Skillprint has released new findings from Signal, its human-AI evaluation platform, revealing a critical challenge as artificial intelligence becomes increasingly integrated into workplaces and everyday life: even the most advanced AI systems still struggle with the communication, coordination and adaptability required to work effectively alongside humans.

The experiment tested popular AI models, including those from OpenAI, Anthropic and Google, alongside human participants in a cooperative platform game. Teammates were connected by a virtual rope and had to work together to complete challenges. Rather than measuring how well AI could solve problems independently, the experiment examined a more practical question: Can AI effectively collaborate with people, adapt to their behavior and help teams achieve better outcomes?

● Humans dramatically outperform AI at collaboration. Human-only teams achieved an 84% win rate, compared with just 14% when different AI models worked together. This sixfold difference highlights a significant gap in AI’s ability to coordinate actions, plan together and achieve shared objectives.

● Human leadership and collaboration styles make a measurable difference. Teams consisting of two humans and one AI agent performed better when a human directed the team’s strategy than when the AI agent led. The findings also suggest that how teammates communicate, coordinate and divide responsibilities can influence outcomes, underscoring the importance of collaboration strategies, not just individual intelligence.

● More AI agents don’t necessarily create better teams. AI win rates declined as teams expanded from two agents to three and four, highlighting how additional agents can introduce communication and coordination challenges rather than automatically improving results.

Source link