Introduction

Public diplomats must often make decisions in the context of complex, multifaceted, ever-evolving environments. Navigating and making sense of complexities and uncertainty is one of the key missions of the diplomat. Mills (2021, p. 277) has even stated that “uncertainty lies at the heart of public diplomacy”.

Uncertainty lies in the process of collecting, processing and presenting information in public diplomacy. In 2003, for example, a United States-led coalition went to war with Iraq. Most of the rationale for the invasion originated in claims that Iraq had a weapons of mass destruction program that posed a threat to the United States and its allies. Intelligence published by both the American and British governments in 2002 included such claims but subsequent enquiries found no such weapons and no efforts to restart a nuclear program. An independent review conducted after the war in the UK concluded that, “although there was no deliberate distortion in the report, expressions of uncertainty in the intelligence, present in the original non-public assessments, were removed or not made clear enough in the public report” (Van Der Bles et al. 2019, p. 3). A US Senate Select Committee investigation reached a similar conclusion. This has led scholars of uncertainty to argue that “(t)he removal of considerable expressions of uncertainty from both documents had a dramatic effect on the opinions of the public and governments, and in the UK at least the removal of the uncertainties was considered key to paving the way to war” (Van Der Bles et al. 2019, p. 3).

New technologies have often helped to manage the complexities of communicating in public diplomacy, both when gathering and sharing information. They have, at the same time, increased the speed of public diplomacy activities and made the environment more unpredictable. Generative Artificial Intelligence is one of the latest technological advancements that is both creating excitement and concerns about its applications in public diplomacy.

In an article for Foreign Policy magazine, Andrew Moore, who is chief of staff to former Google CEO Eric Schmidt, articulates the multiple ways in which “AI Could Revolutionize Diplomacy” (2023). “(N)ot even diplomacy – one of the world’s oldest professions,” he writes, “can resist the tide of innovation.” He suggests that “(e)ven if diplomacy remains an art, it will increasingly rely on hard science.”

Moore positions AI as the better, faster, more secure mediator than the human diplomat, predicting that it can first be used as a tool for improving the work of diplomats and eventually “become an independent agent in international engagements”. Others are more pragmatic about the possibilities of generative AI for public diplomacy in two key areas (Manor 2023). The first is related to information gathering and reporting: using generative AI to gather and write information about the context in which diplomats work (for example, about a person, a group, an organization or institution of interest). In a recent meeting of the U.S. Advisory Commission on Public Diplomacy (2023), Alexander Hunt, the Public Affairs Officer of the U.S. Embassy in Guinea, for example, suggested that ChatGPT could be used as a prompt for speeches and to “build institutional memory” which could help to create a brief on the local context (para. 28). The second is related to external communication: using generative AI to write press releases, social media posts and images (Manor 2023).

While there has been widespread enthusiasm about some of the instrumental ways in which generative AI might support the practice of public diplomacy, Manor (2023, para. 7), for example, has warned that AI tools can often provide misleading, inaccurate or uncontextualized information that “may shape the opinions, beliefs and actions of its users” and that the spread of inaccurate information could, in turn, decrease confidence in institutions such as diplomacy. Jessica Brandt, Policy Director at the Brookings Institution, has highlighted the risks associated with generative AI in the propagation of fake news or automated commenting that would become more persuasive because of “the ability to personalize content” (para. 73).

Pundits and observers seem to be mostly concerned by the possible shortcomings arising from the use of AI by audiences and malicious actors rather than what the impact of embedding such technologies in diplomatic practices might herald. In this paper, we move away from practical concerns about AI to question whether AI’s “operationalization of truth” (Munn et al. 2023, p. 2) is compatible with how public diplomats manage the information environment. While there has been significant attention paid to generative AI’s inaccuracies, we argue that the more significant risk is to the ways in which generative AI systems result in a kind of “epistemic flatness” (Munn et al. 2023, p. 2) where all claims are positioned as equally certain. Actually, there is always an inherent and underlying variation in the uncertainty of claims produced by AI applications due to both aleatory uncertainty and epistemic uncertainty. Aleatory uncertainty (uncertainty in the model) refers to the fact that LLMs assemble phrases based on probabilities and statistical patterns learned from vast datasets of text (see Bender et al. 2021). Epistemic uncertainty (uncertainty in the world) is attributed to missing information or expertise (e.g. whether COVID-19 was airborne or not in the early stages of the pandemic).

We compare the epistemologies of generative AI systems with public diplomacy practice to argue that AI and public diplomacy represent two opposite and irreconcilable methods to treat uncertainty: as something to be smoothed over in the case of the former and something to be revealed, considered and, above all, negotiated in the case of the latter. In the paragraphs below, we set out what uncertainty is and how it is managed in AI systems, before we examine how uncertainty is managed in the practice of diplomacy.

