harness a multitude of cyberspace and computer resources well beyond what a human could enlist, resulting in an attack that is faster, more unpredictable, more sophisticated than even the strongest cybersecurity team can respond against. (Bocetta Citation2020). As AI becomes a more powerful tool in the hands of malicious actors, cybersecurity researchers, practitioners, and governments will need to respond with more inventive solutions to successfully safeguard cyberspace from malicious actors (Hamadah & Aqel, Citation2020). AI-driven attacks employ sophisticated obfuscating algorithms and frequently change identifiable characteristics that could reach a level of adaptability that renders it virtually undetectable to both behavioral and signature-based antivirus tools (Babuta, Oswald & Janjeva, Citation2020). High-security risks and elusive attacks in benign carrier applications, such as DeepLocker, have demonstrated the intentional use of AI for negative objectives. Attackers are constantly improving and changing their attack strategy, with a particular focus on the use of AI-based techniques in the attack process, known as AI-driven cyberattacks, which can be used in collaboration with traditional cyberattack techniques to cause more significant damage in cyberspace while remaining undetected (Kaloudi and Li Citation2020; Thanh and Zelinka Citation2019; Usman et al. Citation2020).

As datasets and models become more complex, the energy needed to train and run AI models becomes enormous. This increase in energy use directly affects greenhouse gas emissions, aggravating climate change. According to OpenAI researchers, since 2012, the amount of computing power required to train cutting-edge AI models has doubled every 3.4 months. By 2040, it is expected that the emissions from the Information and Communications Technology (ICT) industry as a whole will reach 14% of the global emissions, with the majority of those emissions coming from the ICT infrastructure, particularly data centres and communication networks. These data demonstrate the urgent need to address AI’s carbon footprint and role in environmental deterioration.

Recently, a study was conducted by researchers at the University of Massachusetts to determine how much energy is used to train certain popular large AI models. According to the results, training can produce about 626,000 pounds of carbon dioxide, or the equivalent of around 300 round-trip flights between New York and San Francisco – nearly 5 times the lifetime emissions of the average car.

The power demands of the huge data centers that underpin the A.I. revolution keep growing. Wall Street is taking notice — but the climate effect isn’t getting as much attention.

The A.I. boom is supercharging markets’ interest in power. One sign of investor enthusiasm: The S&P 500’s utilities sector is up nearly 8 percent this year, outpacing the benchmark index overall.

