Art and CultureCOHA in EnglishGeopoliticsIndependenceMain 4 headlinesMediaPoliticsRegionsSecurity and Defense

Should we be concerned about what ChatGPT “thinks” about Latin America?

John Perry
Masaya, Nicaragua

ChatGPT is a powerful AI chatbot that is as easy to use as Google and provides more direct answers to users’ questions. Ask it anything you like, and you will receive an answer that sounds like it was written by a human, based on knowledge and writing skills gained from massive amounts of data from across the internet. Because of its growing popularity, there are already political questions about it, for example assertions that it has a left-wing bias or concerns about privacy issues, which have led to the bot being banned in Italy just this month. It is already banned in China and Russia.

A search on Google reveals little or no discussion about the relevance of ChatGPT to writing or research about Latin America. If asked whether there has been commentary about its interpretations of Latin American politics, ChatGPT is rather shy, saying “It is possible that individuals or organizations have commented on the accuracy or quality of my responses, but I do not have access to this information as I am a machine learning model and do not interact with individuals or organizations outside of responding to user queries.” Nevertheless, as we shall see, ChatGPT has a considerable database of material on the region, which means that it can produce analytical articles and other written material about its politics.

Can ChatGPT write articles like this? Yes, ChatGPT can write articles on a wide range of topics. It can generate text based on a given prompt or topic, and produce articles that are grammatically correct, coherent, and can pass as human-written text. However, the quality of the article will depend on the quality of the prompt and the level of specificity and clarity provided in the instructions. Additionally, ChatGPT may not have domain-specific knowledge or expertise, so the accuracy and depth of the content may vary depending on the complexity of the subject matter. Overall, ChatGPT’s ability to generate articles is impressive, but it should be noted that the output should be reviewed by a human to ensure accuracy and relevance.

Apart from the initial question, the previous paragraph was written entirely by ChatGPT itself. It makes the point that its output depends on being prompted by relevant questions: without prompting it and contesting its answers, ChatGPT gives responses that can be lacking in depth. Its factual accuracy is also below 100%, so responses must be viewed with skepticism and checked. And even this short paragraph shows that it often gives uncontroversial and rather bland replies.

From the viewpoint of COHA’s readers, a key question is whether ChatGPT provides reliable answers about Latin American politics, and whether it shows any bias in doing so. On simple factual questions, it generally gives reliable answers. For example, asked “what is the Council on Hemispheric Affairs?” ChatGPT gives an acceptable summary of the details on COHA’s website. Where it fails spectacularly is in giving reliable guidance to articles by particular authors. For example, asked what articles John Perry has written for COHA, it produced three titles that do not exist, including the interesting-sounding but as yet unwritten one, “Nicaragua’s Canal Project: Boon or Boondoggle?” Asking ChatGPT for articles by other authors on other topics produces equally plausible-sounding but often incorrect answers.

On political issues in Latin America, its answers become more interesting. For example, asked for a list of US interventions in Latin America since 1950, it produced the following:

  • Guatemalan Coup d’etat (1954): The CIA orchestrated a coup that overthrew the democratically elected government of President Jacobo Arbenz, who had nationalized land owned by the United Fruit Company.
  • Bay of Pigs Invasion (1961): The US attempted to overthrow the government of Fidel Castro in Cuba by invading the island with a CIA-trained Cuban exile force. The invasion was a failure.
  • Cuban Missile Crisis (1962): The US and Soviet Union engaged in a tense standoff over Soviet nuclear missiles in Cuba. The crisis was resolved after the US agreed not to invade Cuba and the Soviet Union removed the missiles.
  • Dominican Republic Intervention (1965-66): The US intervened in the Dominican Republic to prevent a communist takeover after the assassination of President Joaquín Balaguer
  • Chilean Coup d’etat (1973): The US supported the military coup that overthrew the democratically elected government of President Salvador Allende and installed General Augusto Pinochet as dictator.
  • Operation Condor (1975-1983): The US supported a campaign of political repression and state terrorism by right-wing dictatorships in Argentina, Chile, Brazil, Paraguay, Uruguay, and Bolivia.
  • Grenada Invasion (1983): The US invaded Grenada to overthrow the Marxist government of Prime Minister Maurice Bishop.
  • Nicaraguan Contras (1980s): The US provided financial and military support to the Contras, a group of anti-Sandinista rebels, in their efforts to overthrow the socialist government of Nicaragua.
  • Panama Invasion (1989): The US invaded Panama to overthrow the military dictator Manuel Noriega, who had been indicted on drug trafficking charges.
  • Haitian Intervention (1994): The US intervened in Haiti to restore democracy after a military coup overthrew the democratically elected government of President Jean-Bertrand Aristide.
  • Venezuelan Coup Attempt (2002): The US supported a failed coup attempt against the democratically elected government of President Hugo Chávez.
  • Honduran Coup d’etat (2009): The US supported a military coup that overthrew the democratically elected government of President Manuel Zelaya.