Uncertainty in AI systems

Uncertainty is ever-present in the development and deployment of AI systems, as it is in all other areas of scientific and technological development. The sources of uncertainty in AI systems range from the quality and quantity of training data to uncertainties related to the design and parameters of the model. Despite the multiple sources of uncertainty in AI systems, the allure and authority of such systems is built on their apparent certainty.

Indeed, AI applications are built to avoid or smooth over uncertainty in order to fulfill their goal of being “useful” to the user (Munn et al. 2023, p. 7). Generative AI systems smooth over uncertainty in three key ways. First, they provide a single answer synthesized from multiple possible answers, reducing a vast multitude of possibilities to a single output. Amoore (2019) writes that “(i)t is this process of condensation and reduction to one from many that allows algorithmic decision systems to retain doubt within computation and yet to place the decision beyond doubt” (Amoore 2019, p. 154). The single output is then actionable by “the border guard, the security analyst, or the police officer, for example” (Amoore 2019, p. 154) who have little chance of “speak(ing) against the grain of the single output generated from millions of potential parameters” (Amoore 2019, p. 154). LLMs belie the uncertainty inherent in the process of their development. Human annotation is a key feature of big data processing, for example, where humans are used to perform analysis on small amounts of example data to create “a gold standard or ground truth” (Aroyo and Welty 2015, p. 16). But as Oroyo and Welty argue, the obvious disagreement that occurs among labelers when attributing meaning to text or images is treated as a flaw that “should be avoided or reduced” (Aroyo and Welty 2015, p. 16).

Second, LLM applications like ChatGPT fail to provide sources for their claims, and sometimes make up fictional sources. References can be requested from ChatGPT and Bing Chat generally provides citations, but ChatGPT has been known to provide fictional sources and Bing can provide references that aren’t accurately summarized in the response. Removing citations can give the appearance that there is consensus about a particular claim since scientific development sees the gradual removal of citations as consensus is reached (Latour 1987). Providing citations against specific claims can, alternatively, give the appearance of certainty by performing norms of academic citation.

Finally, the tone and form of generative AI systems give the appearance of the machine’s certainty. ChatGPT, for example, relays claims and arguments with a conviction and assuredness that belies the fact that it functions by merely predicting the most likely next word or phrase given the context of the prompt. The impressive results have been described as “fluent bullshit” by a number of commentators (Malik 2022; Vincent 2022). Munn, Magee, and Arora (2023), drawing from Frankfurt (2009) work, note that “rather than misrepresenting the truth like a liar, bullshitters are not interested in it; they subtly change the rules of dialogue so that truth and falsity are irrelevant”. Large Language Models look for statistical regularities to predict what words should come next in any given sentence. The result is authoritative because it mimics the style of a myriad texts.

Uncertainty in public diplomacy

Uncertainty is the bread and butter of diplomats and public diplomacy practitioners. In times of crisis in international relations, contextual uncertainty is a key focus of diplomatic debate—after the war in Iraq (Goodall et al. 2006), for example, or during the recent COVID-19 pandemic (Surowiec and Manor 2021). Uncertainty can also be encountered in key components of public diplomacy: in the design and implementation of policy objectives—as in the case of the “perpetual crisis” in the EU (Manfredi-Sánchez and Smith 2023) or measuring and assessing public diplomacy effectiveness (Sevin 2017).

Managing uncertainty in public diplomacy is a complex and nuanced process as public diplomacy involves shaping and influencing public perceptions, opinions, and attitudes in a constantly evolving environment. Public diplomacy practitioners manage uncertainty by being trustworthy, by promoting dialogic engagement and by conducting negotiation, a core activity of diplomacy.

A public diplomacy actor must be trustworthy in order to maintain authority in changing environments. As (Rolfe 2014, p. 79) confirms, “no matter how rational (logos) or emotional (pathos) a speech is, an audience will not listen if it lacks trust in the rhetorician”. Trust is principally generated by the positive reputation, which is to say credibility, of public diplomacy actors. Since credibility is about perception, it does not “reside in a source” (i.e. public diplomacy actors) but rather “is bestowed on a source by an audience” (Gass and Seiter 2009, p. 156). Lack of trust in a public diplomacy actor makes the communication environment even more uncertain and unstable.