Large language models (LLMs) have a dirty secret: they require vast amounts of energy to train and run. What’s more, it’s still a bit of a mystery exactly how big these models’ carbon footprints really are. AI startup Hugging Face believes it’s come up with a new, better way to calculate that more precisely, by estimating emissions produced during the model’s whole life cycle rather than just during training. It could be a step toward more realistic data from tech companies about the carbon footprint of their AI products at a time when experts are calling for the sector to do a better job of evaluating AI’s environmental impact. Hugging Face’s work is published in a non-peer-reviewed paper. To test its new approach, Hugging Face estimated the overall emissions for its own large language model, BLOOM, which was launched earlier this year. It was a process that involved adding up lots of different numbers: the amount of energy used to train the model on a supercomputer, the energy needed to manufacture the supercomputer’s hardware and maintain its computing infrastructure, and the energy used to run BLOOM once it had been deployed. The researchers calculated that final part using a software tool called CodeCarbon, which tracked the carbon dioxide emissions BLOOM was producing in real time over a period of 18 days. Hugging Face estimated that BLOOM’s training led to 25 metric tons of carbon dioxide emissions. But, the researchers found, that figure doubled when they took into account the emissions produced by the manufacturing of the computer equipment used for training, the broader computing infrastructure, and the energy required to actually run BLOOM once it was trained. While that may seem like a lot for one model—50 metric tons of carbon dioxide emissions is the equivalent of around 60 flights between London and New York—it's significantly less than the emissions associated with other LLMs of the same size. This is because BLOOM was trained on a French supercomputer that is mostly powered by nuclear energy, which doesn’t produce carbon dioxide emissions. Models trained in China, Australia, or some parts of the US, which have energy grids that rely more on fossil fuels, are likely to be more polluting. After BLOOM was launched, Hugging Face estimated that using the model emitted around 19 kilograms of carbon dioxide per day, which is similar to the emissions produced by driving around 54 miles in an average new car. By way of comparison, OpenAI’s GPT-3 and Meta’s OPT were estimated to emit more than 500 and 75 metric tons of carbon dioxide, respectively, during training. GPT-3’s vast emissions can be partly explained by the fact that it was trained on older, less efficient hardware. But it is hard to say what the figures are for certain; there is no standardized way to measure carbon dioxide emissions, and these figures are based on external estimates or, in Meta’s case, limited data the company released. “Our goal was to go above and beyond just the carbon dioxide emissions of the electricity consumed during training and to account for a larger part of the life cycle in order to help the AI community get a better idea of the their impact on the environment and how we could begin to reduce it,” says Sasha Luccioni, a researcher at Hugging Face and the paper’s lead author. Hugging Face’s paper sets a new standard for organizations that develop AI models, says Emma Strubell, an assistant professor in the school of computer science at Carnegie Mellon University, who wrote a seminal paper on AI’s impact on the climate in 2019. She was not involved in this new research. The paper “represents the most thorough, honest, and knowledgeable analysis of the carbon footprint of a large ML model to date as far as I am aware, going into much more detail … than any other paper [or] report that I know of,” says Strubell. The paper also provides some much-needed clarity on just how enormous the carbon footprint of large language models really is, says Lynn Kaack, an assistant professor of computer science and public policy at the Hertie School in Berlin, who was also not involved in Hugging Face’s research. She says she was surprised to see just how big the numbers around life-cycle emissions are, but that still more work needs to be done to understand the environmental impact of large language models in the real world. "That’s much, much harder to estimate. That’s why often that part just gets overlooked,” says Kaack, who co-wrote a paper published in Nature last summer proposing a way to measure the knock-on emissions caused by AI systems. For example, recommendation and advertising algorithms are often used in advertising, which in turn drives people to buy more things, which causes more carbon dioxide emissions. It’s also important to understand how AI models are used, Kaack says. A lot of companies, such as Google and Meta, use AI models to do things like classify user comments or recommend content. These actions use very little power but can happen a billion times a day. That adds up. It’s estimated that the global tech sector accounts for 1.8% to 3.9% of global greenhouse-gas emissions. Although only a fraction of those emissions are caused by AI and machine learning, AI’s carbon footprint is still very high for a single field within tech. With a better understanding of just how much energy AI systems consume, companies and developers can make choices about the trade-offs they are willing to make between pollution and costs, Luccioni says. The paper’s authors hope that companies and researchers will be able to consider how they can develop large language models in a way that limits their carbon footprint, says Sylvain Viguier, who coauthored Hugging Face’s paper on emissions and is the director of applications at Graphcore, a semiconductor company. It might also encourage people to shift toward more efficient ways of doing AI research, such as fine-tuning existing models instead of pushing for models that are even bigger, says Luccioni. The paper’s findings are a “wake-up call to the people who are using that kind of model, which are often big tech companies,” says David Rolnick, an assistant professor in the school of computer science at McGill University and at Mila, the Quebec AI Institute. He is one of the coauthors of the paper with Kaack and was not involved in Hugging Face’s research. “The impacts of AI are not inevitable. They’re a result of the choices that we make about how we use these algorithms as well as what algorithms to use,” Rolnick says.

