Genesis of
the Project
Policymakers often come to Google DeepMind to understand more about the latest technologies, how they’ll be used and misused, and what we recommend they consider as they draft policies. Sometimes our recommendations are clear cut - they involve areas we know well and have expertise in. Others involve areas with which we are less familiar, or that are changing rapidly, things like how AI might affect jobs, the planet or historically excluded groups.
In order to understand and build safe and beneficial AI in different contexts and geographies, we must engage meaningfully with civil society, experts and practitioners so that we can learn from one another.
In 2022, we partnered with the Aspen Institute to convene a discussion about how the benefits of AI might be distributed equitably. The conversation was summarised in the report A Blueprint for Equitable AI, which identifies some key questions that require further exploration and collaboration. This report was the starting point for convening a series of in-depth discussions on the challenges facing policymakers, and possible ways to address them. The concept of equitable AI, which (COPY TBC)
The rapid pace of AI’s development brings unique challenges for policymakers. The advent of more powerful systems with more general-purpose capabilities gives rise to governance challenges, as well as the need to manage complex societal impacts, which may be hard to predict fully. The capabilities of these models and their successors are emerging in parallel with understanding of their implications, meaning policymaking must be iterative and dynamic. Meanwhile, the risks and opportunities of current AI systems are unfolding now, requiring policymakers to respond to both current and future scenarios and needs simultaneously.
If the development and deployment of ‘frontier models’ are steered responsibly in positive directions, the collective benefits will be huge. The aim of this report is to surface and share the most pertinent questions, insights and perspectives that emerged from our roundtable series, in the hope that they will serve policymakers’ shaping of AI regulation.
While each conversation took a slightly different format, the program overall sought to explore the following questions:
Where should AI be used?
What are the highest priority areas where AI can have a positive impact on society? Where should AI not be used? Which areas might need more time to develop meaningful solutions that don’t exacerbate existing problems?
Who should be accountable for the use of AI?
What role do companies, governments, and individuals play in the safe use of AI? How can society hold each group accountable?
How should AI be built?
Where is deep collaboration needed? What institutions should we create?
How should AI be governed?
What are the key regulatory challenges? What other technologies and regulations can we draw from as a model for AI regulation?
How do current governance regimes compare to those we may need to support equitable AI in the future?
The answers to these questions are context specific and each sector faces its own challenges - deploying AI safely will require close collaboration with those with on the ground experience and expertise to avoid exacerbating existing inequalities and structural problems. The insights from these conversations are not exhaustive, and nor were the conversations. Many more sectors and communities than are discussed here are impacted by the risks and opportunities of AI. Our aim is to continue these conversations, broadening as well as deepening them and helping to translate insights into concrete and measurable policies. For this series of engagement, one of the best outcomes is to have started the conversation and connected people who will continue to work together.
It was clear from the discussions that, alongside serious concern about the current harms and future risks of the technology, there is undeniable excitement about the possibilities of AI. Participants were inspired by the promise of medical breakthroughs, climate solutions, improvements in educational attainment and the reaching and expanding of human potential. Greater equality was felt to be possible in some areas - for example, for disabled people and other previously marginalised communities in the workforce. Even when the risks appear more concrete, the potential for progress is taken seriously and supported by evidence.
A sense of commonality emerged from the fact that all sectors are grappling with rapidly advancing technology that will effect significant changes.
Each has a great deal of expertise, but anticipating the future and designing policies that can respond to emerging technology is challenging and daunting as well as exciting. While it is certainly true that incentives can be misaligned and different groups can talk past each other without intending to, the conversations overall reflected humility, seriousness and a sense of shared experience that bodes well for future collaboration.