The Citizen Developer

Another Layer of Abstraction: Citizen AI

• By Howard M. Cohen
• August 23, 2024

In my last column I asked the question, "Where Did Low Code/No Code Go?" suggesting that platforms enabling business people to move icons around a screen to create applications may have a very short half-life, giving way quickly to AI-powered platforms.

As I anticipated, the advance seems to be accelerating. Already more and more analysts are using the term "Citizen AI" instead of Citizen Developer. Some may be referring to AI user interfaces that allow non-technical people to take advantage of artificial intelligence. Others, however, already share the vision of business users creating new and often very sophisticated applications by simply requesting functionalities from an AI/ML engine.

Requirement: An Effective Layer of Abstraction

One of the bellwether documents we all live by, the Definition of Cloud Computing, from the National Institute of Standards and Technology (NIST Special Publication 800-145) discusses how the cloud model provides a layer of abstraction that provides "transparency for both the provider and consumer of the utilized service."

The objective of such an abstraction layer is to enable users to do their work without concern for where their resources are coming from, or how to access them.

Citizen AI users will require, and actually be defined by, a similar layer of abstraction between themselves and the various system elements that provide artificial intelligence, machine learning, and other technologies. Put more simply, they must be able to use AI/ML without knowing anything about how AI/ML works.

Suggests a New Organizational Structure

The introduction of low-code/no-code (LCNC) applications and the citizen developer brought substantial change in the organizational structure of many organizations. Shifting more of the application development tasks to non-technical users reduced the need for scarce, difficult-to-find technologists, including coders and architects. Many IS shops became focused on the "last mile" of application development, fine-tuning what the citizen developers had produced to create a production version of the software.

Citizen AI will take this division of labor even further. Augmented by well-trained large language models (LLM), the capabilities of citizen developers will reduce the need for fine-tuning by coders to a bare minimum. However, it will also create a need for more systems experts who can manage the many disparate components of advance AI/ML systems.

Each of the major cloud service providers and many others have introduced their own AI/ML implementations. Each consists of many modules, applications, engines, and subsystems, each of which enables various aspects of the platform. To maintain the desired layer of abstraction, all of this must be handled by technology experts, not Citizen Developers or Citizen AI users. Citizens should only need to interface with the UI to make their requests and specify their desired business functions.

What may begin to fade away is the concept of an Information Systems team as application development writ large becomes the province of the citizen developers, augmented by their AI/ML tools.

Impact on User Acceptance Testing

These developments may also affect User Acceptance Testing (UAT) profoundly. It has become common knowledge that AI systems can "hallucinate" and even make things up. In general use, these false results are often relatively harmless. But in the course of creating production applications, they can create enormous delays. These developments will almost certainly require the creation of a new, higher level of testing that will expose falsehoods that may have found their way into the code generated by the AI.

Relationship Between People and Processors

Perhaps the most important question that arises from these developments (and the hardest to answer) is What will be the impact of Citizen AI on the fundamental relationship between businesspeople and the information assets they use"

It's reasonable to assume that the time between concept and working software will be significantly truncated. A businessperson with an idea will be able to realize that idea with computer support potentially within the same day. Libraries of modular sub-programs will very likely emerge that will make these development processes even faster.

Hints from a Visionary

I usually avoid going here, but I can't help myself. As long as we're considering potential futures, let's note for a moment the remarkable prescience of Gene Roddenberry, creator of the incredibly successful "Star Trek" franchise, who imagined versions of technologies that are commonplace today, from cell phones to virtual assistants. The crews of his starships included a wide range of roles, everything from bridge officers to scientists, engineers, and medical personnel—but no programmers, software developers, or any real coders. Most interactions with the ship's computers in Roddenberry's vision involved a voice interface—humans talking to intelligence machines and systems.
 
Some of the things he imagined have already been realized; others are in development now or awaiting the next technological surge. But I think it’s a safe bet that computer/human interaction will continue to evolve radically.
 
It may turn out that the ultimate function of artificial intelligence and machine learning will be to enable machines that can understand us well enough to program themselves to do what we ask in a conversational mode. The interaction will become far more casual, far more flexible, and far more powerful.
 
In other words, we may someday find that Citizen AI is the only kind of AI.