Artificial Intelligence (AI) models such as ChatGPT and Google’s Bard have demonstrated remarkable capabilities in generating natural language texts, images, and other forms of content. These models are based on large-scale neural networks that are trained on massive amounts of data, often collected from the internet or other sources. While these models can offer many benefits for various applications and domains, they also pose significant risks and challenges for society.
One of the main concerns is the lack of transparency and accountability of these AI models. Transparency means being able to understand how the models work, what data they use, what assumptions they make, and what outcomes they produce. Accountability means being able to monitor, audit, evaluate, and regulate the models’ performance, behaviour, and impact. Without transparency and accountability, these AI models may cause harm or injustice to individuals or groups, violate ethical principles or legal norms, or undermine trust and confidence in the technology.
Therefore, there is an urgent need for effective governance of these AI models, both at the national and international levels. Governance refers to the set of policies, laws, regulations, standards, and practices that aim to ensure that AI is used in a responsible, ethical, and beneficial manner. Governance also involves the participation and collaboration of various stakeholders, such as governments, industry, academia, civil society, and users.
However, governance of AI models is not a simple or straightforward task. It requires balancing multiple and sometimes conflicting objectives, such as innovation, security, privacy, fairness, human rights, and social welfare. It also requires addressing complex and dynamic technical, legal, ethical, and social issues that may arise from the development and deployment of these models. Moreover, it requires adapting to the rapid and evolving nature of AI technology and its applications.
To address these challenges, I propose six suggestions for how the world can police the creation of AI models such as ChatGPT and Google’s Bard:
- Establish clear and consistent principles and guidelines for AI ethics
There is a need for a common framework that defines the core values and principles that should guide the design, development, and use of AI models. These principles should reflect universal human rights and ethical standards, such as fairness, accountability, transparency, privacy, safety, and human dignity. Several initiatives have already proposed such principles, but more efforts are needed to harmonise them across different regions and sectors. - Develop technical standards and best practices for AI quality
There is a need for a set of technical specifications and methods that ensure the quality and reliability of AI models. These standards should cover aspects such as data quality, model validation, explainability, robustness, security, and performance. They should also provide guidance on how to test, monitor, audit, and debug AI models throughout their lifecycle. Several organisations have already developed or proposed such standards, but more coordination and adoption are needed among different stakeholders. - Implement legal frameworks and regulations for AI accountability
There is a need for a set of laws and rules that define the rights and responsibilities of AI developers, users, and affected parties. These laws should address issues such as liability, consent, ownership, intellectual property, and oversight of AI models. They should also provide mechanisms for redress, remedy, and enforcement in case of harm or violation caused by AI models. Several countries have already enacted or proposed such laws, but more harmonisation and compliance are needed at the global level. - Promote ethical culture and education for AI developers and users
There is a need for a culture of responsibility and awareness among those who create and use AI models. This culture should foster ethical values and norms, such as honesty, integrity, respect, and empathy. It should also encourage critical thinking and reflection on the potential impacts and implications of AI models. Moreover, there is a need for education and training programs that equip AI developers and users with the necessary skills and knowledge to understand and apply ethical principles and standards to their work. Several initiatives have already launched or supported such programs, but more resources and outreach are needed to reach a wider audience. - Engage diverse and inclusive stakeholders in AI governance
There is a need for a participatory and collaborative approach to AI governance that involves multiple and diverse stakeholders, such as governments, industry, academia, civil society, and users. These stakeholders should have a voice and a role in shaping the policies, laws, regulations, standards, and practices that govern AI models. They should also have access to information and tools that enable them to understand and evaluate AI models. Moreover, they should have opportunities to provide feedback and input on the development and deployment of AI models. Several platforms have already facilitated or supported such engagement, but more representation and empowerment are needed for marginalised or vulnerable groups. - Strengthen international cooperation and coordination on AI governance
There is a need for a global dialogue and action on AI governance that fosters mutual learning, sharing, and collaboration among different countries and regions. This dialogue should aim to align and harmonise the principles, guidelines, standards, and regulations that apply to AI models across borders. It should also address common challenges and opportunities that arise from the global nature of AI technology and its applications. Moreover, it should promote joint initiatives and projects that advance the common good and public interest of humanity. Several forums have already initiated or supported such dialogue, but more commitment and leadership are needed from global actors.
These suggestions are not exhaustive or definitive. They are meant to stimulate discussion and action on how to police AI models for transparency and accountability. I believe that by implementing these suggestions, we can ensure that AI models such as ChatGPT and Google’s Bard are used in a way that respects human values and serves human needs.
The CENTER FOR AI SAFETY website is a good place to start if you are interested in AI safety.
Is anyone thinking about the emotional impact of ai-powered virtual personal assistants?