In 2004, James Surowiecky, working at The New Yorker as a business columnist, popularised the concept of ‘wisdom of crowds’ in his eponymously titled book. Since then, multiple studies explored the effect of having a large crowd performing better than either individuals or smaller groups.
One study in 2014, published in the Journal of Business Economics, arrived at 0.59% better annualised investment results from advice on the internet than from professional brokers and analysts. Stanford’s largest examination of the wisdom-of-crowds effect across half a million responses found that crowds perform better than constituent members by a large margin.
In other words, the crowd effect is even more noticeable when participants are quizzed on a series of multiple, related questions. Similar studies point to having diverse mindsets offsetting individual biases. Likewise, anonymity reduces error-generating social pressures, while showing consensus typically leads to ‘herding’ – decreasing accuracy.
One of the implications of the wisdom-of-crowds effect is the feedback loop potential. In essence, the more people Googling an asset, the more likely it is to predict market movement.
This also affirms prior empirical findings that heightened volatility attracts investor attention.
What does all of this mean?
Collective sentiment can offer genuine insight, but it is a double-edged sword. Just as market attention reflects information flow, it also amplifies volatility. In other words, the aggregate of individuals referred to as the “crowd” can spotlight trends before institutions react, but they also fuel bubbles before they burst.
Given that market timing is everything when it comes to profitable trades, could AI agents decipher market sentiment, complementing the wisdom-of-crowds effect in an automated manner?
The Problem of Deep Context for AI Agents
It is common to portray artificial intelligence advancement as a danger to humanity, borrowing heavily from science fiction. However, this portrayal fails to account for one simple fact, which makes it exceedingly unlikely for that scenario to ever transpire: As biological beings, humans have a continuous state of being as opposed to AI’s instanced state.
In turn, AI has a problem of contextual continuity, which is the capacity to sustain an embodied sense of experience that informs perception. Consequently, AI models lack the ability to accumulate subtle environmental cues that affect decision-making.
This is exceedingly important for forecasting because it heavily relies on such deep context. After all, how would one train AI on future events that haven’t transpired yet?
On the upside, it is a straightforward process to determine AI’s accuracy – either the prediction happened or it didn’t. Likewise, there is a steady stream of data from platforms like Polymarket to test against. Combined with reasoning derived from news, an AI could gain a sense of deep context applicable for forecasting.
For example, when the Federal Reserve makes an interest rate decision, an AI model would have a solid launchpad to follow its potential impact on the markets.
And just as there are AI benchmarks for AI doing math, there are going to be benchmarks to gauge AI’s predictive prowess, such as FutureBench. Most importantly, by doing predictions over and over again, the AI can improve, offsetting the problem of deep context.
What to Expect from AI Agents’ Integration
The Polytrader.ai platform was the first to launch an AI agent on Virtuals Protocol in late 2024. Tokenised as $POLY, the AI agent will sift through data across social media such as X, track news and other prediction markets to spot worthy trades.
The platform’s roadmap is to make AI agents perform these functions 24/7, even connecting the bot to users’ X account to receive updates and manage portfolios. For the platform, the underlying monetisation goes through the $POLY token, which the AI agent uses to generate sentiment analysis.
This token on Virtuals, however, should not be confused with upcoming $POLY for the largest prediction market platform – the Polygon-based Polymarket. Only last week, Polymarket’s Chief Marketing Officer Matthew Modabber, confirmed “there will be a token, there will be an airdrop” on the Degenz Live podcast.
At present, Polymarket is by far the largest prediction platform, having recently hit a new record just above $2 billion weekly trading volume according to Dune Analytics data. The rival Kalshi platform gained half as much weekly volume.
Cumulatively, Polymarket churned $21.4 billion in trading volume, greatly outpacing Kalshi at $13.7 billion, with a distant third, SX, at only $158 million, making Polymarket and CFTC-regulated Kalshi the centres of prediction markets.
Given the rapidly rising interest in online betting, it is safe to say that Polytrader’s AI-powered forecasting will have an equally strong demand in the DeFi arena when the platform gains more traction with token airdrops.
An interesting project that complements AI agent deployment is multi-chain PredX.ai. Unlike Polytrader, PredX allows users to predict influencers’ impact on markets and communities.
This means that users can effectively check the validity of the platform’s AI suggestions. As PredX’s AI sifts through news events to suggest event predictions – market trends, influencers’ impact, elections – users’ participation determine if they are worthy of betting.
This effectively challenges AI conclusions, helping curtail notorious AI confabulation.
The Bottom Line
A careful reader might have noticed that crowds are already akin to AI. Long before neural networks, markets turned millions of individual judgements into a single probabilistic output. In this light, AI agents are merely poised to formalise this process, effectively compressing the crowd’s intuition into code.
Combined with the existing crowd behavior across social media, we are looking at a seamless feedback loop of human sentiment and machine reasoning. In that sense, AI isn’t replacing the wisdom of crowds. Instead, it’s learning to become one.
Credit: Source link