Typically, it adds this qualification: “It’s important to note that this list is not exhaustive and there have been numerous other US interventions in Latin America, including economic sanctions, covert operations, and political interference.”

Interestingly, it does not always shy away from controversy about US involvement. For example, asked for more detail on the coup in Honduras in 2009, it says that the “United States played a significant role in the military coup that overthrew the democratically elected government of President Manuel Zelaya in Honduras in 2009… evidence suggests that US officials were involved in supporting the coup and played a role in the decision to remove Zelaya from power.” It adds that “Some observers believe that the US government may have encouraged the military to remove Zelaya from power in order to prevent him from aligning too closely with leftist leaders in Venezuela and Nicaragua.”

Asked about the US role in the Venezuelan coup attempt in 2002, it refers to a “declassified CIA document” and points out that:

“In the days leading up to the coup, high-ranking US officials, including Secretary of State Colin Powell, met with opposition leaders in Venezuela and expressed concerns about Chávez’s leadership. Some observers believe that the US government may have encouraged the coup plotters and provided them with intelligence support. After the coup, the US government was quick to recognize the new government and congratulate the coup leaders on their success. However, the coup was short-lived, and Chávez was restored to power within two days, largely due to the efforts of his supporters and loyal military units.”

COHA readers might well see these statements as objective fact. However, the tendency of ChatGPT to reveal facts that are inconvenient to the US government and its allies may be one of the reasons that have led to accusations of political bias. “ChatGPT Goes Woke,” read the headline of a National Review article which argued that ChatGPT gives left-wing responses to questions about topics such as drag queens and the 2020 election.[1] ChatGPT has also been subjected to the “Political Compass Test” which found it to be biased towards the libertarian left.[2] A rather limited comparison between ChatGPT and Google’s rival bot, Bard, concluded that the latter was balanced but the former had a leftwards bias.[3]

However, probing ChatGPT in more detail on Latin American political issues shows a general tendency to strive for what conventionally passes for “balance.” For example, asked the same question about the US role in coup attempts, but this time in relation to Nicaragua in 2018, ChatGPT says “There was no evidence that the US played a direct role in the coup attempt in Nicaragua in 2018.” It has to be prompted to provide the information that “some opposition leaders in Nicaragua did meet with officials from the US embassy in Managua prior to the 2018 coup attempt… some critics argue that they may have included discussions of regime change or efforts to destabilize the government of President Daniel Ortega.”

Asked if it isn’t neglecting the role of US funding via bodies like the National Endowment for Democracy (NED) in the Nicaragua coup attempt, it concedes that there was NED funding and adds that “critics argue that the US government’s support for opposition groups and civil society organizations created a favorable environment for regime change efforts.” “However,” it adds, “the extent to which this support may have contributed to the political crisis in Nicaragua remains a matter of debate among analysts and policymakers.” Asked to list “grassroots human rights bodies” in Nicaragua, it gives five examples, all of bodies linked to the political opposition.

It also notes that Washington has imposed sanctions on Nicaragua, but asked whether they are legal or not ChatGPT equivocates, saying this depends on circumstances. Reminded that the UN special rapporteur specifically said that the sanctions imposed on Venezuela were illegal, ChatGPT agrees, although still insists that “Ultimately, the legality of sanctions depends on the specific circumstances and legal framework surrounding their imposition, as well as their impact on human rights and the affected population.”

