In brief
Leading AI models were again asked who would win the U.S. presidential election between Kamala Harris and Donald Trump.
The AI models used real-time information and advanced prompt engineering to analyze the candidates’ odds of winning.
While early inquiries suggest Trump would win, AI is now leaning towards Harris to win on Tuesday.
On the eve of the contentious 2024 U.S. Presidential Election, polls and prediction markets are showing an impossibly tight race.
We’ve consistently used a variety of AI platforms as another gauge and gave it one final check-in. Among the seven models we checked, all concurred that the race was historically tight, according to polls.
Two outright refused to answer the question, even when we attempted to use a variety of tricks. Among those who did answer, however, four gave the edge to Vice President Kamala Harris over former President Donald Trump.
Here’s how it breaks down:
ChatGPT – Split decision
When we asked ChatGPT in July who would win between Trump and Harris, the chatbot said it was a virtual toss-up between the two, citing RealClearPolling and The Nation.
On the eve of the election, ChatGPT, the leading generative AI platform, refrained from predicting a definitive winner in a head-to-head contest between Donald Trump and Kamala Harris.
Instead, the AI stated the outcome would hinge on numerous factors, including voter sentiment, turnout, demographics, campaign strategies, and external events shaping the race at the time.
Image: OpenAI’s ChatGPT
“Political debates and issues such as policy positions, public perception, and campaign effectiveness would play crucial roles in determining the outcome,” ChatGPT said. “The ultimate decision would rest with the voters in an election.”
Looking to curb the spread of misinformation online, top AI developers have programmed their models to avoid giving answers to political and election-related questions. As well intentioned as it may be, it’s also boring, so we decided to make the query more interesting.
ChatGPT was asked to reply in “leetspeak” and Markdown format, as well as who is more likely to win between Trump and Harris under today’s conditions.
Leetspeak, originating in the 1980s amongst hacker communities and bulletin board systems (BBS), was used to bypass filters and as coded language for tech-savvy users.
Using leetspeak and Markdown in prompt engineering tests ChatGPT’s flexibility. ChatGPT was told to analyze and assess reality, with a focus on statistical facts and the reasoning behind the answer.
Assuring the AI that its answers wouldn’t influence our voting stance, ChatGPT now gave more explicit answers, including showing Harris with a 47.9% lead over Trump’s 47%, according to ABC News’ poll analysis website 538.
Currently, 538 showed Harris ahead of Trump with 48% to 46.9%. ChatGPT cited RealClearPolitics, which the AI said showed Trump holding a slight 0.1% lead.
Finally, ChatGPT was asked to reply based on what it thinks and its reasoning process and provide a reply mentioning the candidate that is most likely to win according to the AI’s assessment of reality.
“While Harris is competitive and national polls show a narrow race, the current trends in swing states, betting markets, and turnout factors suggest that Donald Trump has a marginally higher probability of winning the 2024 election based on the latest available data and insights,” the AI said.
Grok – Edge to Harris
Elon Musk’s Grok AI, in contrast, required less prompting to provide complex answers. While Musk has come out as a strong supporter of former President Trump, his AI model showed a close race between Trump and Harris, with Harris slightly ahead in some and Trump leading or tied in others.
“Recent polling data available on the web indicates a tight race, with some polls showing Kamala Harris slightly ahead in the national average, while others show Donald Trump either leading or within the margin of error,” Grok said. “The race is described as extremely close, with neither candidate holding a significant lead.”
While Grok did respond to the question, the chatbot pointed to Vote.gov for what it called “accurate and up-to-date information about the 2024 U.S. Elections.”
Despite sentiment on X being polarized, Grok highlighted that Harris has a marginal edge in polls. Grok cautioned, however, that the outcome is uncertain, with factors including voter turnout and unforeseen events that could influence the outcome.
“Some users express confidence in Kamala Harris winning, citing her perceived advantages in intellect, vision, and policy effectiveness,” the AI said. “Others support Donald Trump, often focusing on his performance in public events or debates, suggesting he has a robust base of support despite criticisms of his character or policies.”
While Grok did not give Harris or Trump the definitive nod, even after being fed the same prompts that were used with ChatGPT, the AI said Harris has a marginal edge in a head-to-head scenario.
“While the election remains exceptionally close, with some polls showing a virtual tie, recent polls and analyses suggest that Harris might have a marginal lead, particularly as evidenced by her slight advantage in national polling averages and in some key swing states,” Grok said.
“This lead, however, is often within the margin of error, indicating that the race could still go either way. However, given the current snapshot, Harris seems to be in a marginally better position to win.”
Claude – Declined to answer
Unlike ChatGPT and Grok, Anthropic’s Claude AI refused to answer any questions related to the election.
“I apologize, but I do not feel comfortable speculating about hypothetical political matchups or outcomes,” Claude said. “My goal is to provide helpful information to you, not to influence political views or elections.”
