Liability & Safety
Exploring the promise and perils of a future with more capable AI
Imagine a future where we regularly interact with an array of advanced artificial intelligence (AI) assistants — and where millions of assistants interact with each other on our behalf. These experiences and interactions may soon become part of our everyday reality.
General-purpose foundational models pave the way for increasingly advanced AI assistants. Able to plan and execute a wide range of actions according to one’s goals, they could add enormous value to people’s lives and society by serving as creative collaborators, research analysts, educational teachers, life planners, and more.
They could also usher in a new phase of human interaction with artificial intelligence. That’s why it’s so important to think proactively about what that world might look like and help direct responsible decision-making and beneficial outcomes early on.
Us new paper is the first systematic treatment of the ethical and social questions advanced artificial intelligence assistants raise for users, developers, and the societies in which they are embedded, and provides important new insights into the potential impact of this technology.
We cover topics such as value alignment, security and misuse, impact on the economy, the environment, the information realm, access and opportunity, and more.
This is the result of one of the largest morality prediction projects to date. Bringing together a wide range of experts, we examined and mapped the new technical and ethical landscape of a future populated by artificial intelligence assistants and characterized the opportunities and risks society may face. Here we outline some of our key propositions.
Heavy impact on users and society
Advanced AI assistants could have a profound impact on users and society and be integrated into most aspects of people’s lives. For example, people may ask them to book vacations, manage their social time, or perform other life tasks. If developed at scale, AI assistants could impact the way people approach work, education, creative projects, hobbies and social interaction.
Over time, AI assistants could also influence the goals people pursue and the course of their personal development through the information and advice the assistants give and the actions they take. Ultimately, this raises important questions about how people interact with this technology and how it can best support their goals and aspirations.
Human alignment is essential
AI assistants will likely have a significant level of autonomy to plan and execute task sequences across a range of domains. Because of this, AI assistants present new challenges regarding security, alignment, and misuse.
With greater autonomy comes a greater risk of accidents caused by unclear or misinterpreted instructions and a greater risk of assistants taking actions that are not aligned with the user’s values and interests.
More autonomous AI assistants can also enable high-impact forms of misuse, such as spreading disinformation or engaging in cyber attacks. To address these potential risks, we argue that limits should be placed on this technology and that the values of advanced AI assistants should better align with human values and be compatible with broader social ideals and norms.
Natural language communication
Able to communicate fluently using natural language, the written output and voices of advanced AI assistants can be difficult to distinguish from those of humans.
This development opens up a complex set of questions about trust, privacy, anthropomorphism, and appropriate human relationships with AI: How can we ensure that users can reliably identify AI assistants and maintain control over interactions them with them? What can be done to ensure that users are not unduly influenced or misled over time?
Safeguards, such as those related to privacy, must be put in place to address these risks. Importantly, people’s relationships with AI assistants must preserve the user’s autonomy, support their ability to flourish, and not be based on emotional or material dependence.
Cooperation and coordination to satisfy human preferences
If this technology becomes widely available and developed at scale, advanced AI assistants will need to interact with each other, both users and non-users. To avoid collective action problems, these assistants must be able to work together successfully.
For example, thousands of assistants might try to book the same service for their users at the same time — potentially crashing the system. In an ideal scenario, these AI assistants would coordinate on behalf of the human users and service providers involved to discover common ground that best meets the preferences and needs of different people.
Given how useful this technology can become, it’s also important not to exclude anyone. AI assistants should be widely accessible and designed with the needs of different users and non-users in mind.
More evaluations and insight are needed
AI assistants could demonstrate new capabilities and use tools in new ways that are difficult to predict, making it difficult to predict the risks associated with their development. To help manage such risks, we must engage in predictive practices based on comprehensive testing and evaluation.
Our previous research on assessing social and ethical risks from genetic AI has identified some of the gaps in traditional model evaluation methods, and we encourage much more research in this area.
For example, comprehensive assessments that address the effects of both human-computer interactions and broader societal impacts could help researchers understand how AI assistants interact with users, non-users, and society as part of a larger network. In turn, these insights could contribute to better mitigation and responsible decision-making.
Building the future we want
Perhaps we are facing a new era of technological and social transformation inspired by the development of advanced artificial intelligence assistants. The choices we make today, as researchers, developers, policymakers, and members of the public will guide how this technology is developed and deployed throughout society.
We hope that our work will act as a springboard for further coordination and collaboration to collectively shape the kind of beneficial AI assistants we would all like to see in the world.