美国电子隐私信息中心：2023生成式人工智能的潜在有害影响与未来之路（英文版）_86页_13mb

Summary of "Generating HARMs"

Core Content

This paper, titled Generating HARMs, explores the various harms associated with generative AI and outlines potential legal and regulatory interventions to mitigate them. It was published in May 2023 and authored by a group of experts, including Grant Fergusson, Caitriona Fitzgerald, and others. The paper emphasizes that while generative AI is a relatively new technology, its potential for harm is not, and that the rapid deployment of AI tools without adequate safeguards has led to significant risks for individuals, democracy, and cybersecurity.

Main Viewpoints

• Generative AI amplifies existing risks: The technology accelerates the spread of harmful content, including scams, disinformation, misinformation, cybersecurity threats, and clickbait.
• Privacy and consent are compromised: AI-generated content often uses publicly available images and videos, making it difficult to apply traditional privacy torts.
• Legal frameworks are inadequate: Existing laws, such as those related to defamation, fraud, and cyberstalking, are not well-suited to address the unique challenges posed by generative AI.
• The need for transparency and accountability: Companies should be required to disclose the use of AI, ensure informed consent, and build in fairness, accountability, and transparency from the start.
• Public interest and democratic integrity are at risk: AI tools can be used to manipulate public opinion, spread false information, and interfere with elections.

Key Information

Types of AI-Generated Harm

• Economic Loss: Scams and malware can lead to direct financial harm, affecting credit and causing long-term consequences.
• Reputational/Relationship/Social Stigmatization: Misinformation can damage an individual's reputation, relationships, and social standing.
• Psychological Harm: Disinformation and clickbait can cause emotional distress, shame, and feelings of manipulation.
• Autonomy: The spread of false information and surveillance-based advertising can undermine individual autonomy and decision-making.
• Discrimination: AI-generated content can target vulnerable groups, such as the elderly, immigrants, or those in debt, increasing their susceptibility to harm.

Examples of Harm

• Fake bomb threats: AI was used to create false threats against public institutions.
• Impersonation of loved ones: AI voice generators were used to trick individuals into sending money for bail or legal assistance.
• Deepfake pornography: Nonconsensual sexual imagery was generated using AI and circulated online.
• AI-generated misinformation: Google's Bard chatbot replicated 78 out of 100 conspiracy theories without context or disclosure.
• Inaccurate AI-written articles: CNET and Buzzfeed published AI-generated content that was later found to be misleading or useless.

Legal and Regulatory Interventions

• Enact laws against election disinformation: Such as the Deceptive Practices and Voter Intimidation Prevention Act.
• Pass the American Data Privacy Protection Act (ADPPA): To limit the collection and use of personal information for targeted advertising and scams.
• Implement FTC Commercial Surveillance rules: To enforce data minimization and prevent the misuse of personal data for AI training.
• Strengthen existing criminal laws: Including those related to cyberstalking, fraud, and defamation, to address AI-generated harassment and impersonation.

Structure of the Report

The paper is organized into the following sections:

• Introduction: Highlights the rapid rise of generative AI and its potential for harm.
• Turbocharging Information Manipulation: Discusses the spread of disinformation, misinformation, and malicious content.
• Harassment, Impersonation, and Extortion: Focuses on the use of deepfakes and AI-generated content for personal and political harm.
• Profits Over Privacy: Examines the increased data collection and its implications for privacy.
• Increasing Data Security Risk: Explores how AI can be used to create and refine malware.
• Confronting Creativity: Addresses the impact of AI on intellectual property and creative industries.
• Exacerbating Effects of Climate Change: Discusses the environmental costs of AI training and operation.
• Labor Manipulation, Theft, and Displacement: Reviews how AI may affect employment and labor practices.
• Spotlight: Discrimination: Examines the role of AI in reinforcing stereotypes and targeting vulnerable groups.
• The Potential Application of Products Liability Law: Proposes holding AI developers accountable for harms caused by their tools.
• Exacerbating Market Power and Concentration: Analyzes the concentration of AI development power in a few companies.
• Recommendations: Offers a range of policy and legal measures to address AI harms.
• Appendix of Harms: Provides a comprehensive list of AI-related harms.
• References: Cites sources and studies that support the paper's arguments.

Conclusion

The paper underscores the urgent need for regulatory and legal frameworks that can effectively address the harms of generative AI. It calls for greater transparency, accountability, and legal clarity to ensure that AI is used responsibly and that its negative impacts are mitigated. The authors emphasize that without such measures, the risks of AI will continue to grow, especially as the technology becomes more accessible and sophisticated.

Key Sections Summary

Turbocharging Information Manipulation

• AI tools facilitate the spread of false, misleading, and harmful content.
• Examples include scams, disinformation, misinformation, cybersecurity threats, and clickbait.
• AI-generated content can be used to influence public opinion, harass individuals, and interfere with elections.

Harassment, Impersonation, and Extortion

• Deepfakes are a major concern, used to impersonate, harass, and extort individuals.
• AI-generated content often uses public material, making it hard to apply privacy laws.
• Legal challenges include determining intent, privacy, and consent in cases of deepfakes.
• Victim redress is complicated by the First Amendment and the difficulty of identifying creators.

Recommendations

• Enact laws to prevent election disinformation.
• Pass the American Data Privacy Protection Act.
• Implement FTC rules on commercial surveillance.
• Strengthen existing criminal and civil laws to address AI-related harms.

Legal and Ethical Challenges

• First Amendment issues: AI-generated content can blur the line between free speech and privacy violations.
• Privacy torts: Traditional legal claims like intrusion upon seclusion and publication of private facts may not apply to AI-generated content.
• Consent and intent: Determining the intent of AI creators and whether they should be held legally accountable is complex.
• Economic interests: Some laws focus on economic harm rather than emotional or reputational damage, which may not fully protect victims.

Final Thoughts

The paper serves as a call to action for policymakers, regulators, and the public to recognize the potential for harm from generative AI and to develop appropriate legal and ethical responses. It highlights the importance of transparency, accountability, and the need for updated legal frameworks that can address the unique challenges of AI-generated content.