As Generative AI Drives Urgency for AI Regulations, Most Companies Unprepared for New Requirements

Verta, the Operational AI company, previewed findings from the 2023 AI Regulations study, which surveyed more than 300 AI and machine learning (ML) practitioners to benchmark the awareness of current and pending regulations covering artificial intelligence and companies’ preparedness to comply with regulatory requirements around Responsible AI, ML model transparency, and data and model lineage, even as widespread concerns about generative AI, and specifically ChatGPT, drive increased urgency around regulating AI.

AiThority Interview Insights: AiThority Interview with Vova Kyrychenko, CTO at Xenoss

Organizations that traditionally have not been under significant regulatory scrutiny must now have MRM practices in place to identify, assess, mitigate and monitor risks associated with their use of ML models, and to ensure their models are fully compliant

The study was conducted in March-April 2023 by Verta Insights, the research practice of Verta Inc., and found that while a majority of organizations view AI regulatory compliance as a priority and believe that they will face an increasing number of AI regulations in the near future, few companies today are well prepared to meet current or future regulatory requirements.

The European Union is set to pass the EU AI Act this year, the US Congress has taken up the American Data Privacy and Protection Act (ADPPA) and the Algorithm Accountability Act in recent sessions, and US states like California, Illinois, New York and Texas have passed or proposed laws regulating AI. These laws create new compliance and reporting requirements around companies’ use of AI and machine learning, intended to protect consumers against privacy violations, bias in automated decision-making and other potential harms.

“Generative AI technology like ChatGPT and Stable Diffusion are raising issues around copyright, Responsible AI and cybersecurity that are helping to drive increased urgency around regulating AI. This trend is converging with the continued increase in regulations like the EU’s GDPR covering data privacy to make AI regulatory compliance a high priority for CEOs and board members at more than half the companies in our study,” said Manasi Vartak, Founder and CEO of Verta. “This is true for organizations in industries like insurance and financial services that traditionally have had ML models regulated via Model Risk Management practices. New regulations are increasing the complexity of compliance for these companies. At the same time organizations that traditionally have not been under significant regulatory scrutiny must now have MRM practices in place to identify, assess, mitigate and monitor risks associated with their use of ML models, and to ensure their models are fully compliant with a growing number of regulations.”

Most Companies Unprepared for AI Regulatory Compliance

In the research study, 55% of companies in the study said that regulatory compliance is a C-level or board-level priority. Only 11% said that it is a “low priority.” In addition, 89% of participants agreed that AI regulations will increase over the next 3 years, and 77% believe that AI regulations will be strictly enforced.

Despite these findings, the study revealed a disconnect between the high priority that companies are placing on regulatory compliance and their current level of preparedness for compliance. For example, nearly two-thirds (62%) of participants were not confident that their company would be able to complete the algorithm impact assessments called for in the ADPPA. Only 28% were highly confident they could complete the necessary reporting.

Alarmingly, the median time that study participants said they would need to complete an algorithm impact assessment for a single model was 40 hours. For organizations with tens or even hundreds of models or model versions in production, the costs of providing fully compliant reports to satisfy ADPPA requirements could be significant given the high salaries that AI/ML roles are commanding in today’s market.

The Verta Insights research also found that close to 90% of companies have little or no automation in place for the AI governance processes that they will need to rely on to ensure regulatory compliance, like bias detection and mitigation, model explainability and transparency, and model validation and testing. Similarly, just 10% of companies have fully or highly automated their model documentation process, which would make it challenging to comply with regulatory requirements around model transparency and explainability.

“Companies typically react to regulatory pressures in a predictable curve, where we see leaders and fast followers making substantial early investments in the people, processes and technology necessary for compliance. Laggards and late followers, on the other hand, tend to wait until a regulation is imminent or has taken effect before they prepare for compliance. This means that leaders are prepared for compliance on Day One when the regulations go into effect, while laggards are frequently left scrambling to comply, resulting in additional cost, lost business or even legal risks,” explained Rory King, Head of Verta Insights Research and Verta Go-to-Market.

Read More about AiThority Interview: AiThority Interview with Ahmad Al Khatib, CEO and Founder at Qudo

King added that, because leaders have made early investments with the goal of ensuring compliance, they also accrue ancillary benefits because they have established standardized processes, transparency and traceability that make their machine learning pipelines more efficient and effective.

Advanced AI/ML Maturity Gives Companies an Edge with Regulations

In the study, 84% of participants said that their organizations viewed regulatory compliance as equally important as other priorities driving AI/ML investments, including 43% who said that compliance was a higher priority driving investments. The study also broke out results for companies classified as “leading performers” that always or usually meet their financial targets, against “lagging performers” that sometimes, rarely or never meet targets. Leading performers were more than twice as likely to view regulatory compliance as a “much more important” driver of AI/ML investments than lagging performers.

The research further showed that leaders have invested in certain technologies that support compliance to a greater extent than lagging performers. Specifically, leading performers were more than twice as likely (53% versus 24% for laggards) to already be using tools for ML monitoring, technology supporting model governance and risk management (leaders: 38%; laggards: 24%), and a model catalog (32% vs 19%). These technologies help organizations implement controls and processes to ensure that ML models are developed, tested and deployed in a way that aligns with ethical and regulatory standards, and that the organization can demonstrate regulatory compliance when necessary.

“The investments that companies make in ML tooling have impacts when it comes to regulatory compliance. For example, leading organizations are 50% more likely to have invested in a model catalog to inventory models. This helps explain why leaders are more than twice as likely to be highly confident that they can produce a complete list of all their organization’s models currently in production to meet regulatory requirements. Companies need a model catalog as a centralized system of record for their ML assets before they can begin to manage those assets effectively for compliance,” said Vartak.

 Latest AiThority Interview Insights : AiThority Interview with Brad Anderson, President of Product and Engineering at Qualtrics

 [To share your insights with us, please write to sghosh@martechseries.com]