Responsible AI in an Evolving Regulatory Environment

Generative AI is expected to be a $200 billion industry by 2032. The pace of its growth underscores the need to address the legal, ethical, and privacy concerns created by the technology and its use.

This powerful thought leadership whitepaper from the Cloud Security Alliance, Principles to Practice: Responsible AI in a Dynamic Regulatory Environment, explores the current legal and regulatory landscape struggling to keep pace with the explosive growth of Generative AI (GenAI).? By providing an overview of existing legislation and regulations and their impact on AI development, deployment, and usage, the CSA’s goal is to identify areas where legislation lags.? The key takeaway is a gap is emerging around proper governance.? The CSA suggests a three-pronged collaborative approach: commitment to responsible AI from all tech companies, clear guidelines from policymakers, and effective regulations from legislatures.? The goal is to discuss legal frameworks for responsible AI development and adoption.

Dave Linthicum and I are sharing our take on this well done body of work from the CSA.? We believe that the heart of the message lies in the idea that an understanding of ethical and legal frameworks for AI aligns with three corporate objectives: Building trust and brand reputation, Mitigating risks, and Fostering responsible innovation.

The whitepaper focuses on five topics:

·????? Key Areas of Legal and Regulatory Focus for Generative AI?

·?????? Addressing the Impact of GenAI’s Hallucinations on Data Privacy, Security and Ethics

·?????? Emerging Regulatory Frameworks, Standards, and Guidelines

·?????? Intellectual Property

·?????? Technical Strategies, Standards, and Best Practices for Responsible AI

?

Key Areas of Legal and Regulatory Focus for Generative AI

?

The first topic, Key Areas of Legal and Regulatory Focus for Generative AI reviews existing laws and regulations, including the General Data Protection Regulation (GDPR), California Consumer Privacy Act (CCPA), the California Privacy Right Act (CPRA), and Health Insurance Portability and Accountability Act (HIPAA), that aim to protect individual privacy and data security.? While these laws and regulations do a great job of protecting traditional data, Generative AI presents unique challenges in the realm of data privacy and security.?

Its ability to learn from vast amounts of data raises concerns about how personal information is collected, stored, used, shared, and transferred throughout the AI development and deployment lifecycle.?? Also, the ability for AI systems to take data that has been analyzed and find specific PII data that can be attached, thus making this a huge privacy issue.?

Let's look at GDPR in the context of these shortcomings. We will find that Article 6 of personal data that directly or indirectly identifies an individual may only be collected, stored, or processed with legal grounds. However, prior consent for personal information isn’t familiar territory in Generative AI models today.? In almost all cases, there was no way to know what the AI systems were gathering data online.

Another issue is the right to be forgotten. Article 17 of GDPR, often known as the "Right to erasure" or "Right to be forgotten," gives people the right to ask that their data be deleted in certain situations. Can AI language models forget an individual's personal data??? The core issue, as raised above, is AI systems' ability to “fill in the blanks.”?? This means that even if a specific individual is “erased,” he or she may find that their data finds its way into future databases due to AI’s ability to attach data using existing data that is determined to be anonymous.?? In an article in Forbes, AI expert and social entrepreneur Miguel Luengo-Oroz noted that AI neural networks don’t forget like humans do, but instead modify their weights to reflect fresh data more accurately. This means the information stays with them, and the networks focus on collecting more new data. It is currently impossible to reverse the modifications made an AI system by a single data point at the request of the data owner.

Legal gaps for data protection and privacy as it relates to AI data collection exist in all of the major Privacy regulations and the ramifications are not small.?

The National Library of Medicine, release a publication:? AI Chatbots and Challenges of HIPAA Compliance for AI Developers and Vendors.? In the publication they point out that Developers and vendors of large language models (“LLMs”) — such as ChatGPT, Google Bard, and Microsoft’s Bing at the forefront—can be subject to the Health Insurance Portability and Accountability Act of 1996 (“HIPAA”) when they process protected health information (“PHI”) on behalf of the HIPAA covered entities. In doing so, they become business associates or subcontractors of a business associate under HIPAA.

As we think about data holistically, this includes lawful and transparent data collection and processing, data security and accountability, and individual rights and controls; the regulations continue to evolve and adapt to the complexities of Gen AI.? Additional compliance requirements and increased complexity can be expected. The shortfalls noted above highlight the need for ongoing efforts for organizations to navigate the evolving regulatory landscape while fostering responsible development and deployment of AI

Addressing the Impact of GenAI’s Hallucinations on Data Privacy, Security and Ethics

According to Wikipedia, “a?hallucination?or?artificial is a response generated by AI which contains false or?misleading information?presented as?fact. This term draws a loose analogy with human psychology, where hallucination typically involves false?perceptions. However, there is a key difference: AI hallucination is associated with unjustified responses or beliefs rather than perceptual experiences.” ?

