Applications of generative artificial intelligence to influence climate change decisions
Ethical complexity
Previous research has highlighted the nefarious potential of GenAI in society2,3, and within the field of climate change decisions there has been a particular focus on the risks of large-scale dissemination of climate misinformation4,22. However, GenAI may be employed to influence climate change decisions with diverse perspectives and objectives, from the pro-environmental to outright denial perspectives, and from moderate to more radical actors on all sides28. These potential applications of GenAI span an ethical gradient, ranging from the normatively virtuous aim of providing objective information to support decisions5, to the illegal and unethical cases of blackmail and harassment24,25,29.
The role of GenAI in determining whether an activity is deemed ethical varies across uses. In cases such as blackmail, it is the act itself which is considered unethical and is illegal in most jurisdictions. Therefore, the use of GenAI to perform blackmail may be irrelevant from an ethical standpoint, but the technology may allow a larger number of cases to be attempted and successfully achieved3. Conversely, the production of personally-targeted advertising material is commonplace and tolerated by society. In this case, the empathic nature of GenAI, its unlimited patience, and ability to persuade people in an increasingly human-like way, are among the ethical concerns30. It will be critical to define legal and ethical boundaries for applications of GenAI to climate decision-making, to protect democracy and the integrity of climate change decision-making worldwide. These boundaries may vary across cultures and legal jurisdictions, and in developing such frameworks we must be cognisant of the ethical concerns of groups under-represented in policy and governance. It is crucial to consider the ethical concerns of Indigenous people, who will be disproportionately influential in developing solutions to climate change, have contrasting ethical frameworks that influence their perceptions of GenAI, yet are under-represented in developing GenAI regulation5,31.
Research needs
Many urgent research needs regarding nefarious and criminal applications of GenAI have been well-documented2,3,22,25,26, including research into technological and legislative approaches to apprehending specific intents24,27,29. For example, algorithms have been developed to detect synthetic media, and these have been incorporated into content verification systems, which may reduce the risk of deepfake content32. Instead of repeating these general discussions, we instead highlight two urgent research areas specific to applications of GenAI to influence climate decisions. First, we lack a basic understanding of the current state of climate debate and decision influence by GenAI. There has been no systematic attempt to quantify the key actors using GenAI to influence climate change decisions, and the volumes of material being created or actions undertaken. Furthermore, there has been little quantification of the success rates of GenAI influence on real climate change decisions. It is challenging to quantify activity in influencing decisions due to the intrinsically secretive nature of many activities, yet digital signatures of GenAI content exist and can be quantified2,25. In some cases, it may also be possible to trace financial signatures of GenAI use, in the form of payments made to commercial GenAI providers, or investment in the advanced computer hardware required to train and run models25,27.
Second, we need to better understand the potential of GenAI to influence different types of climate change decisions. In particular, we need to quantify the potential impact and likelihood of the various risks discussed above. Climate change is an intergenerational issue, with decisions made now having far reaching impacts5. It is also an issue in which coordinated and sustained efforts are required to be made by billions of people at a global scale5. At the system level, it is thus important to understand how increasing GenAI-based influence of information, debate, and decisions may have long-term implications. For example, there is the potential for greater volumes of misinformation and active persuasion to contribute to greater polarisation of opinion and radicalisation of action on both sides4. Alternatively, a greater availability of objective climate information may lead to consensus-building that balances the needs of varied stakeholders5. Projecting the impacts of GenAI on climate decisions will require integrated approaches that allow us to analyse these system-level outcomes, for example using network analyses, serious games, and agent-based models6.
Conclusions
Here we have outlined the potential for GenAI to influence decision processes regarding climate change, at scales ranging from the individual to the global. We outlined a typology of examples of such influence based on the current state of the technology, and discuss the deep ethical and legal complexity involved. We would like to reiterate that none of the methods, intents, or societal scales mentioned here are hypothetical or possible only in some distant future. With the current state of technology, every example here is possible, and each is being employed in criminal, political, or industrial realms at present25. Every example could immediately be deployed to influence climate change decisions at low cost and with little technical expertise. In the face of this disruptive technology, we require both technical research on specific use cases, and holistic thinking to analyse the broader implications and potential long-term impacts.
Responses