To Build or Not to Build? Factors that Lead to Non-Development or Abandonment of AI Systems
This paper investigates the underexplored factors leading to the non-development or abandonment of AI systems by synthesizing a scoping review and empirical case data to reveal that, beyond ethical concerns, diverse practical levers such as resource constraints, legal issues, and organizational dynamics significantly influence these critical pre-deployment decisions.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the world of Artificial Intelligence (AI) as a massive, bustling construction site. Usually, when we hear about AI, the news is all about the buildings that got finished, the skyscrapers that went up, and how people are living in them. We talk about the "use" and the "impact" of these finished structures.
But this paper asks a different question: What about all the construction sites that were abandoned? What about the half-built houses, the blueprints that were crumpled up, and the cranes that were turned off before the first brick was even laid?
The authors, Shreya Chappidi and Jatinder Singh, argue that we are ignoring a huge part of the story. They want to understand why organizations decide not to build an AI system, or why they stop building one halfway through.
Here is a simple breakdown of their findings, using some everyday analogies.
The Big Misunderstanding
Most people think companies stop building AI because they realize, "Oh no, this is unethical!" or "This is dangerous!"
The paper says: That's only part of the story.
While ethical concerns (like bias or privacy) are important, they are often just the tip of the iceberg. In the real world, companies often abandon AI projects for much more boring, practical reasons—like running out of money, running out of time, or realizing the math just doesn't work.
The Six Reasons Construction Stops (The Taxonomy)
The researchers looked at hundreds of reports and talked to engineers to create a "menu" of reasons why AI projects get cancelled. They found six main categories:
Ethical Concerns (The "Conscience" Check):
- The Analogy: The architect realizes the house is being built on a graveyard or will block a neighbor's sunlight.
- The Reality: The team worries the AI might discriminate against people, invade privacy, or be used for bad things (like weapons).
Stakeholder Feedback (The "Town Hall" Meeting):
- The Analogy: The neighbors, the workers, or the future residents come together and say, "We don't want this building here," or "We are going on strike if you build it."
- The Reality: Employees, customers, or activists push back. They might protest, refuse to use the tool, or threaten to quit, forcing the company to stop.
Development Lifecycle Challenges (The "Blueprint" Problems):
- The Analogy: You try to build a bridge, but you realize the river is too wide, the steel is too weak, or you can't figure out how to pour the concrete.
- The Reality: The data is messy or missing, the model doesn't learn correctly, or the math is too hard to solve. The project hits a technical wall that can't be climbed.
Organizational Dynamics (The "Office Politics"):
- The Analogy: The boss changes their mind, the department budget gets cut, or the team fighting over who is in charge can't agree on what to build.
- The Reality: The company's strategy shifts, leadership loses interest, or different teams have conflicting goals. The project just loses its internal support.
Resource Constraints (The "Empty Wallet"):
- The Analogy: You have a great plan for a mansion, but you only have enough money for a shed. Or, you don't have the right tools or the right workers.
- The Reality: It's too expensive to build or maintain. They don't have enough powerful computers (GPUs), enough data, or enough skilled engineers. Sometimes, it's just cheaper to buy a tool from someone else than to build it themselves.
Legal/Regulatory Concerns (The "Permit" Issues):
- The Analogy: The city zoning board says, "You can't build here," or "You need a permit we can't get."
- The Reality: New laws (like the EU AI Act) or data privacy rules (like GDPR) make the project too risky or legally impossible to finish.
What the Data Actually Showed
The researchers looked at two different groups of stories to see what really happens:
Group 1: The "News" Stories (Incident Databases)
- These are the famous failures that made headlines (like a facial recognition system that got shut down because it was racist).
- Finding: In these public stories, Ethical Concerns and Stakeholder Feedback are the big stars. It makes sense because these are the stories that get reported in the news.
Group 2: The "Backstage" Stories (Survey of Practitioners)
- These are the quiet failures that happened inside companies and never made the news.
- Finding: Here, the story is totally different. The most common reasons for stopping were Technical Challenges (it didn't work), Resource Constraints (too expensive), and Organizational Dynamics (management changed its mind).
- Key Insight: Many projects were killed before they ever got to the public stage because they were too hard, too expensive, or simply not a good fit for the company.
The "Ghost" Projects
The paper also notes that sometimes, a project gets cancelled, but then comes back to life later.
- The Analogy: A builder stops work on a house because of rain. Two years later, the sun comes out, and they start digging the foundation again.
- The Reality: Companies often pause a project, wait for better technology or cheaper costs, and then try to build it again. Or, they might cancel a specific feature (like facial recognition) but keep the rest of the system.
The Main Takeaway
The authors are saying to the AI community: "Stop only looking at the finished buildings and the ethical scandals."
To truly understand how to make AI safe and responsible, we need to understand the construction phase. We need to help companies realize that deciding not to build is a valid, smart, and sometimes necessary choice.
They suggest that we need better tools and checklists to help organizations ask:
- "Do we actually have the money and skills to finish this?"
- "Is the technical problem solvable?"
- "Is this the right time to build this?"
If we can make it easier for companies to say "No, we shouldn't build this" for practical reasons, we might prevent a lot of future problems before they even start.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.