Generative AI and its potential impact on environmental reporting/ journalism

Risks, opportunities and possible solutions

Environmental reporting and generative AI (Image by leonardo.ai)

(This newsletter post profited from input received from Anja Krieger, an environmental podcaster and communicator)

We are faced with several interrelated global environmental crises, such as global environmental change (including climate change and environmental pollution) and biodiversity decline. There is an urgent need for solutions in terms of restoration and conservation, and there is an increasing demand for sustainable transformation in various aspects of human activities. This all highlights the need for excellent environmental reporting and journalism, because we cannot rely just on scientists to fulfill part of this role of keeping the public informed. Environmental reporting is at the interface between science, policy and the public.

When it comes to text and language processing and generation, large language models (LLMs) are entering a wide variety of domains, wherever text can be important, and journalism is no exception. Some news outlets have disclosed that they use LLMs already  (LLMs increasingly used in journalism: Is artificial intelligence a threat to journalism or will the technology destroy itself? | Samantha Floreani | The Guardian). 

Given the different types of media outlets reporting on environmental topics, from more specialized to very general, there is a potential that outlets for which environmental reporting is not a main focus are particularly susceptible to inaccuracies and problems associated with LLMs.

Risks and opportunities of LLMs for the environmental sciences have previously been discussed (Rillig et al. 2023), and some of these will also map onto the domain of environmental reporting. In the following I discuss first the risks of LLMs in environmental reporting, then the opportunities, and some suggestions for solutions.

First, what are possible risks?

• LLMs can generate texts that are incorrect, biased, or misleading, but they will always appear authoritative and convincing. Environmental issues are often highly complex, results come with a degree of uncertainty and findings are often context-dependent, LLMs may not capture or convey this properly. For example, if a LLM is asked to extract and summarize information from various sources, it may omit or distort some important details or nuances.

• If incorrect reporting based on inappropriate LLM use becomes public, reputational harm for the media outlet may ensue. Such damage can come at a high price in terms of the public’s trust.

• Abuse by malicious actors is always a possibility. Such people or organizations could be: creating wrong or misleading information at scale with a certain purpose, for example to discredit environmental causes or concerns. When such actors use tex-to-image models or other models that use plain text input, there is additional potential for misinformation through the use of fake images or other media.

• Reporting on solutions is an important part of journalistic reporting. LLMs will likely favor mainstream views and/or the systemic status quo when it comes to solving environmental issues - the views of actors with high communication budgets and/or skills. 

• LLMs cannot generate new knowledge on the real world that they haven’t been trained on. This means they cannot fulfill an important goal of journalism. Environmental science and the debate around pros and cons of different theories, solutions or policies is always evolving. This needs human reporters who do interviews, reportage and in-depth investigation.

• Some LLMs might lack the ethics and human judgment needed to understand what to report on and what not, how to navigate fine lines of disclosure and privacy. 

• As in some parts of the tech sector and other industries, job losses may ensue caused by replacements with LLMs. Such a reduction in the workforce that could otherwise deliver quality environmental reporting would be worrisome and potentially dangerous.

The use of LLMs can also offer particular opportunities; what are they?

• LLMs can offer unprecedented opportunities to tailor texts (or images) for different target audiences and output formats, at least to propose texts suited for different groups of people, including different educational levels. 

• LLMs and other AI tools operate in many different languages, and thus there could be increasing support for multilingual services. This in turn may translate into better reporting at the regional level, in a variety of different regions of the world.

• LLMs can speed up text generation, freeing up time for other tasks such as those involved in investigative journalism

• LLMs could help generate engaging texts, and coupled with generative AI for images, also help with visual support

• LLMs can be the human-machine mediators when it comes to SEO (search engine optimization) helping human authors to easily optimize a text for searchability.

• LLMs could help summarize technical information from original sources, but this is tricky and comes with risks. Nevertheless, this is very likely to happen, and it may increase access to highly technical information

Finally, what are possible solutions (for all parties involved)?

• It will become paramount for the consumer of news and for the journalists themselves to verify accuracy of the text proposals generated, and it will be important to instill the use of LLMs as assistants to the journalist or writer

• There should be conventions on inserting warnings and disclaimers disclosing how LLMs were used in the generation of a news piece

• For all of us, it becomes increasingly important to support quality environmental reporting by actual human reporters in the field in whatever way possible

• For scientists: perhaps we should forge tighter relationships with environmental writers, and understand their ethics and processes, making their jobs as effective as possible

What do you think about the role of LLMs in environmental reporting? And, if you’re working in environmental communication, please share your take on this!