Generative AI Governace

What is Generative AI

Generative artificial intelligence (also Generative AI or GenAI), describes algorithms (such as ChatGPT) that can be used to create new content such as text, images, audio, music, and videos.  

Generative models work by analyzing large amounts of data to learn patterns and then use that information to predict what would come next in a sequence.

GenAI applications are distinct from traditional AI in that they leverage machine learning techniques to create new, novel, content by understanding and mimicking the patterns of the training data. Traditional AI is primarily concerned with analyzing and making decisions based on input data without creating new content.

Economic Potential of Generative AI

Growth in the number of Gen AI projects is expanding rapidly. McKinsey estimates that Generative AI solutions could add the equivalent of $2.6 trillion to $4.4 trillion annually to global economic output.  The interest and excitement associated with GenAI has been fueling the growth of not only generative models but also that of traditional AI models as well.

Top Generative AI Use Cases

• Chat Based Interfaces Embedded in Products
• Customer Communications and Contact Centers
• IT Automation & Cybersecurity
• Coding Assistants
• Knowledge Assistants (e.g. sales copilot)
• Content Development (e.g marketing personalization)

New Risks Specific To Generative AI

While GenAI offers enterprises the opportunity to create value by applying AI to a new class of problems, this opportunity does not come without risk.  Generative models carry all of the risks inherent with traditional AI models that are also part of the Enterprise AI portfolio.  

For example, they can have issues associated with fairness, bias, or safety.
‍
But beyond the fundamental risks that are common to all AI uses, GenAI and Large Language Models (LLM) in particular, introduce additional risk elements.  

A few of the most significant incremental risks include:

1. Democratization of Model Building - GenAI’s ability to understand natural language through the use of Large Language Models (LLM) introduces the potential for many more individuals (non data scientists) to create models that in turn require some level of oversight.
2. Novel Content Creation
3. Intellectual Property Leakage
4. Copyright Infringement‍
   

The Need for AI Governance

Increasingly GenAI is becoming tied to business revenue generation and cost reduction initiatives.  Just as with traditional AI model development, businesses will need to manage the risks associated with generative models by putting in place an effective AI governance framework.

The regulatory trend is clear.  A significant number of AI specific guides and regulations have been published in the last 5 years.  It is a near certainty that the AI efforts of enterprises, both traditional and generative, will increasingly be the focus of audits and potentially significant penalties for non-compliance.  

For example, failure to comply with provisions of the EU AI Act as it relates to what the act defines as high-risk AI uses, can result in fines up to 20 million euros or 4% of turnover.

Good Governance - Not Just Compliance

AI governance is not only important to ensure compliance with relevant AI regulations. Ensuring that Generative AI applications produce valid and accurate results is also a key benefit of good governance.  

Consider the experience of Air Canada which launched a virtual assistant to help improve the customer experience and lower customer service costs.  Chat bots of this type have become one of the most widely taken on-ramps for enterprises starting GenAI initiatives.

In a widely reported incident that occurred in 2022, Air Canada’s virtual assistant told a customer that they could buy a regular priced ticket and apply for a bereavement discount within 90 days of purchase.  

When the customer submitted their refund claim, the airline turned them down, saying that bereavement fares couldn’t be claimed after ticket purchase. The customer took the airline to a court, claiming the airline was negligent and misrepresented information via its virtual assistant.  

The court found in favor of the plaintiff and ordered Air Canada to reimburse the customer for the full price of the tickets.  While the dollars involved in this example were not significant, it is easy to imagine scenarios where the financial risk could have been much greater.

‍