Each time you use AI to generate an image, write an email, or ask a chatbot a question, it comes at a cost to the planet. In fact, generating an image using a powerful AI model takes as much energy as fully charging your smartphone, according to a new study by researchers at the AI startup Hugging Face and Carnegie Mellon University. However, they found that using an AI model to generate text is significantly less energy-intensive. Creating text 1,000 times only uses as much energy as 16% of a full smartphone charge. Their work, which is yet to be peer reviewed, shows that while training massive AI models is incredibly energy intensive, it’s only one part of the puzzle. Most of their carbon footprint comes from their actual use. The study is the first time researchers have calculated the carbon emissions caused by using an AI model for different tasks, says Sasha Luccioni, an AI researcher at Hugging Face who led the work. She hopes understanding these emissions could help us make informed decisions about how to use AI in a more planet-friendly way. Luccioni and her team looked at the emissions associated with 10 popular AI tasks on the Hugging Face platform, such as question answering, text generation, image classification, captioning, and image generation. They ran the experiments on 88 different models. For each of the tasks, such as text generation, Luccioni ran 1,000 prompts, and measured the energy used with a tool she developed called Code Carbon. Code Carbon makes these calculations by looking at the energy the computer consumes while running the model. The team also calculated the emissions generated by doing these tasks using eight generative models, which were trained to do different tasks. Generating images was by far the most energy- and carbon-intensive AI-based task. Generating 1,000 images with a powerful AI model, such as Stable Diffusion XL, is responsible for roughly as much carbon dioxide as driving the equivalent of 4.1 miles in an average gasoline-powered car. In contrast, the least carbon-intensive text generation model they examined was responsible for as much CO2 as driving 0.0006 miles in a similar vehicle. Stability AI, the company behind Stable Diffusion XL, did not respond to a request for comment. The study provides useful insights into AI’s carbon footprint by offering concrete numbers and reveals some worrying upward trends, says Lynn Kaack, an assistant professor of computer science and public policy at the Hertie School in Germany, where she leads work on AI and climate change. She was not involved in the research. These emissions add up quickly. The generative-AI boom has led big tech companies to integrate powerful AI models into many different products, from email to word processing. These generative AI models are now used millions if not billions of times every single day. The team found that using large generative models to create outputs was far more energy intensive than using smaller AI models tailored for specific tasks. For example, using a generative model to classify movie reviews according to whether they are positive or negative consumes around 30 times more energy than using a fine-tuned model created specifically for that task, Luccioni says. The reason generative AI models use much more energy is that they are trying to do many things at once, such as generate, classify, and summarize text, instead of just one task, such as classification. Luccioni says she hopes the research will encourage people to be choosier about when they use generative AI and opt for more specialized, less carbon-intensive models where possible. “If you’re doing a specific application, like searching through email … do you really need these big models that are capable of anything? I would say no,” Luccioni says. The energy consumption associated with using AI tools has been a missing piece in understanding their true carbon footprint, says Jesse Dodge, a research scientist at the Allen Institute for AI, who was not part of the study. Comparing the carbon emissions from newer, larger generative models and older AI models is also important, Dodge adds. “It highlights this idea that the new wave of AI systems are much more carbon intensive than what we had even two or five years ago,” he says. Google once estimated that an average online search used 0.3 watt-hours of electricity, equivalent to driving 0.0003 miles in a car. Today, that number is likely much higher, because Google has integrated generative AI models into its search, says Vijay Gadepally, a research scientist at the MIT Lincoln lab, who did not participate in the research. Not only did the researchers find emissions for each task to be much higher than they expected, but they discovered that the day-to-day emissions associated with using AI far exceeded the emissions from training large models. Luccioni tested different versions of Hugging Face’s multilingual AI model BLOOM to see how many uses would be needed to overtake training costs. It took over 590 million uses to reach the carbon cost of training its biggest model. For very popular models, such as ChatGPT, it could take just a couple of weeks for such a model’s usage emissions to exceed its training emissions, Luccioni says. This is because large AI models get trained just once, but then they can be used billions of times. According to some estimates, popular models such as ChatGPT have up to 10 million users a day, many of whom prompt the model more than once. Studies like these make the energy consumption and emissions related to AI more tangible and help raise awareness that there is a carbon footprint associated with using AI, says Gadepally, adding, “I would love it if this became something that consumers started to ask about.” Dodge says he hopes studies like this will help us to hold companies more accountable about their energy usage and emissions. “The responsibility here lies with a company that is creating the models and is earning a profit off of them,” he says.

