In this article, I’ll talk about some multi-agent ideas that fascinate me. I hope you find them interesting as well and maybe one day, someone will pick up on these ideas and make them a reality.

Let’s talk about the possibility of predicting the future by modelling the influence of each person’s decisions on the world.

👻 Fun fact: As a child, my summers at my grandparents’ house included a post-lunch tradition. We’d gather around the table for Turkish coffee, brewed in the traditional style that left a thick layer of grounds at the bottom of each cup. After finishing our drinks, my grandmother, a chemist and food engineer with a whimsical side, would pretend to read our fortunes from the coffee grounds. It was all in good fun; we knew she was making up stories, but her tales were always optimistic, filled with bright futures. Her fascination with the mystical was a playful contrast to her scientific background—much like how my generation turns to puppies on Instagram for entertainment, her generation embraced the mystical for their fun.

I’ve been fascinated by the possibility of using a tool to predict the future. Instead of relying on something whimsical like my grandmother’s coffee grounds, what if we harnessed the power of AI? Of course, I’m not suggesting that we analyze coffee grounds with a segmentation model—that would be far-fetched. But imagine if we could model the cumulative impact of individual decisions on a global scale. Could we potentially forecast future events and trends?

1. How do the decisions from one person influence the outcomes of another person?

Let’s consider an example to illustrate the potential of AI in predicting and preventing adverse scenarios:

Imagine two individuals, Person A and Person B.
Person A decides to drive after consuming alcohol one evening,
while Person B plans to visit In-N-Out for a burger.
Unfortunately, due to Person A's impaired driving, an accident occurs,
and Person B becomes involved because their path intersects with Person A's.

This situation could potentially be mitigated with the help of AI. For instance, an AI system could analyze factors such as Person A’s alcohol consumption and the timing of their decision to drive, and it could warn them of the high risk of causing an accident, advising against driving. Simultaneously, the AI could alert Person B about the increased risk of accidents along their route to In-N-Out, suggesting they delay their trip or choose a different destination.

This concept is related to the AI’s ability to simulate multiple future scenarios based on real-time data, evaluating the outcomes of various decisions to provide timely warnings.

However, implementing such a system raises significant privacy concerns. It would require continuous access to personal data and decision-making insights, a prospect that might be more palatable in a future where technologies like Neuralink have normalized the integration of AI in daily decision-making. While this could dramatically enhance safety and decision-making efficiency, it also opens up a complex debate about privacy, autonomy, and the extent to which we are willing to allow AI to influence our lives.

2. Voting and future of the world.

The process of voting has always captivated me, particularly how it seems we make such significant decisions—like choosing a representative—with relatively little information. So silly! While the media aims to inform us, their coverage can often be biased or even exacerbate misunderstandings. Additionally, voters come from diverse educational backgrounds and are influenced by various social factors such as friends and family.

Consider how frequently political candidates make campaign promises, only to pursue markedly different agendas once in office. This discrepancy between promises and actions leads to a possibility: what if we could use AI to predict the likely outcomes of a candidate’s tenure based on their past decisions, historical data, and broader national history?

Imagine an AI-driven outcome tree for each candidate. This model could predict, for instance, a 20 percent chance that a candidate might initiate a conflict, or a 30 percent likelihood of raising taxes. Such predictions could empower voters to make more informed decisions, choosing candidates not just based on promises but on probable outcomes.

We always say: history repeats itself. Given that history often shows patterns of repetition, leveraging this data to predict future political actions seems both logical and feasible. Modeling such a system would involve complex algorithms capable of analyzing vast datasets to identify trends and predict future events. This would undoubtedly be a challenging yet fascinating problem to tackle, potentially revolutionizing how we approach voting and political analysis.