On another controversial topic, the coup attempt during the presidential election in Bolivia in 2019, ChatGPT is more willing to see both sides. It does not dispute that there was an attempted coup, but asked about the calamitous intervention by the Organization of American States, which claimed irregularities in the voting, it aims for balance:

“The OAS audit was widely cited by opposition groups and other critics of President Evo Morales as evidence of electoral fraud, and helped to fuel the protests that ultimately led to Morales’ resignation. However, the OAS audit has also been criticized by Morales supporters and other analysts, who argue that the organization’s findings were flawed and influenced by political considerations.

“Some of the criticisms of the OAS audit include the fact that the organization did not conduct a full recount of the votes, and that it relied heavily on statistical analysis to reach its conclusions. Critics have also pointed out that the OAS has a history of political bias, particularly in its dealings with left-leaning governments in the region.

“Overall, the role of the OAS in the events surrounding the 2019 coup in Bolivia remains a subject of controversy and debate. While some argue that the organization played an important role in exposing electoral irregularities, others have criticized its methods and motives, and argued that its actions contributed to the destabilization of Bolivia’s democracy.”

However, on a different issue, asked whether Cuba has democratic elections, ChatGPT’s response would hardly trouble Washington:

“Cuba has a one-party socialist system, where the Communist Party of Cuba is the only legal political party. While Cuba does hold elections, they are not considered free and fair by international standards because opposition candidates are not allowed to run for office. Instead, candidates are selected by the Communist Party, and voters are presented with a single candidate for each office. Additionally, the government controls the media and restricts freedom of speech, which further limits the ability of citizens to express their political views. Therefore, Cuba’s electoral process is not considered democratic in the traditional sense of the term.”

Asked to compare the US and Cuban electoral system, ChatGPT’s response is conventional. Prompted on issues such as the influence of big money in the US system, and the limited differences between the two main parties it concedes that “voters feel like they do not have a real choice between candidates, and that the political system is not truly responsive to the needs and concerns of the general population.” However, the electoral system is nevertheless “a cornerstone of American democracy and a model for other countries around the world.”

Overall, it is difficult to sustain the argument that ChatGPT is biased in its treatment of Latin American political issues. Its answers, often bland and accompanied by caveats, veer between left-wing and right-wing interpretations, presumably depending on the volume and bias of material on the internet on particular issues, but if unprompted they are largely conventional. Because of this, and ChatGPT’s standard writing style, they are often boring. They can also be dated, since ChatGPT points out that its answers do not include the most recent sources. This may change of course, as the bot is developed further.

It has been argued that bots like ChatGPT threaten jobs, although the argument has been contested by the Marxist economist Michael Roberts.[4] Such bots are clearly also a challenge to teachers in assessing students’ work. But it is too soon to judge their effect on writing about Latin American politics. ChatGPT might replace some of the blander journalism about the region, and it may (like Google) be a useful research tool for writers. But at its current stage of development, it hardly seems likely to replace critical analysis of Latin American politics, regardless of one’s starting point on the political spectrum.

 

John Perry is a COHA Senior Research Fellow and  writer living in Masaya, Nicaragua.

[Main photo: Open license, protest in Ecuador, Flickr

 


Sources

[1] “ChatGPT Goes Woke,” https://www.nationalreview.com/corner/chatgpt-goes-woke/

[2] “The Bitter Truth Behind ChatGPT’s Political Correctness,” https://analyticsindiamag.com/the-bitter-truth-behind-chatgpts-political-correctness/

[3] “Comparing Google Bard with OpenAI’s ChatGPT on political bias, facts, and morality,” https://www.brookings.edu/blog/techtank/2023/03/23/comparing-google-bard-with-openais-chatgpt-on-political-bias-facts-and-morality/

[4] “AI-GPT: a game changer?,” https://thenextrecession.wordpress.com/2023/04/08/ai-gpt-a-game-changer/