WASHINGTON – Today, partners in the Climate Action Against Disinformation coalition released a report that maps the risks that artificial intelligence poses to the climate crisis. Topline points: AI systems require an enormous amount of energy and water, and consumption is expanding quickly. Estimates suggest a doubling in 5-10 years. Generative AI has the potential to turbocharge climate disinformation, including climate change-related deepfakes, ahead of a historic election year where climate policy will be central to the debate. The current AI policy landscape reveals a concerning lack of regulation on the federal level, with minor progress made at the state level, relying on voluntary, opaque and unenforceable pledges to pause development, or provide safety with its products. “AI companies spread hype that they might save the planet, but currently they are doing just the opposite,” said Michael Khoo, Climate Disinformation Program Director at Friends of the Earth. “AI companies risk turbocharging climate disinformation, and their energy use is causing a dangerous increase to overall US consumption, with a corresponding increase of carbon emissions.” “We are already seeing how generative AI is being weaponized to spin up climate disinformation or copy legitimate news sites to siphon off advertising revenue”, said Sarah Kay Wiley, Director of Policy at Check My Ads, “Adtech companies are woefully unprepared to deal with Generative AI and the opaque nature of the digital advertising industry means advertisers are not in control of where their ad dollars are going. Regulation is needed to help build transparency and accountability to ensure advertisers are able to decide whether to support AI generated content.” “The evidence is clear: the production of AI is having a negative impact on the climate. The responsibility to address those impacts lie with the companies producing and releasing AI at a breakneck speed,” said Nicole Sugerman, Campaign Manager at Kairos Fellowship. “We must not allow another ‘move fast and break things’ era in tech; we’ve already seen how the rapid, unregulated growth of social media platforms led to previously unimaginable levels of online and offline harm and violence. We can get it right this time, with regulation of AI companies that can protect our futures and the future of the planet.” “The climate emergency cannot be confronted while online public & political discourse is polluted by fear, hate, confusion and conspiracy,” said Oliver Hayes, Head of Policy & Campaigns at Global Action Plan. “AI is supercharging these problems, making misinformation cheaper and easier to produce and share than ever before. In a year when 2 billion people are heading to the polls, this represents an existential threat to climate action. We should stop looking at AI through the “benefit-only” analysis and recognise that, in order to secure robust democracies and equitable climate policy, we must rein in big tech and regulate AI.” “The skyrocketing use of electricity and water, combined with its ability to rapidly spread disinformation, makes AI one of the greatest emerging climate threat-multipliers, said Charlie Cray, Senior Strategist at Greenpeace USA, “Governments and companies must stop pretending that increasing equipment efficiencies and directing AI tools towards weather disaster responses are enough to mitigate AI’s contribution to the climate emergency.” Previously, the coalition submitted letters to President Biden and Senator Chuck Schumer that call on them to implement climate concerns into proposed AI legislation. The letters echo recommendations made in the report, including: Transparency: Companies must publicly report on energy usage and emissions produced, assess any environmental justice concerns related to developing AI technology and disclose how their AI models produce information in a way that prioritizes climate science. Safety: Companies must be able to publicly demonstrate their products are safe for users and the environment. In addition, governments should develop standards on AI safety reporting and invest in research that maps the risks AI poses to the spread of climate disinformation. Accountability: Governments should enforce rules on investigating and mitigating the climate impacts of AI with clear, strong penalties for noncompliance. Companies and their executives must be held accountable for any harms that occur as a result of their products.

