Recommendations for Generative AI
The World Economic Forum (WEF) proposes 30 recommendations for generative AI to facilitate the use of artificial intelligence.
Generative AI refers to artificial intelligence technology that utilizes algorithms and models to create content through methods such as text, images, and videos. In April of this year, WEF and AI Commons jointly held the Responsible AI Leadership Conference, aiming to guide technical experts in responsible development and application of generative AI.
The recommendations proposed by WEF are the consensus reached by industry experts during the discussion. These suggestions consist of three parts: Responsible Development and Release, Open Innovation and International Collaboration, and Social Progress.
Responsible Development and Release
WEF believes that to protect society from unexpected impacts caused by generative AI, stakeholders should be advocated for responsible implementation of research and development. WEF proposes the following recommendations:
- Use precise and shared terminology: Stakeholders need to use more precise terminology in design, development, and evaluation to promote effective communication and establish effective standards.
- Establish public expectations for artificial intelligence: Stakeholders need to inform users of the uncertainty and randomness of generative AI, so that users have accurate expectations of them.
- Pay attention to human values and preferences: When researching generative AI, developers need to conform to human normative values and preferences.
- Encourage multi-party participation: Policy makers should promote the participation of all parties in artificial intelligence research and development, enhancing diversity and inclusiveness.
- Adopt strict benchmarks: Testers need to experiment with generative AI using established benchmarks and comprehensively evaluate the model’s performance.
- Adopt diverse opposing opinions: Testers can choose diverse opposing opinions (Red Team, personnel who analyze potential weaknesses and vulnerabilities from a critical perspective).
- Adopt a transparent release strategy: The project team needs to responsibly release models in product design to ensure that potential risks can be identified.
- Listen to user’s feedback: The project team needs to enable users to provide real-time feedback on the model output.
- Increase traceability: The project team needs to increase the traceability of the artificial intelligence lifecycle.
- Ensure content traceability: The project team needs to track the way content is generated and provide users with methods to distinguish between human generated and artificial intelligence generated content.
- Disclose artificial intelligence interaction: The project team needs to inform users whether they are interacting with humans or machines.
- Establish human trust in artificial intelligence: The project team needs to consider the characteristics of transparency and consistency, so that users can understand how artificial intelligence achieves results.
- Implement a step-by-step review process: The project team needs to specify a step-by-step review process to supervise artificial intelligence before and after product launch.
- Develop a comprehensive measurement framework: The evaluator needs to adopt a multi-level framework, focusing on the technical performance of the product and its impact on society.
- Adopt a sandbox environment: Stakeholders can collaborate in a sandbox environment to ensure product safety and compliance.
- Update intellectual property policies: With the creation of generative AI, regulatory agencies need to re-examine and update intellectual property related laws to determine the ownership of existing content.
Open Innovation and International Collaboration
In terms of innovation cooperation, WEF believes that countries should strengthen academic and industry communication. WEF proposes the following recommendations:
- Encourage public-private research cooperation: Public and private parties should promote cooperation throughout the technology development lifecycle.
- Establish a universal platform: Researchers should contribute to the joint development and use of code, models, data, benchmarks, and other content.
- Support knowledge sharing: Researchers should provide open access to calculations, data, and models for the public.
- Strengthen international cooperation: International institutions should promote cooperation in artificial intelligence among countries.
- Establish a global artificial intelligence governance initiative: International institutions should develop a global artificial intelligence governance initiative to coordinate the risks and challenges faced by artificial intelligence.
Social Progress
WEF believes that generative AI will have a broad impact on society, and society needs to be more fully prepared in the face of new changes. WEF proposes the following recommendations:
- Prioritize social progress: Stakeholders need to ensure that the social impact of artificial intelligence technology is prioritized.
- Improve public artificial intelligence literacy: Education institutions need to enhance public awareness of generative AI and create an atmosphere of knowledge and participation.
- Cultivate holistic thinking methods: Education institutions need to cultivate diverse thinking among users to prepare for the era of artificial intelligence.
- Address the impact of artificial intelligence: Stakeholders need to understand the impact of artificial intelligence on human interaction, knowledge dissemination, and evaluation mechanisms.
- Address social welfare issues: Stakeholders need to encourage the application of artificial intelligence in social welfare issues such as healthcare and climate change.
- Reduce baseline infrastructure differences: Stakeholders need to increase public investment in infrastructure.
- Cultivate artificial intelligence talents: Countries need to adopt targeted incentives, communication plans, and other methods to cultivate artificial intelligence talents and serve the public interest in the future.
- Increase opportunities for developing countries: Stakeholders need to enhance the participation capacity of developing countries in artificial intelligence technology.
- Protect cultural heritage: Stakeholders need to protect language diversity and cultural heritage while developing artificial intelligence.
Reference:
The Presidio Recommendations on Responsible Generative AI
