Power-shift AI: Human-Centred AI for the Equitable Smart Grid

As part of my Bridging Responsible AI Divides fellowship, we have been exploring a scenario for renewable energy futures. In the scenario a small group of households share solar PV and battery storage. This ‘microgrid’ is managed by a reinforcement learning agent (it charges and discharges the battery, buying and selling energy from the national grid). While there is a lot of work looking at AI optimisation of locally generated renewable, the trend so far has been to abstract out individual household practices or simulate them using game theoretic models. What’s more, designers tend to assume that the AI will be a invisible to people, with no need for individual households to have knowledge or input on its workings.

My fellowship has taken a human-centric approach, building on work done on the Beyond Individual Persuasion project to combine AI with eco-feedback (where users are able to see their own energy data). In particular, we focused on how low-income communities understand and respond to such AI-integrated systems. We installed simple electricity sensors in participants’ homes and created realistic simulations of the scenario. They could log on to a web app to view their own data and coordinate usage of shared solar by pre-booking domestic activities. They were interviewed after a month-long trial. Many participants saw the AI as an authority figure but one that would be more reasonable and trustworthy than the local council (which acts as the landlord or the freeholder for many of our participants). Many perceived the AI as being useful for mediating existing tensions with the council or diffusing their power. There were concerns that the AI could allow new forms of surveillance for the council or commercial organisations. Disabled participants worried that the scenario would make their higher energy needs be seen as undermining the system and again looked to the AI to mediate this social tension. Some viewed the act of viewing and booking their energy use as personally empowering.

This work was complemented with participatory design workshops in which Year 8 and 9 secondary school pupils envisaged apps that could support the scenario in their local council estate. They co-designed lo-fi prototypes that prioritised forms of controlled pro-social communication and target setting without recrimination. Their diverging visions for energy futures also exposed tensions between forms of participation such as individual freedom versus support for the most vulnerable. One school pitched their ideas to their council climate team and there will be a follow-up event to share their ideas and broader study findings with stakeholders involved in green energy transition in the council.

Overall, the work showed opportunities and challenges in AI-managed communally shared resources around differing values and systems of social power. Participants were often keen to defer to the AI to manage not only their own energy but also how their energy use was to be evaluated by others, though the present system could only partially offer such functionality. 

This work was carried out in collaboration with Enrico Costanza, Georgia Panagiotidou, Malak Ramadan and other members of the Beyond Individual Persuasion team.

Link to the Project website