Artificial Intelligence (AI) has exploded into our lives and workplaces in the last 12 months, but as you ask ChatGPT (or any of its competitors) to run a few work tasks for you, or when you get prompted by one of the built-in AI that are becoming ubiquitous across digital platforms, have you stopped to think of the ecological and social impacts of the rush to use this new technology? Last week, the founder of Open AI (the platform behind ChaptGBT) Sam Altman went public with his bid to raise 7 trillion dollars (yes that’s a T not a B!) to ramp up computer chip manufacturing so that the computational power of AI can expand more rapidly. This article is about the impact that AI learning models have on the environment and how this proposed expansion of AI would dramatically increase impacts, so read on to find out more. Just like mining Bitcoin, the creation of AI models requires a significant amount of computational power to get to the point where you are provided with quick (but not always accurate) responses. Although measuring the exact impact is pretty hard (as it is with mining Bitcoin), according to Statista, “The amount of data today’s digitised economy creates is growing annually by 40% and is expected to reach 163 trillion gigabytes by 2025, which will further fuel growth in AI.” Before we get into the details of where and how Generative AI is having impacts on the planet and people, let’s start with the overarching concept of measuring the size of an impact on the environment. One effective way to do this is through what’s called an ecological footprint. What is an Eco Footprint? Ecological footprints are an account of the demand for and the supply of nature’s biocapacity as a result of human activities. The “footprint” is the collective size of the impact, measured against a series of different data sets such as land, water, natural resources and energy use. The approach was developed and is promoted by the Global Footprint Network in the 1990s. The methods used to measure the impacts are designed to gauge the human demand on natural capital (which is the stock of all of nature’s resources), by looking at the amount of natural resources needed to support a country, region, product or the entire planet. We currently use the resources of 1.75 planets to sustain the consumption demands of humanity, so that means we exceed the capacity for the Earth to regenerate itself well before the end of each year. Since the late 1970s, we have been using up resources at a faster rate than the Earth can replenish, which is the concept behind Earth Overshoot Day. Based on the science, this demonstrates what day each year we use up all of the resources that Earth could provide us without having a negative impact. In 2023, it fell on August 2nd, which is not surprising considering that the Ecological Footprint accounts show that humanity collectively demands 75% more from our planet than its ecosystems can regenerate. Essentially, an ecological footprint explains how much nature we use and how much we have to use if we were to be sustainable, accounting for the deficit between these two. And there are spin-offs like water and carbon footprinting methods too, which look at the specific impact on these impact categories. Whilst there is some controversy around carbon footprints (specifically in relation to personal rather than product-level ones), the overarching ecological footprint method has a solid scientific foundation and is a valid mode of understanding generalized impacts. “The Environmental Footprint methods measure and communicate about the environmental performance of products (both goods and services) and organisations across their whole lifecycle, relying on scientifically-sound assessment methods agreed at international level.” — EU Commission Carbon footprints assess the amount of carbon-producing energy used to do something. In fact, it’s not just carbon emissions — there are several greenhouse gases that are measured for their global warming potential. The global standard for measuring carbon impacts is through the Greenhouse Gas Protocol. Given that Generative AI is so new, the science is still catching up to measuring it, but as we enter into a carbon-constrained economy as a result of climate change, the AI sector will have to be more accountable for its impacts because, as you will see below, they are pretty significant. The Carbon Footprint of AI MIT Technology Review recently published an article, stating, “There’s one thing people aren’t talking enough about, and that’s the carbon footprint of AI. One part of the reason is that big tech companies don’t share the carbon footprint of training and using their massive models, and we don’t have standardised ways of measuring the emissions AI is responsible for. And while we know training AI models is highly polluting, the emissions attributable to using AI have been a missing piece so far. That is, until now.” So, AI is just the newest contributor to our resource-hungry society that often ignores the invisible impacts associated with delivering the goods and services that end up in our lives (all of which are interconnected parts of the systems failures that I advocate adopting a systems mindset for in order to solve complex problems!). The tech sector is a significant contributor to climate change, on par with the airline industry. The full scope of AI’s specific impact is still unknown, but we can cobble together a picture of what’s going on and where the industry needs to make changes to avoid AI being an ecological nightmare. A Hungry Growing Machine Needs to be Fed There are many things that go into ensuring you can get a result from your Generative AI search: Devices and sensors that are used to capture the data (which comes from a rapid “crawl” across global, online open text and subsequent extraction of the most relevant digitized text in can access to craft its output) Large banks of servers that power and store the networks for communicating the data (these have to be made from mined minerals and powered