Photo credit: venturebeat.com
Join us for updates and exclusive insights on pioneering AI developments.
Imagine a scenario where a diverse group of 200 individuals can engage in a meaningful conversation, effectively brainstorming ideas, sharing insights, weighing alternatives, and swiftly arriving at AI-enhanced solutions. Is such a scenario achievable, and could it elevate their collective intelligence?
A new generative AI innovation, known as conversational swarm intelligence or hyperchat, makes this possible. It allows teams of varying sizes to partake in real-time discussions and rapidly develop AI-optimized solutions. To explore the effectiveness of this technology, I collaborated with Unanimous AI to assemble 50 sports fans, directing them to create a March Madness bracket through live conversational engagement.
However, before sharing the findings of this experiment, it’s crucial to consider why simply gathering 50 people in a virtual meeting wouldn’t yield efficient results. Research indicates that the optimal number of participants for a productive real-time discussion is between 4 to 7 individuals. Smaller groups ensure that everyone has ample opportunity to voice their opinions with minimal delays, while larger groups see a decrease in participation and engagement. By the time you have a dozen people, discussions can spiral into a series of disjointed monologues, and a group larger than 20 often descends into disarray.
So how can 50 people hold a conversation, or even 250, or 2,500?
Hyperchat addresses this challenge by dividing larger groups into parallel subgroups. Each subgroup includes an AI agent known as a “conversational surrogate,” responsible for synthesizing human insights and communicating them as natural dialogue with other groups. This structure allows for the convergence of local discussions into a cohesive large-group conversation, effectively enabling dynamic brainstorming, prioritization, debate, and consensus in real-time.
The innovation of hyperchat isn’t solely to improve efficiency in communication and collaboration at scale. It aims to significantly enhance group intelligence as well. Progress in this area has advanced swiftly, with enterprise teams now leveraging a commercial platform named Thinkscape®, which facilitates real-time deliberations for large groups.
But does hyperchat technology really enhance team intelligence (and can it predict March Madness outcomes)?
To publicly assess this, I tasked the team at Unanimous AI with gathering 50 random sports enthusiasts on their Thinkscape platform to create a March Madness bracket. This bracket was subsequently entered into the ESPN March Madness contest for comparison against approximately 30 million other entries. Notably, this bracket generated by the group is currently performing in the 99th percentile (top 1.4%) of the contest. Here are the details:
While the tournament results can always be unpredictable, the initial performance of this collaboratively-created bracket has exceeded expectations. This is not an isolated instance; the technology has demonstrated its potential on previous occasions.
A recent 2024 study from Carnegie Mellon University and Unanimous AI showed that groups of 35 individuals completing standard IQ tests via hyperchat scored an effective IQ of 128 (the 97th percentile), despite their individual average IQ being 100 (the 50th percentile). This indicates a jump to what can be regarded as gifted-level intelligence performance.
In a different 2024 study, 75 participants engaged in brainstorming sessions to tackle creative problems, utilizing both traditional chat and hyperchat through Thinkscape. Participants reported feeling significantly more productive and collaborative, credited better solutions to the hyperchat experience (p<0.001), and expressed greater ownership and commitment to the outcomes (p<0.001).
This technology is not only intellectually enriching for human groups. It also fosters collaboration between humans and AI agents, paving the way for decisions optimized at an extensive scale while ensuring human oversight. This requires integrating another type of AI agent within the hyperchat system, termed a “contributor agent.” These agents provide factual content during discussions to bolster human deliberation, aspiring toward a hybrid collective superintelligence.
This hybrid approach was trialed in a 2024 study that paired groups of humans with AI agents to field fantasy baseball teams in a hyperchat framework. Participants indicated a preference for this method as it resulted in enhanced decision-making, with 87% endorsing its effectiveness.
Ultimately, conversational swarm intelligence represents a profound advancement in the application of AI agents, transforming collaborative efforts by enabling spontaneous discussions among teams of various sizes. Given that the typical Fortune 1000 enterprise may consist of over 30,000 employees, often working in functional teams with hundreds of participants, this technology holds the potential to eliminate existing barriers that have traditionally restricted real-time deliberations to smaller contingents. It offers an efficient mechanism to harness AI in pivotal decision-making without relinquishing human agency.
The men’s March Madness tournament is ongoing, and while the outcomes remain uncertain, the collective intelligence derived from the 50 sports fans could lead to impressive results. Time will tell.
Louis Rosenberg is the founder of Immersion Corp and Unanimous AI.
Source
venturebeat.com