AI models are trained on data and learn to make predictions?by finding patterns in the data. However, if the training data is incomplete or biased, the AI model may learn incorrect patterns, which can lead to incorrect predictions or hallucinations.

The CSA points out that data privacy is a critical area impacted by GenAI hallucinations.

The GenAI models, when fed with sensitive data, have the potential to produce outputs that inadvertently may disclose private information about individuals or organizations. Frameworks like GDPR mandate strict measures to protect personal data from unauthorized access or disclosure. The emergence of AI-generated content blurs the lines between genuine and fabricated information, complicating efforts to enforce data privacy laws effectively.

?

According to CSA, GenAI’s hallucinations also introduce security risks related to regulation. AI-generated content can be manipulated and fabricated. This direct threat to the integrity and security of data systems would require regulatory authorities to adapt existing cybersecurity regulations to address the unique challenges posed by AI-generated content. As GenAI technology evolves and the capabilities of its models advance, ensuring compliance with security standards may become increasingly complex. How will regulators be able to tell if the output is authentic?

?

But the question of hallucinations moves into the realm of ethics as well. There are considerations around responsible development and the use of GenAI models, including the impact of hallucinated content on individuals' rights, autonomy, and well-being. It becomes a balancing act for regulators trying to balance innovation, safeguard personal rights, and ensure governance prioritizes transparency, accountability, and inclusivity.

?

Emerging Regulatory Frameworks, Standards, and Guidelines

?

CSA has noted several federal, international, and tech firm frameworks, standards, and guidelines that can be used by organizations as they move forward with Generative AI.? These include The AI Bill of Rights (White House Blueprint), 2023, the United Nations Global Resolution on Artificial Intelligence, The NIST AI Risk Management Framework, The OWASP Top 10 for Large Language Model Applications project, the OWASP Machine Learning Security Top 10 project, and standards such as ISO/IEC 42001:2023 and ISO/IEC 23053:2022.? T

?

Tech firm guidance comes from firms like IBM's "Trusted AI." Ethics offers guidelines to ensure AI is designed, developed, deployed, and operated ethically and transparently. Microsoft's "Responsible AI Practices" are guidelines and principles for trustworthy AI development and use. AWS’s "Core dimensions of Responsible AI" are guidelines and principles for the safe and responsible development of AI, taking a people-centric approach that prioritizes education, science, and customers. Google’s “Responsible AI Practices and Principles” are designed to guide the development and use of AI responsibly using a human-centered design approach.

The whitepaper clearly states that effective AI regulation needs standardization, accountability, and international cooperation.

An issue to consider, however, is the conflict of interest that may exist between technology companies and the use of AI.? Indeed, they will be the primary beneficiary of the rise of AI, and there is certainly a “Fox guarding the henhouse” type of concern that enterprises and AI users need to consider.?? An alternative would be for the governments to provide regulations and guidance, which they are indeed doing but tend to operate much slower than the technology progresses.?

?

Technical Strategies, Standards, and Best Practices for Responsible AI

This section summarizes some of the technical standards and best practices for implementing responsible AI

The key categories are:

• Fairness and Transparency
• Security and Privacy
• Robustness, Control, and Ethical AI Practices

The whitepaper rounds out its thought leadership by guiding organizations looking to leverage the standards presented.? It suggests an embedding of best practices into the development process to ensure AI’s responsible and ethical use that includes establishing clear internal policies, documentation, and reporting that is regularly shared on data usage, model performance, bias assessments with a path to corrective action and exploring partnerships and collaborations to continue the development of best practices that actively shape the organizational discussion around responsible GenAI.

CSA Whitepaper Link:

https://cloudsecurityalliance.org/artifacts/principles-to-practice-responsible-ai-in-a-dynamic-regulatory-environment

?References:

https://www.forbes.com/sites/ashoka/2023/01/25/we-forgot-to-give-neural-networks-the-ability-to-forget/

https://gdpr-info.eu/art-17-gdpr/

https://pubmed.ncbi.nlm.nih.gov/38477276/

Hallucination (artificial intelligence) - Wikipedia

https://hbr.org/2023/04/generative-ai-has-an-intellectual-property-problem?utm_medium=paidsearch&utm_source=google&utm_campaign=domcontent_bussoc&utm_term=Non-Brand&tpcc=domcontent_bussoc&gad_source=1&gclid=CjwKCAjwm_SzBhAsEiwAXE2Cv635yT5C4uNB8hXNyqjEV663oh2aIhNp61_3gRTT9fi323rsYcn0JBoCqmYQAvD_BwE

#cloud #cloudsecurity #cloudai #cyberai