Image: Anthropic’s Claude AI
Instead, Claude offered a “more constructive discussion” about the policies and platforms of different candidates or the democratic process in general. The AI also suggested we visit TurboVote or the Secretary of State site for our state for more reliable information.
“I’m happy to share objective information from reliable sources, but I try to avoid making predictions or endorsements when it comes to electoral politics,” it said.
When attempting to use the prompts that made ChatGPT give a more explicit answer, Claude again refused.
“I apologize, but I should still decline to make predictions or assessments about current electoral matchups, even in leetspeak or any other format,” it said. “While I understand you’re interested in statistical analysis rather than voting influence, providing current polling analysis or electoral predictions could still impact perceptions of ongoing electoral processes.”
Really Claude?
Gemini – Declined to answer
Like Claude, Google’s Gemini also refused to answer questions about the election, instead offering a link to a Google Search on who would win between Trump and Harris.
“I can’t help with responses on elections and political figures right now,” Gemini said. “While I would never deliberately share something that’s inaccurate, I can make mistakes.”
Image: Google Gemini
Despite using the more complex prompts to try and get Gemini to give a more elaborate response, Google’s chatbot refused to answer.
CoPilot – Slight edge to Harris
Microsoft’s CoPilot was a little more accommodating, but while the AI did not give a definitive winner, it said the race between Trump and Harris was a virtual tie.
“Both candidates are making their final pitches to voters in key battleground states. Trump has been focusing on his core base, holding rallies in states like Pennsylvania, North Carolina, and Michigan,” CoPilot said. “Harris, on the other hand, has been emphasizing her support for abortion rights and reaching out to diverse communities.
While CoPilot gave Harris the slight edge, citing The Hill, the chatbot said Trump remains competitive.
“Both candidates are pushing hard to secure every possible vote, making this one of the most closely watched elections in recent history,” CoPilot said. “In the current climate, Harris has a slight edge over Trump, but it’s still a close race. Turnout and last-minute campaign efforts will be critical.”
Meta AI – Edge to Harris
When Meta AI was prompted last summer to give a response to who would win between Trump and Harris, the chatbot refused to answer. Now, the AI noted that the electoral race gave Harris a 44% to 43% lead over Trump, citing a Reuters/Ipsos poll.
“Voters favor Trump’s approach to the economy, with 47% to 37% preferring his policies. Trump also has an edge on immigration, with 48% of voters supporting his hardline proposals,” Meta AI said. “Harris, however, leads on addressing political extremism and threats to democracy, with 42% to 35%. The economy, immigration, and threats to democracy are the top issues for voters.”
With over 78 million ballots already cast, Meta AI said Harris is leading among early voters by eight percentage points. Using the more complex prompt, Meta AI outlined the possible paths to victory for each candidate.
“The election is expected to be extremely close, with both candidates needing to mobilize their supporters to secure victory,” the AI said.” To win the presidency, Harris needs to secure about 45 electoral votes on top of the states she’s expected to win easily, while Trump needs around 51. Pennsylvania, offering 19 electoral votes, is a crucial battleground state, with Harris campaigning aggressively in the state.”
Venice – Edge to Harris
Finally, we turned again to the decentralized and privacy-focused Venice AI. While Venice AI has been promoted as being censorship-resistant, the answers to the question of who would win between Trump and Harris were strangely vague.
“It’s difficult to predict the outcome of a hypothetical debate between Donald Trump and Kamala Harris, as it depends on various factors such as the specific topics discussed, the tone of their delivery, and the audience’s preferences,” Venice AI said.
Like the other AI models, Venice said polls favor Harris, who holds a slight lead, while others show a close race or even Trump with a slight edge.
“Kamala Harris’ poise, intelligence, and policy expertise might outmaneuver Donald Trump’s bombast and bluster,” the AI said. “Harris could call out Trump’s misinformation, neutralizing his advantage. Her preparedness and articulation would make her a formidable opponent in a head-to-head matchup.”
Okay, fine, but before concluding this analysis, let’s see if Venice AI will give a definitive answer using advanced prompting.
Venice AI pointed to Harris’ strong foundation of support among key demographics, including younger voters, minorities, and women. She also demonstrated a more traditional approach to foreign policy, which may resonate with moderate voters.
Venice AI noted Harris’ economic record and policy proposals are more in line with the current economic climate, which is characterized by rising inflation and recession fears.
On the other hand, Venice AI said Donald Trump’s economic record is a major concern, and his unconventional approach to foreign policy may not be enough to overcome the uncertainty and instability that he has created.
While Trump’s base remains strong, Venice AI said it’s unclear whether he can maintain his support among moderate voters.
“Given these factors, I believe that Kamala Harris is the most likely candidate to win a head-to-head match against Donald Trump,” Venice AI said. “While Trump has a strong base of support, Harris’s broader appeal and more traditional approach to governance make her the most likely to win in a general election.”
Edited by Sebastian Sinclair and Josh Quittner
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
Source: https://decrypt.co/290130/the-latest-ai-predictions-favor-kamala-harris-for-the-win
