Key Takeaways
- State DMVs are already deploying AI for practical, low‑risk tasks such as automatic photo‑background removal and natural‑language search, cutting transaction times from minutes to seconds.
- Government agencies adopt generative AI cautiously, favoring deterministic, well‑defined use cases over open‑ended content generation to avoid hallucinations and maintain reliability.
- AI‑driven modernization helps agencies replace legacy systems faster by scanning, analyzing, and extracting insights from massive volumes of paper‑based documents and legal texts.
- The primary metric of success in the public sector is improved citizen experience—measured in “citizen minutes saved”—rather than profit or ROI.
- Robust guardrails, policy‑in‑code frameworks, and cautious implementation help mitigate AI hallucinations, ensuring accurate and trustworthy outputs in high‑stakes government services.
- Partnerships with technology firms like Kyndryl enable governments to keep pace with rapidly evolving AI models and expand AI’s role across entire modernization stacks.
The Perception of Government Tech Adoption
Brian Shell, a senior partner at Kyndryl, acknowledges that the public often views government as slow to embrace new technology. He agrees that this caution is appropriate: durable institutions must vet innovations before wide‑scale rollout. While governments are not moving at the speed of tech startups, many states are already piloting and scaling generative AI applications, showing a steady, thoughtful uptake rather than outright resistance.
AI as an Inevitable Part of the Technology Landscape
Shell likens today’s AI proliferation to the earlier mantra “every business is an IT business.” Just as IT became indispensable three decades ago, AI will soon permeate every sector that interacts with technology, including public services. The pervasiveness of AI means that no area of society can afford to ignore it, and governments are beginning to integrate it into their core operations.
Deterministic Government Meets Probabilistic AI
Government work is rooted in deterministic processes—clear rules, defined outcomes, and little tolerance for guesswork. Generative AI, by contrast, is inherently probabilistic. Shell observes that current government deployments stay on the deterministic side, using AI for specific, tactical problems rather than relying on large language models to create new content. This approach aligns with the need for predictability while still gaining AI’s efficiency benefits.
Small‑Scale Wins: Photo Background Removal
One of the most tangible AI use cases in DMVs is automatic background removal for license photos. Previously, citizens had to step to a dedicated station with a high‑resolution camera and a blue backdrop, a process taking three to five minutes. Now, a camera at each desk captures the image, and AI instantly strips the background, checks for glasses or head tilt, and delivers a compliant photo in five to ten seconds. This modest change has already saved countless minutes per transaction.
Scaling Up: Natural Language Search
Beyond photo processing, DMVs are employing AI to bridge the language gap between bureaucratic jargon and everyday citizen speech. A natural‑language translator lets users type plain questions—such as “How do I renew my license?”—and the system maps the query to the correct internal code or form. Citizens retain the familiar terminology they think with, while the agency keeps its precise internal references, resulting in high accuracy and reduced frustration.
Document Processing Leapfrogs OCR
Traditional optical character recognition (OCR) has been outpaced by agentic AI models capable of scanning, interpreting, and validating paper documents. States are piloting these models for tasks like title transfers, where AI quickly extracts data, verifies legitimacy, confirms required fields are filled, and flags inconsistencies. The shift from manual data entry to AI‑driven processing dramatically speeds up workflows while maintaining high accuracy.
Citizen Experience as the Core Metric
Shell emphasizes that the public sector’s success metric differs from private‑sector ROI. Governments measure value by how well they serve citizens, not by profit. A key indicator Shell cites is “citizen minutes saved.” For a state processing three million transactions annually, saving four minutes per citizen translates to 12 million minutes returned to the public each year—time that would otherwise be lost in bureaucratic queues. This focus on reducing wasted time reflects a renewed commitment to public service.
Mitigating AI Hallucinations
Given the risks of hallucinations—where AI generates false or nonsensical information—Shell outlines a cautious, framework‑based approach. By embedding policies directly into the code, agencies constrain AI agents to operate within defined boundaries, much like giving an artist a canvas with clear edges. This method reduces unpredictable outputs while still allowing AI to perform its useful functions, leveraging the government’s natural inclination toward caution as a strength.
Expanding AI’s Role Across the Modernization Stack
AI capabilities have grown far beyond early narrow tasks like background removal. Modern models can ingest massive corpora—such as state motor‑vehicle law books—compare them to existing systems, and extract actionable business insights. This enables end‑to‑end modernization: AI assists with data ingestion, encoding, process redesign, and even agent management. Shell notes that the technology now supports building and orchestrating AI agents across entire workflows, opening possibilities that were previously unattainable.
Looking Ahead: Stories of Improvement
Shell concludes with optimism about the next decade of AI in government. He points to a specific DMV serving 7.2 million people where the average door‑to‑door time—covering check‑in, transaction, and exit—is 18 minutes. While not fully AI‑driven yet, ongoing efforts aim to push that figure lower, returning more time to citizens. These improvements, he argues, will become compelling narratives of how AI can revitalize public service when applied thoughtfully and responsibly.
This summary reflects the insights shared by Brian Shell of Kyndryl in his interview about AI adoption at state DMVs and other government agencies.