A new generation of commentators is using artificial intelligence to inform their predictions as polls indicate Donald Trump and Kamala Harris are tied for the U.S. presidential election on Tuesday, Nov. 5.
Some data science and machine learning traditions are incorporated into the new field of AI election prediction. How accurate their projections are will only become clear with time.
What are the Various Methods for Election Prediction?
Polls are a useful tool for gauging voter sentiment in the lead-up to an election, but they are a blunt instrument for predicting a race as close as Trump vs. Harris due to their margin of error.
People like Allan Jay Lichtman, the American historian dubbed the Nostradamus of American elections, have been driven by this inaccuracy to create methods based on observations of previous outcomes rather than on current polling data.
According to Lichtman’s model, which has accurately forecasted nine of the previous ten presidential elections, Americans’ voting behavior is influenced by variables including whether the US economy is in a recession or if the current administration is marred by controversy.
Despite being simpler, his method’s fundamental reasoning is similar to that of predictive AI, which is to make predictions from historical data.
Tencent Unveils Hunyuan-Large: An Open-Source 389 Billion Parameter AI Model Excels Over Llama 3.1-70B and 405B
Machine Learning
Modern AI models can analyze considerably more data to produce their forecasts than Lichtman’s system, which was based on thirteen “Keys to the White House.”
Tom Farnschläder, a data scientist, for instance, developed a model based on polls conducted eight months prior to the election in the preceding five cycles.
More advanced systems include the one developed by 24cast, which models more than 100 variables, from campaign financing and voting accessibility in each state to past election outcomes.
Although helpful, these quantitative methods are still restricted. They can be challenging to understand since they produce their predictions as probabilities.
For instance, Harris wins 70% of the time when 24cast’s model is run 100,000 times. But in 28% of simulations, the most likely scenario—in which Hillary wins every battleground state but Arizona—occurs.
In contrast, a Trump victory is the second most likely result, occurring in 7% of simulations.
BioMedGPT: First Open-Source AI Model Set to Transform Healthcare and Research
Simulations of Elections
The probabilistic results produced by 24cast mirror techniques are applied to predictive AI in domains like meteorology and quantitative finance.
However, a new method created by the firm Aaru provides the idea of election simulation with a completely different meaning.
Aaru runs hundreds of AI bots designed to mimic the personality features of voters using census data to duplicate voter districts. After that, it provides them with real-time news updates that are intended to resemble the media diets of the people they are imitating, and it tracks changes in their voting preferences.
Aaru creates a synthetic poll sample that is significantly larger and, in theory, far more accurate than any poll conducted in the real world by continually running the simulation.
According to Aaru’s most likely scenario, Donald Trump wins North Carolina, Arizona, and Georgia but loses Nevada and all of the northern swing states, which results in a Harris triumph.
However, it is important to note that the model only assigns Harris a 50.9% chance of winning Wisconsin, a state whose ten electoral college votes would be essential to victory in the scenario described above.