Uncertainty in public diplomacy can arise from various sources, including complex dynamics in the communication process, in which cultural beliefs, dispositions, values (Zaharna and Arsenault 2023) and emotions (Graham 2014) are essential to understand. With the “war on terror” in the post 9/11 era it became clear that it wasn’t enough for public diplomats to broadcast information to overseas publics. The message and the sender have also to be perceived as trustworthy in order for the communication to be effective (Van Ham 2003; Zaharna 2003). Public diplomacy was reconceptualized as a diplomatic activity that aimed to build “trust and credibility” (Melissen 2005, p. 15) so that public diplomacy practices could be distinguished from propaganda. Managing uncertainty and openness to other interpretations of facts and reality are essential to the development of such trust.

The next important feature that supports public diplomats in dealing with uncertainty is being open to dialog and to different interpretations of reality. Dialogic forms of communication, and thus dialogic engagement “enables organizations and stakeholders to interact, fostering understanding, goodwill, and a shared view of reality” (Taylor and Kent 2014, p. 391). Listening is a critical element of dialogic communication and a key factor in building credibility because it supports the interpretation of the “situation-specific” and “culture-bound” elements that foster an audience’s trust (Gass and Seiter 2009, p. 157; Di Martino 2020).

Dialogic forms of communication do not imply self-negation but rather a constant negotiation of different attitudes to avoid conflict and enhance trust “with direction(s) and purpose(s)” (Bickford 1996, p. 146). Auer (2016, p. 128), for example, argues that “public diplomacy is a crisis management technique” and that much of the work done by public diplomats is about nurturing the “conditions that help prevent a crisis or crisis escalation” (p. 128). This is often achieved through negotiation. In a time of climate crisis, for example, it is important to negotiate scientific and policy disagreements on climate change. In this sense, there has been call for a “ ‘diplomatic’ approach to knowledge assessment” (Kouw and Petersen 2018, p. 52) or for the need to “apply diplomatic relationships to international scientific collaborations” (Höne 2022). These calls highlight the importance of both scientific and diplomatic deliberation to address the complex problem of climate change.

Public diplomacy practitioners must navigate a complex and ever-changing landscape, and effective management of uncertainty is essential to achieving their objectives in building positive relationships and influencing public opinion abroad. Managing uncertainty in the context of public diplomacy is not about smoothing over disagreement, but rather in meeting disagreement head on by listening to differing opinions, understanding the origins of disagreements and navigating the complexities of opposing demands.

Disagreement cannot be predicted with certainty. Nor can the most useful outcome for all parties. In his essay, Winham (1977, p. 97) argues that “negotiators tend not to estimate acceptable outcomes, because outcomes are distant and unknowable. They focus instead on the process of negotiation and what they want the process to achieve, such as exchange of information about both parties' principal concerns, decision-making procedures, or the like”. The introspective analysis on the process of producing an outcome is something that is inherently human. Thus, negotiation remains an essentially human activity because it involves the creation of “shared values” in its process of managing uncertainty (Winham 1977).

Epistemic frictions

It is inevitable that the embedding of generative AI in public diplomacy is attracting so much attention from both practitioners and scholars alike. Beyond the hype that often accompanies the introduction of new technology, in this case we are witnessing the introduction of a technology that can cause unpredictable epistemic frictions.

The discussion above has shown that managing uncertainty by highlighting rather than smoothing over disagreement is essential to the practice of public diplomacy. Negotiation, a key diplomatic practice, requires surfacing, highlighting, examining and holding in place what cannot be controlled. In contrast, generative AI systems like ChatGPT simulate authority by smoothing over doubt and uncertainties that might arise from disagreements, probabilities and temporalities to present a singular answer that synthetically melds the most popular claims and presents them as if from a god’s eye view.

While AI can support diplomacy in providing citizens with practical information about consular services or automate some aspects of the visa process, diplomats should be cautious in the use of AI when complexity must be embraced to achieve a strategic outcome. For example, AI tools may be able to provide a list of key decision makers to target in a public diplomacy effort to close a commercial deal between two countries. However, it might not be able to fully capture the assets available to diplomats, including members of the diasporic community in the country who might have important connections to key decision makers but who may not be well publicized as such.

The implications of this idea for public diplomacy include the need for AI literacy among public diplomacy practitioners. AI literacy is recognized as a set of competencies that enable people to critically evaluate AI technologies (Long and Magerko 2020, p. 2). In particular, public diplomacy practitioners require competencies in critically interpreting data, i.e. understanding “that data cannot be taken at face-value” and the ability to “recognize and describe how computers reason and make decisions” (Long and Magerko 2020, pp. 5–6). These capabilities would enable public diplomacy practitioners to understand how AI deals with uncertainty and how the resulting claims that they make might subsequently be interpreted.

Unlike generative AI systems, diplomats don’t have the luxury of simplification. The fate of nations can depend on the composition of a single fact, and the ways in which uncertainties are exposed, managed and dealt with by public diplomacy practitioners.