5 considerations when developing a AI / LLM corporate policies

Every decade a new technology emerges that is truly disruptive, and Artificial Intelligence (AI) is clearly that new global MEGA trend. The general public’s reaction to all things #AI and #GPT in the last few months is quite similar to the introduction of desktop computing by Apple / Microsoft in the early 90s. It was incredibly impressive to the public at the time, but people had no idea how far and wide it would change how we live, play and work. It opened the floodgates for tech innovation - and given rise to thousands of tech companies that each offer unique capabilities for their customers. And just like then, I know that we are at the beginning of the AI journey. I expect there will be many different providers, hundreds of different models, and lots of different use cases and needs served.

As AI and #LargeLanguageModels (#LLM) advance, they present an array of opportunities for productivity, operations, decision-making, and services for your customers. #GenerativeAI, in particular, has been gaining traction in the corporate world for its ability to generate human-like language, which can be used in various use cases, such as chatbots, content creation, and customer service.?

However, integrating AI into corporate policies requires careful execution to ensure the benefits outweigh the #risks. In this blog, I will explore five considerations when looking at using Generative AI / LLMs in your business and what to consider when developing corporate policies around their usage.

#1 Data Security and Privacy

Data security and privacy means keeping important information safe and not letting anyone who shouldn't have access to it, see it. When we use LLMs to help us at work, these models learn from lots and lots of data inputs to work properly. For a model like Open AI chat GPT, it is collecting vast amounts of data from the open internet, but also all the folks who use it on a daily basis.?

The risk for businesses is this - sometimes the data inputted into the LLM might have confidential or private information in it. Thus, it is important to ensure the LLM only learns what it's supposed to and doesn't accidentally share information it shouldn't. It's like keeping a secret safe - we only tell the people who need to know and make sure nobody else finds out.

Any easy way to solve this is to employ data classification, where you classify your data based on the level of sensitivity or confidentiality. This classification will help determine which data can be used to train LLMs and which data should not be used.

#2 Legal Implications

Generative AI can be used for both positive and negative purposes. Therefore, it's essential to ensure that the use of LLMs aligns with your governmental regulations. For example, in Canada, the use of AI for employment decisions is governed by the Canadian Human Rights Act and various provincial human rights laws, which prohibit discrimination based on protected characteristics such as race, gender, and disability.?

For example, individual income often correlates with the prohibited grounds, such as race and gender, but income is also relevant to decisions or recommendations related to credit. The challenge, in this instance, is to ensure that a system does not use proxies for race or gender as indicators of creditworthiness. For example, if the system amplifies the underlying correlation or produces unfair results for specific individuals based on the prohibited grounds, this would not be considered justified.

A way to solve this problem is to consider adopting data anonymization, whereby you anonymize any personal or sensitive data used to train LLMs. This means removing any identifiable information from the data before it is used to train the LLMs. Further you can have a data destruction policy in place for the secure destruction of any data used to train LLMs that is no longer required.?

#3 Open vs. Closed Models

A closed LLM model is trained only on company data and is not accessible to the public. This approach reduces the risk of disclosing sensitive or confidential information inadvertently. Additionally, a closed LLM model can be customized to meet the specific needs of your employees and can provide better accuracy and efficiency. The cost to run these models would be far cheaper than an open model as well.

#4 Integration into existing stack

It is inevitable that these tools will be built to integrate with your other apps, such as CRM, marketing automation platforms, and communication tools. These integrations allow for a seamless and efficient workflow, reducing manual data entry and the likelihood of errors.?

Before integrating AI and LLMs tools within your current environment, ensure your consider and ask:?

• What data am I making available and to whom (inside and outside my organization)??
• Is the data sync bidirectional or unidirectional??
• Will we need additional resources to ensure the integration is adequately set up or is the integration automatic (pre built)??
• Training and maintaining LLMs requires specialized skills and expertise, is there a cost and availability of such resources associated with this??
• LLMs require continuous training to maintain their effectiveness, do we need to allocate resources for ongoing training and maintenance?

#5 Involve X-functional stakeholders

It's important to involve stakeholders across different departments, such as HR, legal, sales, marketing, business development, operations, and IT, when creating a corporate IT policy for using AI and LLM tools. Each group may have different perspectives on the use cases and consequences.?

Additionally, the policy should be a living document that's updated regularly based on emerging use cases, market conditions, and new developments. It's also important to have all stakeholders sign the policy or incorporate it into an existing policy manual signed by the CEO to show that it's an essential part of the company's overall governance approach for technology.

Integrating generative AI and LLMs into business workflows can be a game-changer for businesses, but it requires planning. By weighing the risks against the benefits, companies can take advantage of LLMs to enhance their operations, make better decisions, and foster innovation. With the right strategy and approach, LLMs can take your company to the next level.