Module 8: Ongoing AI Issues and Concerns: Awareness of Legal Issues Flashcards
What are the key elements of the EU Commission’s AI Liability Directive?
- Provide similar compensation for AI-related claims as for damage incurred by other products
- Address specific characteristics of AI, such as opacity, autonomous behavior, complexity and limited predictability
- Prevent companies from contracting away their liability for the products to which they contribute
Name 2 examples of initiatives to reform tort law in the US.
1) State regulations on autonomous vehicles with specific safety standards
2) FTC guidance urging businesses to:
- Be transparent with consumers about AI use and how algorithms work
- Ensure decisions are fair, robust and empirically sound
- Hold themselves accountable for compliance, ethics, fairness and non-discrimination
What are the challenges surrounding AI models and data licensing?
- IP rights
- Data licensing terms
- Model licensing and infringement indemnity
What are key elements of negotiating IP rights for AI models?
- Defining the licensor’s ownership
- Defining the licensee’s permitted use of the data
Data licensing terms are used to regulate which concerns between parties?
- Designating certain model components as trade secrets
- Protecting model components by:
- Limiting the right to use the component(s)
- Designating AI components as confidential information in the terms and conditions
- Restricting the use of confidential information
- Including assignment rights in model evolutions from one party or the other
- Determining the license and use rights the parties want to establish between the provider and the user for each model component
- Establishing rights in the terms and conditions
- Liability and indemnification
What are some typical exceptions to the IP infringement indemnity in traditional software or technology licensing agreements?
- Modifications to the software or technology
- Unauthorized combination of the software or technology with other software or technology
- Use of the software or technology beyond the scope authorized in the agreement
Why don’t typical exceptions to IP infringement indemnity work well?
Because modifications and combinations occur WITHIN the model.
An AI model:
- Must be trained, which means modifications to the AI model solution
- Must be combined with training data and production data
- May evolve and exceed a predetermined authorized scope over time
What should AI model licensees insist upon?
- Minimum performance metrics to ensure that the licensed model provides adequate accuracy, reliability and robustness
- Contractual warranties and indemnities from the licensor to mitigate the risk of underperformance
- Thorough testing and validation, in advance of the license and on an ongoing basis
Name some ways in which AI is affecting IP.
- Copyright ownership
- Trademark ownership
- Detection and enforcement of IP rights violations
- Data scraping can lead to IP infringement, trade secret misappropriation or unfair competition claims
- Protection of raw data - can be protected if it:
- Meets specific national law conditions for qualifying as a protected work (a literary, artistic, or other type of work).
- Is part of a database
Name a prominent US court ruling on AI.
Thaler v. Vidal determined that ONLY humans can be named as inventors on a patent, not AI systems.
Name a prominent EU decision on AI.
The European Patent Office (EPO) published two decisions on the refusal of patent applications designating AI as an inventor. The EPO concluded that for the mandatory designation of an inventor for a European patent application the inventor must have a legal personality.
What is a common copyright infringement defense used by AI providers in the US?
The legal doctrine of “Fair Use”: This doctrine promotes freedom of expression by allowing unlicensed use of copyright-protected works in certain circumstances (like criticism, news reporting and research) without permission from or payment to a copyright holder. Courts evaluate fair use claims on a case-by-case basis.
Is raw data protected under current IP law?
Generally no, though it may be protected if it meets specific national law conditions for qualifying as a protected work (a literary, artistic or other type of work), or is part of a database.
How do healthcare non-discrimination laws apply to AI?
- The U.S. Department of Health and Human Services Office of Civil Rights final rule (Section 1557) prevents discrimination based on race, color, national origin, sex, age or disability in health care settings. It requires covered entities to proactively identify and address potential discriminatory impacts of AI tools, taking corrective actions if discriminatory treatment is found.
- The 21st Century Cures Act is not AI-specific, but promotes the use of advanced technologies, including AI. The Act aims to enhance accessibility and transparency of health data to prevent discrimination.
How do insurance non-discrimination laws apply to AI?
- In the US, insurance is state regulated.
- Some discrimination is inherent, e.g., younger drivers being charged more for auto insurance, or people with specific health conditions paying higher health insurance premiums
- NAIC Model Law (2020): Guidelines for regulating AI to ensure algorithms (among other things) do not perpetuate unlawful or unethical discrimination
- NY state has AI-specific guidance for all insurers authorized to provide insurance in that state
How do employment non-discrimination laws apply to AI?
- EEOC Guidance on AI and Hiring (2021): AI and algorithms used in hiring must comply with existing federal non-discrimination laws; AI tools must not disproportionally disadvantage people based on protected characteristics
- NYC local law mandates specific AI audit for automated employment decisions for NYC employees
How do credit and lending non-discrimination laws apply to AI?
- FCRA is the foundational law for anti-discrimination requirements in financial services, but no specific AI guidance
- Consumer Financial Protection Bureau issued a request for information and comments on how AI is used in credit decision-making in 2021
- Additional guidance from the Federal Reserve, Office of the Comptroller of the Currency and other agencies around how AI can be responsibly included in these practices
How do housing anti-discrimination laws apply to AI?
- Fair Housing Act (1968): still the foundational law governing housing discrimination
- Must still be able to demonstrate compliance with anti-discrimination rules even when ranked or scored by an AI system
- U.S. Department of Housing and Urban Development (2020): Guidance focused on automated decision-making in housing and how algorithms used in rental or mortgage lending decisions must adhere to the FHA’s non-discriminatory practices
Which US laws require interpretation to determine how and when they apply to AI technologies?
- Employment: Title VII and EEOC regulations
- Consumer finance: Equal Credit Opportunity Act, the Fair Credit Reporting Act
- SR 11-7: A regulatory standard set out by the U.S. Federal Reserve Bank that gives guidance on model risk management
- OSHA’s guidelines for robotics safety and “hazard analysis”
- The Food and Drug Administration’s (FDA) systemic approval processes for software as a medical device
