QUICK SUMMARY
The session explores how automated behavioral science can quantify the unspoken emotional and non-conscious drivers of human behavior, which are essential for preventing AI model collapse and ensuring effective innovation. The speaker warns that as generative AI models are increasingly trained on their own outputs rather than fresh human data, they experience “model collapse” that leads to regression to the mean and loss of creativity. By continuously injecting AI models with real human behavioral data, especially non-conscious emotional insights, companies can accelerate innovation processes from weeks to days while maintaining creative differentiation and more accurately predicting market success.
KEY QUOTES
- “Automated behavioral science is going to give us causal evidence, scientific evidence of the drivers of behavior, not relying solely on what people say, but observing what they do and doing it within a controlled environment.”
- “The work that we do in understanding humans and getting deeper insight into the true drivers of behavior is going to become increasingly important as AI accelerates a regression to the mean, we’re the champions of the future.”
- “We have to continually inject these AI models with fresh human data, not to mention that human values and what we want tends to change over time.”
FULL SESSION SUMMARY
Understanding the Unspoken Layer of Human Behavior
The session begins by highlighting how human behavior is significantly driven by emotion, intuition, and implicit bias that operate below conscious awareness. The speaker emphasizes that nonverbal communication is often the most important form of communication, yet brands frequently lack real-time access to this “unspoken layer” of consumer insight. Automated behavioral science offers a solution by providing causal evidence of behavior drivers through controlled environments, quantifying attention, emotion, and non-conscious thought processes (referred to as System 1 and System 2 data). This approach is positioned as more accurate at predicting product success in the market compared to traditional methods that rely solely on what consumers say rather than observing what they actually do.
The Innovation Acceleration Promise
By combining automated behavioral science with emotion AI and predictive AI, companies can dramatically accelerate their innovation processes. The speaker cites Estee Lauder as an example, noting their innovation process was reduced from 4-6 weeks to just 2-3 days using these techniques. The innovation framework presented includes several key stages:
- Discovery – Understanding what is implicitly true, distinct, and valuable about a brand at the non-conscious level to identify promising extension categories
- Concept Screening – Rapidly evaluating hundreds of ideas based on emotional reactions through automated behavioral science
- New Product Development – Refining winning ideas with features, benefits, and positioning informed by behavioral insights
- Launch Optimization – Using accumulated insights to optimize messaging at market launch
The speaker provides real-world examples including Flowers Foods (Dave’s Killer Bread, Wonder Bread) identifying 31 potential extension categories, Doritos successfully launching “Doritos Loaded” at 7-Eleven, and Heineken effectively positioning its non-alcoholic “Zero Zero” beer.
The AI Model Collapse Problem
Despite the benefits of AI-accelerated innovation, the speaker warns of a significant challenge: model collapse. Referencing a 2024 paper in Nature by Shemi et al., the presentation explains how generative AI models trained on data increasingly generated by those same models experience degradation over time. This collapse happens in two forms:
- Early Stage Collapse – Models lose information about the “tails” of distribution where creativity often resides
- Late Stage Collapse – Successive model generations converge to point estimates with minimal variance
The speaker demonstrates this problem with examples showing how text degrades into nonsense over multiple generations of AI training on its own outputs. This regression to the mean is particularly problematic for innovation, which requires creativity and differentiation rather than averages.
Behavioral Science as AI’s Missing Link
The solution proposed is continuous injection of fresh human data, particularly from the non-conscious realm. The speaker cites research showing that adding just 10% real human data to training sets significantly reduces model uncertainty and maintains greater variance in outputs. Automated behavioral science is positioned as “AI’s missing link” because it can quantify non-conscious and emotional data that has significant influence on behavior but is typically missing from AI training data.
The session concludes by emphasizing that behavioral science also serves as a “natural bot killer,” eliminating fraudulent survey responses and ensuring data quality. The speaker advocates for responsible implementation through whole human representation (both conscious and non-conscious data), testing diverse audiences, using true implicit measures, eliminating bots, and conducting always-on testing to continually generate new learning.
KEY TAKEAWAYS
- Non-conscious data is critical for innovation – Emotion, intuition, and implicit bias drive human behavior below conscious awareness, making automated behavioral science essential for accurate consumer insights.
- AI models face regression without fresh human data – Without continuous injection of real human data, AI models trained on their own outputs experience “model collapse,” losing creativity and variance critical for innovation.
- Automated behavioral science dramatically accelerates innovation – Companies implementing these techniques can reduce innovation cycles from weeks to days while improving prediction accuracy for market success.
Delivery on Event Focus: Innovation and Business Strategy
This session directly addresses the event’s focus by demonstrating how automated behavioral science and emotion AI can transform innovation processes to be both faster and more effective. The presentation shows how these techniques align innovation with business strategy by reducing risk, accelerating time-to-market, and more accurately predicting consumer behavior. The examples from Flowers Foods, Doritos, and Heineken illustrate how companies can use these insights to make strategic decisions about brand extensions and new product development that align with their core brand values and market opportunities.
Delivery on Event Theme: Harvesting Innovation and Future Growth
The session supports the event theme by addressing both immediate innovation harvesting (accelerating current processes from weeks to days) and sowing seeds for future growth (preventing AI model collapse through continuous human data). The speaker positions behavioral science as essential for sustainable innovation, warning that without it, AI-driven innovation will increasingly regress to the mean, losing the creative differentiation necessary for future growth. By quantifying the unspoken aspects of human behavior, companies can maintain creative innovation capabilities even as AI becomes more prevalent.
Action Steps for Innovation Experts and Corporate Changemakers
- Implement automated behavioral science to capture non-conscious consumer insights alongside traditional research methods
- Establish continuous human data collection systems that regularly inject fresh behavioral and emotional insights into AI innovation processes
- Apply implicit testing methodologies to understand what’s truly distinct and valuable about your brand at a non-conscious level
- Integrate emotion AI and behavioral science throughout the innovation process from discovery to launch optimization
- Develop protocols to eliminate bot responses in consumer research to ensure data quality
- Create always-on testing frameworks that continually generate new learning and improve models rather than relying on one-time research
- Balance AI efficiency with human creativity by using automated behavioral science to identify where true innovation opportunities exist beyond the statistical mean