This article focuses on how we can achieve collective intelligence – where humans and AI work together as a more intelligent whole. The authors argue we need new ways of designing systems that allow humans and AI to effectively share knowledge, pay attention to the right things, and reason through problems together.

​

They propose a system with three interconnected parts:

​

Transactive Memory System: This helps both humans and AI remember information by efficiently storing and retrieving relevant knowledge for the task at hand. 

1. A transactive memory system (TMS) is a cognitive process where individuals within a group develop a shared understanding of who knows what.1 This system relies on two key aspects: (1) knowing the expertise of other group members, and (2) being able to reliably retrieve that knowledge.1

​

​

Transactive Attention System: This ensures humans and AI are focusing on the most important aspects of a problem by coordinating and aligning their attention.

1. A transactive attention system (TAS) is a process for coordinating collaborators' attention to the most important aspects of a task. It involves aligning the shared beliefs of collaborators about what deserves their collective attention.3

2. TAS is crucial in human-AI collaboration because it ensures that both humans and AI are focusing on the right information at the right time.3 For example, in a complex decision-making scenario, an AI could filter vast amounts of data and direct human attention towards the most relevant insights.

3. Developing effective TAS for human-AI collaboration is challenging. Current AI tools, like calendar-based reminders, are limited in their ability to understand the dynamic nature of human attention.3 More sophisticated approaches are needed to model human attention processes and to develop AI tools that can seamlessly integrate with human workflows without being disruptive.3

​

Transactive Reasoning System: This allows for a shared reasoning process where humans and AI can combine their strengths to make better decisions together.

1. A transactive reasoning system (TRS) goes beyond just information and attention sharing. It focuses on aligning the goals, preferences, and beliefs of human and AI collaborators to achieve a shared objective.

2.  This system involves negotiation and understanding different perspectives to arrive at a joint plan of action that all parties are committed to.

3. TRS is crucial for navigating complex, uncertain environments where individual reasoning might be insufficient. By combining human intuition and experience with AI's ability to process vast datasets and explore multiple possibilities, a more robust and effective decision-making process can be achieved.

4. Two markers of a highly effective TRS are:

- Goal Interdependence: Human and AI goals are aligned and interconnected, working towards a shared outcome.

- Metareasoning: This involves reflecting on the reasoning process itself. Both human and AI collaborators should be able to evaluate their own reasoning strategies and adapt them based on feedback from the other.​​

​