How To Spot Trends with AI
After Sarah DaVanzo and I leveraged NWO.ai’s invaluable AI to score and re-rank the Meta Trends, we were left stuck with one finding:
Both the global and U.S. AI data-driven ranks were significantly different from the original human rank. The AI declared that what we humans thought was most important was not actually the case.
Were we just splitting hairs of importance here, or were these divergent rankings a signal that our Meta Trends (which came from source material) were not as important as we once thought? Maybe more influential cultural shifts are out there waiting to be exposed.
And if so, how can we find them?
We debated important but missing Meta Trends for weeks — but, how important could these be if the experts couldn’t agree upon their importance by not collectively highlighting them within their reports which we analyzed? But simultaneously, according to our work analyzing the last five years of Meta Trends, the “most important trends” being reported haven’t changed much.
There was no denying, though: important, nuanced cultural shifts were missing from our list of 14 Meta Trends. So, how could we identify and highlight these overlooked trends… and further, in a way that isn’t subjective (Sarah’s opinion against mine)?
We considered just naming our favorite cultural phenomenon not included in the original Meta rank, or we could have just surfaced interesting leftover trends from the 40+ reports that didn’t make their way into one of the 14 Meta Trend themes, but both approaches would have thrown us into the same trap which we immediately called out after publishing the most recent annual Meta Trend report: the prevalence of bias and scarcity of risk in the trends and foresight field is concerning at best…
While Sarah and I both have historical proof and a pedigree of accurate trend forecasting, our life experiences and methods differ. Just listing our favorites felt too qualitative. So we designed another experiment with NWO.ai.
Experiment 04.
AI Meta Trend Identification
_ Comparison
Left on its own, could AI identify similar or different — perhaps missing — Meta Trends?
This time we fed all of the text from the original 40+ sourced trend reports into the NWO.ai AI platform. Nearly one million words of text. We figured that the AI could process this information with a different, extraordinary comprehension than us humans, who attempted to do the same when creating that original 2022 rank. We hypothesized the AI would make more connections — ergo identify Meta Trends completely overlooked by the humans.
By crunching all of the reports and instructing the AI to identify Meta Trend patterns (clusters, themes, etc.), would it come back with missing valuable, social shifts? Answer: Not even close.
We were very wrong to believe AI could complete this exercise similar to that of an expert trend spotter.
From the one million words of text inputted, the AI used Natural Language Processing (NLP) and clustered like-with-like, arriving at 72 clusters of “trends.”
Interestingly, there was very little overlap with our 14 Meta Trends — a handful at best, which were really just optimistic stretches.
Further, the AI’s clustered “trends” weren’t even trends, but rather general topics like “technology” and “pandemic.” It’s not to say that these themes weren’t impressive — they were — but these findings aren’t helpful to an experienced cultural strategist who can arrive at more provocative groupings.
So to answer the question:
Could AI identify overlooked Meta Trends: No.
But feeling we were onto something we asked a follow up…
Experiment 05.
AI Micro-Trend Identification
_ Extrapolation
Rather than identifying large patterns which we’d call Meta Trends, could we use the AI to identify and rank smaller, perhaps overlooked micro-trends from within the reports?
To figure this out, instead of having the AI merely organize the reports’ text, we instructed the AI to take its newly created meaning of the one million words (i.e. it’s 72 clusters) and use diverse internet data sources to measure and rank each and every signal. The goal of the experiment was to understand what the consumer energy is behind every micro-trend and rank them accordingly. We’d called these the AI-identified trends.
It was a complicated process, but essentially we asked the AI to take the signals it captured from the 40+ industry reports, use them as a launching off point, and then use all the available online information to easily rank and validate them.
The AI came back with 1,062 newly scored micro-trends.
This was a ranking of AI-identified, human-overlooked trends, via abstracted meanings and associations all from the 40+ reports’ text. It turned up gold.
Precisely, these were trends buried — or, hidden — within the industry reports that the AI pulled out using advanced NLP techniques and a vast amount of data.
Here is a curation of the top 40 ranked, overlooked micro-trends discovered by the AI:
Inclusive Insurance, Disruptive Winds, Food Inflation, DAO’s (Decentralized Autonomous Organizations), Rising Energy, Super Apps, Longevity Food, Unisex Fragrance, Dating Fatigue, Clothing Rental, AI-Music, Biodynamic Farming, Gender Affirmation, Rainwater Harvesting Systems, Wearable Robotics, Dopamine Dressing, Sleep Coaches, Financial Coaches, Gut Health, Caregiver Leave, Self-Hypnosis, Alcohol-Free Beer, Psychoactive Tea, Sperm Freezing, Touch-Free, Post traumatic, Carbon Pawprint, Sustainability Calculator, Land Stewards, Privacy Enhancing Tech, Anonymous Marketplace, Subscriptions, Fluid Fashion, Land Availability, Period Products, Paid Menstrual Leave, Mutual Aid, Workplace Conditions, Banned Advertising, and Media Anxiety
Perhaps most noteworthy: “Russia Initiative” was a buried “trend” identified by the AI from the structured text of the industry trend reports. The AI then used various online data sources to measure and score the energy behind this (and all of the other 1061 signals). While the AI picked up this shift, not a single report explicitly mentioned a pending war at the time of their writing. The AI was literally able to give voice to cultural change indirectly alluded to from within the human-authored reports.
Conclusion:
Humans for Sensemaking & AI For Discovery and Inspiration
Our experiments found that humans outperform AI in decoding the zeitgeist and defining cultural shifts at large (Meta Trends, Mega or Macro Trends). We’d call this “sensemaking.” This skill is essentially being able to synthesize wide-ranging, already structured data, and intuitively pattern match and creatively stitch narratives. Humans have an edge over AI when it comes to seeing the big picture and making non-obvious connections.
Humans can derive Meta Trend patterns. But as we found out, the AI cannot. Humans bring context to the table: historical knowledge, existing understanding of worthy trend criteria, and most important, ties to business use cases and priorities. Simply, we humans know what to look for. But… This is also our fatal flaw as it translates into bias…
Meanwhile, we uncovered AI has a distinct advantage when it comes to unifying, processing and analyzing diverse unstructured data sets at scale and with unmatched speed. AI beats humans when finding the most noteworthy weak and emerging signals (aka micro-trends) — concepts undetectable to the human eye due to the sheer volume of data. AI’s superpower in this context is “discovery” and “inspiration.”
We also learned AI works well when it deconstructs and analyzes both human-structured data (ex. our original Meta Trends) and massive troves of unstructured data, using them as source material or trailheads in its own search for novelty.
Ultimately, with insight from AI’s more precise rankings, its detection of signal vs. noise, and its delivered inspiration, it’s undeniable:
AI is a crucial fixture in a successful cultural intelligence system.
This series of experiments run by Sarah, NWO.ai and myself demonstrate the need and role of AI and cultural data at scale.
This is the future of cultural intelligence.
That’s the clearest takeaway here.
The optimal cultural intelligence system combines Humans-and-Machines in a series of orchestrated hand-offs, repeating the pattern of construction and deconstruction.
Humans are best utilized for sourcing and defining large, complex social themes, while AI is best utilized for prioritizing these weighty trends, sourcing micro-trends, and checking humans’ sometimes messy, qualitative approaches.
Nobody’s perfect.
And together is better than alone.
But one question remains: Are there other trends out there unreported by the industry’s published trend reports, unidentified by the Meta Trend analysis, and undiscovered by the AI.
Answer: No doubt.
Matt Klein is a cultural theorist and strategist analyzing the psychosocial implications of our technology to consult on business decisions.