Will the U.S. Congress pass, and the President sign, a bill by January 10, 2025 11:59 PM EST that meets one or more of the criteria for 'Major AI Regulation'?
Criteria:
- Privacy and Data Rights
- Mandates the explicit informed consent of individuals before their personal data is collected
or processed by AI systems.
- Restrictions on the sale, sharing, or third-party use of personal data processed by AI
systems without clear and explicit user consent.
- A right to data erasure, allowing individuals to request the deletion of their data from AI
systems.
- Requirements for timely notification to affected individuals in the event of a data breach or
unauthorized access to their personal data within AI systems.
- Transparency and Explainability
- Requirements for AI systems to provide user-friendly explanations for their decisions,
particularly when used in critical sectors like health, finance, or judicial processes.
- Obligations for AI systems that emulate humans to reveal to people / users that they are AI,
and / or generic disclosure rules for outputs of generative AI.
- Obligations for AI developers or deployers to disclose the datasets and methodologies used
in training the AI, ensuring external auditors can assess the system's fairness and reliability.
- Bias and Fairness
- AI developers must implement tools and methodologies to identify and rectify biased
decision-making in AI systems.
- Periodic third-party audits to ensure AI systems are not perpetuating or amplifying societal
biases, with findings made available to the public.
- Safety and Reliability
- Guidelines or standards for the validation and testing of AI in sectors where human safety is
at risk.
- Ongoing monitoring requirements for deployed AI systems to track potential drifts from
expected behavior, with mandated corrective actions when anomalies are detected.
- Targeted bans or other legal restrictions on the use of “high risk” AI systems, e.g. bans on
voice cloning or models with a biosafety risk.
- Accountability and Liability
- Clear definitions of legal responsibilities for AI system malfunctions or erroneous decisions,
whether they lie with developers, deployers, or operators.
- Amendments to Section 230 of the CDA to clarify that the liability protections for internet
platforms do not extend to AI outputs / companies
- A framework for affected individuals or entities to seek redress or compensation in the event
of harm or damages caused by AI systems.
- Human Oversight
- Requirements for AI systems, particularly those with autonomous decision-making
capabilities, to have human-in-the-loop mechanisms where a human can intervene in real-time
decisions.
- Obligations for regular review and validation of AI decisions by human experts in sectors
deemed critical.
- Training Runs and Compute
- Disclosure mandates for AI developers to report on the compute resources used during
model training, to ensure environmental sustainability and ethical use of computational
resources.
- Restrictions on training runs that utilize datasets in violation of privacy and data rights or that
exceed certain environmental thresholds.
- Guidelines on the acceptable sources and methods for data collection for training AI models,
ensuring ethical sourcing and data quality.