How government adoption of GenAI is different

Governments will follow more than one path to adopting GenAI (image courtesy of Google’s Gemini)

Governments are traditionally slow to adopt new technology innovations. For example:

• Cloud created a debate about security and most jurisdictions were slow to adopt
• Traditional AI (i.e. machine learning) is hard to hire for and implement, including responsible use

But the adoption of GenAI is different. It is following at least three internal adoption curves and one by proxy. Each curve differs in:

1. degree of control over adoption,
2. visibility and control of ongoing use, and
3. ability to manage and mitigate risk.

Adoption curve 1: Individual use

Individual use is already here. Unlike many other technologies, individuals can use GenAI right now. It is more akin to a new search tool in terms of access and adoption (and is even embedded in the standard search tools). What this means is that:

• You can’t stop use (though security teams will try the usual web blocks)
• You can’t control how it’s used and you may have no visibility into use
• With proper training and education, you can mitigate some risks

Adoption curve 2: Embedded use

Embedded use has already started and will grow rapidly. Embedded use is the deployment of GenAI tools into existing tools and services. Examples include Microsoft’s suite of tools and Adobe’s embedded reader.

Governments that have already procured and use these tools will have GenAI embedded. Instead of a new product, GenAI is a set of features or feature enhancements in existing products. What this means is that:

• You can’t stop adoption unless service providers provide the option to “turn off” the feature or put it behind a paywall
• You have more visibility into when/if/how it is used but not necessarily control
• Your risk mitigation will be dependent on the service provider

Adoption curve 3: Systems use

Systems use will most closely follow the traditional government adoption curve. Systems use is when governments buy software and systems via traditional RFPs. Governments may buy a GenAI tool, a tool with GenAI embedded, or develop a GenAI tool based on their own content and data. What this means is that:

• You can control adoption in the initial purchase, but then control will vary depending on the nature of the tool
• You will have the most control and visibility into how this is used at least initially
• You can most closely follow your standard procurement and risk mitigation procedures

Adoption curve 4: Proxy use

A final adoption curve is via proxy. Governments contract with or delegate many things. This can be to companies, consultants, other levels of government, and nonprofits.

But governments do not yet have contractual terms to manage GenAI use. In practice, it will be difficult to gain insight or control in its use. Crafting GenAI contract terms will be fraught with challenges around visibility and enforcement. For comparison, many governments still struggle to address cloud concerns via terms. GenAI terms will be much more complex.

For each adoption curve, governments will need to tailor their approach to policy, training, procurement, and risk management.

***

Acknowledgements: To edit this post for readability, I used mostly Hemingway Editor and a bit of Chat GPT 3.5. I used Google’s Gemini to create the image. Special thanks to Blake Valenta for reading and providing feedback.