The VerifyWise open-source AI governance team attended the AI Accelerator Institute’s Generative AI Summit in Toronto this week. It was a great event with a full 

Thank you to Tim Mitchell and the team for hosting such an insightful and well-organized event!

A lot of industry and sector leaders such as Brijesh Mandan, Manav Gupta, Himanshu Joshi, Amna Jamal, Nassim Tayari, Parth Dave, Akash Sharma, Christie Mealo, Eli Goldberg, Rob Dunlop, Amit Satpathy, Helen Oakley and Amin Atashi provided with their insights and suggestions about how GenAI will shape our future. 

Their thoughts on AI governance and generative AI were so inspiring. We would like to thank them for driving these important conversations forward.

Here are some of the takeaways from the summit:

Governance

• Generative AI presents non-trivial governance challenges that need careful resolution.
• Governance processes often move slowly, complicating the ability to address these issues effectively.
• Achieving a positive ROI in generative AI is difficult due to inherent complexities.
• Data biases are unavoidable, requiring large language models (LLMs) to filter and manage biases effectively.

ROI and investment trends

• ROI in generative AI has a time series component, making it harder to predict success.
• $250 billion has been invested into AI-related ventures, but this capital tends to move between companies.
• A harsh reality is that many AI companies may never reach profitability.
• VCs may eventually realize that exit opportunities are limited due to intense competition and long timeframes.
• Building a regulated, well-governed application without creating conflicts or “stepping on toes” in the industry is particularly challenging.
• When developing generative AI, start with the problem, not the solution. Think ahead by planning for 2 years into the future.
• Begin with a small, focused approach—a simple API or minimal user interface—and improve incrementally.

Trust and security

• Generative AI feels like living in a sci-fi reality—many predictions from 20 years ago are now becoming true.
• However, AI comes with unique issues like vulnerable dependencies, trojan attacks and inference manipulation
• A proper risk assessment framework for generative AI should involve identifying and evaluating risks, deciding mitigation strategies and monitoring risks over time.
• Generative AI increases attack surfaces, necessitating robust security measures.
• The concept of AI Bill of Materials (AI-BOM) is crucial, listing third-party software dependencies (since 90% of software is open source), information on models, datasets, and training processes.