美国电子隐私信息中心-生成式人工智能的潜在有害影响与未来之路（英）-2023.5-86页_738kb

Summary of "Generating Harm: Generative AI's Impact and Paths Forward" (EPIC, May 2023)

Overview:
This report, edited by Grant Fergusson, Calli Schroeder, Ben Winters, and Enid Zhou with contributions from various authors, examines the documented and anticipated harms of generative AI as of May 15, 2023. It highlights how generative AI tools (e.g., ChatGPT, image generators) amplify risks like misinformation, harassment, privacy invasions, data breaches, intellectual property theft, climate change impacts, labor displacement, and discrimination. Due to the rapid evolution of AI, the report notes it is inherently dynamic and subject to future updates.

Key Findings:

• Information Manipulation: AI accelerates the spread of scams, disinformation (purposeful false information), and misinformation (less purposeful false information), exploiting technical capabilities for manipulation, such as phishing, deepfakes, and clickbait, often blurring the lines between accuracy and deception.
• Harassment and Extortion: AI enables impersonation, harassment, and extortion through deepfakes and manipulated media, causing physical, economic, reputational, psychological, and autonomy harms. Legal challenges arise from intent requirements and evidence authentication.
• Privacy and Data Risks: Generative AI relies on large-scale scraping of personal data without consent, leading to privacy violations, data breaches, and misuse of information. Data minimization and transparency are critical for mitigating these risks.
• Security and Labor Concerns: AI increases security threats, such as malware creation and cyberattacks, while displacing labor through automation, potentially exacerbating economic inequality and job loss. Labor union actions and ethical implications are discussed.
• Intellectual Property: AI training on copyrighted works raises issues of infringement and lack of compensation for creators, with debates around fair use and ownership of AI-generated content.
• Environmental Impact: Training AI models consumes vast energy and resources, contributing to carbon emissions, water usage, and environmental damage, amplifying climate change risks.
• Discrimination and Social Effects: AI perpetuates biases in hiring, content generation, and societal stigmas, disproportionately affecting marginalized groups.
• Economic and Market Power: Monopolization by tech giants and unfair concentration of power limit competition, raising concerns about antitrust violations and unequal access to AI benefits.

Recommendations:
The report proposes comprehensive paths forward, including:

• Legislative actions, such as banning broad immunity under Section 230, strengthening data privacy laws like the ADPPA, and enacting laws for consumer protection.
• Regulatory interventions, including audits, environmental footprint requirements, and prohibitions on unfair AI practices.
• Administrative and enforcement measures, such as FTC and CFPB oversight for deceptive practices and product liability applications.
• Private actor responsibilities, such as transparent data use, tools for detecting AI-generated content, and fair labor practices to redistribute profits.

Overall, the report advocates for proactive measures by policymakers, regulators, and industry to mitigate harms, ensuring AI development balances innovation with accountability, ethical standards, and equity.