When Artificial Intelligence Is No Longer a Government Assistant but Part of Its Work

For years, the question remained: How can governments use artificial intelligence? But that question is beginning to lose its precision. The most important development today lies not in an employee’s ability to ask an intelligent system for a faster answer, but in the shift of some systems from providing answers to executing chains of tasks and taking steps within government work itself.

The UAE has set a goal to transform at least half of federal government services and operations into models based on agentic AI within two years, alongside training 80,000 federal employees. In September, the government launched the Cabinet AI Advisor system, which operates through 32 specialized assistants to analyze policies, legislation, and strategies, evaluate their various impacts, and provide recommendations.

These numbers are important, but what matters more to me is what lies behind them. We are facing a change in the nature of the relationship between technology and administration. When AI provided information or a summary, the distance remained clear between the tool and the decision-maker. Yet when a system can track a process, retrieve data, coordinate steps, propose a decision, and perhaps execute part of it, we approach a new zone: the zone of operational authority.

Here begins the question that must accompany any transformation of this kind: When does the system work alone, and when must it stop and hand the matter back to a human?

In operations rooms, the quality of information alone is not enough. The right information might reach the wrong person, arrive at the wrong time, or reach someone who lacks the authority to act. Therefore, the efficiency of an operations room is measured not just by the number of screens or the speed of systems, but by the clarity of both the information path and the authority path.

This same logic must carry over to government artificial intelligence.

If a system is capable of executing a clear, routine task, it naturally performs it quickly. But if an exceptional case arises, data conflicts, or a decision exceeds its usual impact, the response must not be just another automated attempt. There must be a clear threshold where the decision shifts to the appropriate human.

I call this “decision security.” Cybersecurity protects systems and data from breaches and tampering, whereas decision security concerns what happens after information arrives: How was it verified? How was it interpreted? Who has the authority to act upon it? And when must it be escalated?

This is not an issue for the UAE alone. Egypt, given the size of its administrative apparatus, the number of people dealing with public services, and the diversity of its sectors, has a strong reason to follow these transformations. The larger the state, the greater the value of automation, but the greater the importance of distinguishing between a routine transaction and a case that requires human judgment.

The future will not be a government without employees, as some simplified discourses about artificial intelligence imagine. It is more likely that we are heading toward a government where the role of the employee changes. The value of manually transferring data between systems will decrease, while the value of verification, supervision, handling exceptions, and bearing responsibility for complex decisions will rise.

Here, training humans becomes part of the artificial intelligence infrastructure itself. An advanced system operating within an organization unprepared to manage it may increase speed without increasing quality. Meanwhile, an organization that redesigns authorities, training, and escalation paths in parallel with technology is the one that can transform artificial intelligence from a dazzling tool into a genuine institutional capability.

The coming challenge for governments will not be proving that artificial intelligence can work. That phase is advancing rapidly. The tougher challenge is proving that the institution knows when to let it work, when to limit its authority, and when to return the decision to a human.

Only then does artificial intelligence become part of a smarter government, not merely a faster one.

Best regards,

Abdullah Saeed Al Habsi (A.S. Alhebsi)
Writer and researcher specializing in security, heritage, risk management, and operations rooms

Published by Hany Sallam

Hany Sallam, Founder and Editor-in-Chief of Scientific Magazine Ahram
An independent Arabic science magazine operating since 2008, dedicated to simplifying science, technology, space, natural phenomena and scientific stories for Arabic readers. The magazine includes sections such as “Zodiac Group” and “Stories and Tales”, alongside science topics that combine knowledge with a spirit of discovery. It operates independently and is not affiliated with Al-Ahram Press Foundation or any governmental body.