The use of AI to facilitate CI presents specific challenges. CI methods can be used to enable more sophisticated learning, for example. By moving beyond the mere analysis of new data to the generation of new categories for that data, we can create a more effective pathway for both understanding and solving problems. Currently, AI does not do these "learning loops" well.
Finding ways for human-computer partnerships to share agency effectively and to delegate control in an optimal manner will be important.24 Human decision-making is highly context-dependent and that context is not static, so CI technology must be able to adapt on-the-fly to the ever-shifting circumstances in which human groups are operating. It will be crucial for humans to understand what CI technology is doing, and why.
This will require significant breakthroughs. Even such basic tasks as getting machines to understand who is speaking, or what they are saying, are proving difficult: imbuing machines with social intelligence remains a distant goal. Human-AI collaborative teams will also require machines that contain some kind of theory of mind --- the ability to model the mental states of others.25 As technology increasingly mediates group interactions it will also be necessary to find ways to measure how machines are changing human behaviour, to ensure they are enhancing human capabilities rather than causing them to atrophy.