by vast amounts of energy, and then, since they generate a lot of heat, cooled with water in cooling towers) The computational power needed to train the algorithm to start with (not to mention all the resources needed to set up the AI companies, etc.) People who helped to train the datasets (it’s actually surprising how much human there is in AI!) Your internet-connected device needed to conduct the search (and all the internet network cables and infrastructure required to enable you to connect, power your inputs, etc.) At every step of the data collection and transference life cycle, there are many natural materials required to create the physical assets. These are manufactured and made into usable goods, with lots of energy needed to power and cool the devices. All these assets need to be mined, processed, produced, distributed, sold, installed, used and then end up as waste when no longer functional. There are countless ecological and social impacts across this value chain, from the conflicts and conditions of the workers mining the valuable materials that go into making chips to the pollution caused from the processing of these materials. There are also reports of underpaid, overexploited workers being used to support the responses in Generative AI systems and significant issues with bias embedded in the training methods used. But by far, the biggest ecological bit is the amount of energy required to train and power the algorithms. Just as a growing child needs more food inputs to power the transformations in their body, so does the growing learning system of artificial intelligence. OpenAI, the creators behind the wildly popular ChatGPT, say themselves that the energy used to train the average AI model will increase 10-fold each year. Some believe machine learning is on track to consume all the energy that can be supplied. To get AI to be able to generate responses to your prompts requires a lot of training. Just like many everyday consumers are deceived into thinking there’s some magical process that whisks away their recycling and turns it into something new with no issues, something similar is happening with how we think we can just input a prompt into an AI platform and it magically creates content that passes for being written by a living, breathing human without any significant issues or impacts. Indeed, the training phase of AI is extremely energy-intensive, as the system is being fed human produced data that it is required to remember. It’s during the training phase that a growing AI must gobble up a lot of existing content (which is where all the IP issues stem from) and process the content into a bank so that it can respond to your requests. With advanced systems such as the newer versions of ChatGPT, this training is extensive and requires thousands of Graphics Processing Units (GPUs) and high processing chips, all working in harmony for days or months to get the AI trained up ready for commercial use. The more complex the system, the more energy the training requires. Once trained, the AI is then ready for inference (the process of running live data through a trained AI model to make predictions or solve tasks). This stage uses less energy than the training phase, but it’s still enough to generate a significant impact. Models do use less energy after they have been trained. Fascinatingly, AI can forget things it’s already learned as it learns new things! So, this may increase the impact, as it would need to be retrained on things it had previously known but forgot. The Graphic Processing Units (GPU) are the ones in hot demand for expanding the AI ecosystem and what the 7 Trillion dollars is for. Without these, the high energy computational processing for training can’t be done, and so this is why Sam Altman is trying to raise a lot of money to supercharge the chip manufacturing sector. Knowing that there are a lot of things (from nature) that need to go into making these tiny computer chips, there is concern about the astronomical increase in impact this would have. These chips are made from silicon, which is in hot demand for many different sectors. Interestingly, one of the major producers of these chips is using them to help model environmental disasters, but they themselves don’t currently report their environmental footprint. And using Graphic Processing Units are the most energy-hungry parts of the training phase. These demand a lot of electricity and specialized data center rack space to run them. “In rough terms, the average rack to support AI requires around the same amount of power as 25 houses.” Whilst some cloud computing companies have started to focus on reducing the footprint of their data centers by setting up facilities in climate-optimal areas (such as Iceland) and using more renewable energy, it still leaves a significant carbon footprint. A 2020 article in the scientific journal Nature said that it’s “estimated that the carbon footprint of training a single big language model is equal to around 300,000 kg of carbon dioxide emissions.” That’s the same as 125 round-trip flights between Beijing and New York. In training AI models, GPT-3 “resulted in 552 metric tons of carbon emissions, equivalent to driving a passenger vehicle for over 2 million kilometres. Nvidia, a chip giant supplying processors for training these AI models, plays a significant role in the carbon equation.” Despite the data we do have, many experts in the field of analyzing impacts of new technology say it’s still too early to know the full scope of the impact of AI. But what we do know is that it’s pretty bad and it’s likely to get worse unless there are significant changes to the way the industry works. The Guardian reported that, “By 2030, machine learning training and data storage could account for 3.5% of all global electricity consumption. Pre-AI revolution, datacentres used up 1% of all the world’s electricity demand in any given year,” whereas Swedish researcher Anders Andrae has forecast that data centers could account for 10% of total electricity use by 2025.

Recently, a study was conducted by researchers at the University of Massachusetts to determine how much energy is used to train certain popular large AI models. According to the results, training can produce about 626,000 pounds of carbon dioxide, or the equivalent of around 300 round-trip flights between New York and San Francisco – nearly 5 times the lifetime emissions of the average car.

The e-waste produced by AI technology poses a .  contains hazardous chemicals, including lead, mercury, and cadmium, that can contaminate soil and water supplies and endanger both human health and the environment. 

A group of industry leaders warned on Tuesday that the artificial intelligence technology they were building might one day pose an existential threat to humanity and should be considered a societal risk on a par with pandemics and nuclear wars.

“Mitigating the risk of extinction from A.I. should be a global priority alongside other societal-scale risks, such as pandemics and nuclear war,” reads a one-sentence statement released by the Center for AI Safety, a nonprofit organization. The open letter was signed by more than 350 executives, researchers and engineers working in A.I.

I’ve been researching artificial intelligence and machine learning for more than 30 years. Some people call me one of the “godfathers of AI.” And yet I have grown very concerned that this technology—like the atomic bomb—will grant humanity powers that could destabilize the world. If we do not create policy and guardrails around the use of AI, we could be looking at new and major threats to democracy and humanity.

Right now, AI-powered systems are advanced enough to both pilot drones and use high-powered facial recognition technology to target specific people. There have already been reports of such drone attacks in civilian areas in Syria and Ukraine. This is just the beginning. It is becoming shockingly easy, for example, to use AI-generated code to invent new toxic compounds—potential chemical weapons. Even more concerning to me are biological weapons, which can replicate themselves in the form of viruses or bacteria.

Many researchers steeped in these issues, including myself, expect that the most likely result of building a superhumanly smart AI, under anything remotely like the current circumstances, is that literally everyone on Earth will die. Not as in “maybe possibly some remote chance,” but as in “that is the obvious thing that would happen.” It’s not that you can’t, in principle, survive creating something much smarter than you; it’s that it would require precision and preparation and new scientific insights, and probably not having AI systems composed of giant inscrutable arrays of fractional numbers. Without that precision and preparation, the most likely outcome is AI that does not do what we want, and does not care for us nor for sentient life in general. That kind of caring is something that could in principle be imbued into an AI but we are not ready and do not currently know how. Absent that caring, we get “the AI does not love you, nor does it hate you, and you are made of atoms it can use for something else.” The likely result of humanity facing down an opposed superhuman intelligence is a total loss. Valid metaphors include “a 10-year-old trying to play chess against Stockfish 15”, “the 11th century trying to fight the 21st century,” and “Australopithecus trying to fight Homo sapiens“. To visualize a hostile superhuman AI, don’t imagine a lifeless book-smart thinker dwelling inside the internet and sending ill-intentioned emails. Visualize an entire alien civilization, thinking at millions of times human speeds, initially confined to computers— in a world of creatures that are, from its perspective, very stupid and very slow. A sufficiently intelligent AI won’t stay confined to computers for long. In today’s world you can email DNA strings to laboratories that will produce proteins on demand, allowing an AI initially confined to the internet to build artificial life forms or bootstrap straight to postbiological molecular manufacturing. If somebody builds a too-powerful AI, under present conditions, I expect that every single member of the human species and all biological life on Earth dies shortly thereafter. There’s no proposed plan for how we could do any such thing and survive. OpenAI’s openly declared intention is to make some future AI do our AI alignment homework. Just hearing that this is the plan ought to be enough to get any sensible person to panic. The other leading AI lab, DeepMind, has no plan at all. An aside: None of this danger depends on whether or not AIs are or can be conscious; it’s intrinsic to the notion of powerful cognitive systems that optimize hard and calculate outputs that meet sufficiently complicated outcome criteria. With that said, I’d be remiss in my moral duties as a human if I didn’t also mention that we have no idea how to determine whether AI systems are aware of themselves—since we have no idea how to decode anything that goes on in the giant inscrutable arrays—and therefore we may at some point inadvertently create digital minds which are truly conscious and ought to have rights and shouldn’t be owned. The rule that most people aware of these issues would have endorsed 50 years earlier, was that if an AI system can speak fluently and says it’s self-aware and demands human rights, that ought to be a hard stop on people just casually owning that AI and using it past that point. We already blew past that old line in the sand. And that was probably correct; I agree that current AIs are probably just imitating talk of selfawareness from their training data. But I mark that, with how little insight we have into these systems’ internals, we do not actually know. If that’s our state of ignorance for GPT-4, and GPT-5 is the same size of giant capability step as from GPT-3 to GPT-4, I think we’ll no longer be able to justifiably say “probably not self-aware” if we let people make GPT-5s. It’ll just be “I don’t know; nobody knows.” If you can’t be sure whether you’re creating a self-aware AI, this is alarming not just because of the moral implications of the “self-aware” part, but because being unsure means you have no idea what you are doing and that is dangerous and you should stop.