Licensing AI: If We Trust It With Lives, We Should Certify It Like Professionals
The Illusion of Passing Exams
Headlines love a milestone. Every few weeks we see them, loud and proud: AI passes the bar exam. AI clears the medical boards. AI performs at pharmacist standards. AI is at PhD level. AI is now smarter than humans.
These stories capture attention because they make AI look as if it has crossed into professional territory. Yet they miss an obvious truth: passing a test is not the same as being licensed.
Here is my case. I appreciate the progress of AI and I respect the milestones it has reached, but we should not confuse success on exams with readiness to carry responsibility in the real world.
What Licensing Really Means
When we license professionals for specific fields, we are entering into a social contract. A license tells the public that someone has been tested, vetted, supervised, and is accountable for their actions. It is not only about passing knowledge checks. It is about demonstrating ethical behavior, sound judgment under pressure, and the ability to know when to defer to others.
If we expect AI to operate in law, medicine, pharmacy, or education, then it must be treated the same way. It must be certified in the eyes of the public, just as we certify human professionals.
Licensing AI means certifying it for specific, limited tasks with clear boundaries. A system might be licensed to perform legal research but not to advise clients. It might be licensed to check drug interactions but not to recommend treatments. It might be licensed to provide students with practice quizzes but not to grade high-stakes exams.
This approach recognizes both the power and the limits of AI. It ensures that when AI is placed in high-stakes environments, the public can trust that it has earned its place, and that human oversight remains where it matters most.
Building a Tiered Path to Trust
Professional licensing is not a single test. It is a pathway. A long one. I know this from my own experience as a social worker, case manager, advisor, trainer, and project manager. You do not simply pass an exam and walk away with full responsibility. You are guided, tested in practice, and held accountable over time.
We should build the same path for AI, at least in the professions where certification is required for humans.
The first step is knowledge certification. AI must prove that it can answer domain-specific questions with accuracy, calibration, and transparency. Passing an exam is not enough unless the model can also show uncertainty and explain its reasoning.
The second step is simulation-based certification. AI should be tested in realistic scenarios where the stakes are high and the answers are not straightforward. In medicine, that means standardized patients with complex symptoms. In law, it means mock cases where confidentiality and ethics matter as much as factual recall. In education, it means classroom vignettes where fairness is tested as rigorously as subject knowledge.
The final step is operational and ethical certification. This involves risk management plans, independent audits, incident reporting, and recertification on a regular schedule. AI changes quickly. Without continuous oversight, today’s safe model could become tomorrow’s liability.
This tiered approach is not about bureaucracy that slows innovation. It is about applying the same discipline we already demand of human professionals. The point is not to create obstacles for the sake of it. It is to build structures that make innovation safe, trustworthy, and sustainable.
Guardrails Across Professions
The core principles of licensing apply everywhere, but the lines of responsibility differ by field.
In law, AI can support document review and e-discovery, but it must never give unsupervised legal advice. In medicine, AI can suggest differentials or cite guidelines, but a doctor must always make the final decision. In pharmacy, AI can check interactions and dosage ranges, but the pharmacist must remain responsible for counseling the patient. In education, AI can provide practice feedback, but teachers must remain at the center of grading and assessment.
Guardrails like these do not stifle innovation. They make innovation usable. They bring clarity to what AI is for and what it is not. They give professionals the confidence to adopt tools without feeling that they are surrendering their judgment or their duty of care.
Critics will say that licensing AI risks slowing innovation. In reality, it does the opposite. Licensing builds the trust that allows innovation to scale. Hospitals, schools, and courts will not adopt AI at scale without credible assurance that it is safe and accountable. A certification framework provides that assurance.
Moreover, licensing AI gives developers incentives to focus on quality, transparency, and accountability rather than on speed alone. Shared safety benchmarks and staged permissions would allow both startups and large firms to prove competence in a fair and transparent way. The competition would not be over hype or marketing, but over who can earn trust.
A Final Reflection
What I have shared here is an argument, not a final verdict. I am still exploring these questions, testing them against my own experience, and searching for the right balance between innovation and safety. I may be right in some parts and wrong in others, but the point is to return to first principles and ask: how do we build trust in systems that are already stepping into roles once reserved for licensed professionals?
I do not claim to have the full answer. What I hope to do is to encourage you to think with me, to question what certification should mean for AI, and to imagine how we can shape this moment responsibly. If we approach it with honesty and care, we may discover not only how to license AI, but how to renew our own sense of accountability in the professions that matter most.
Passing exams may impress us. Licensing is what ultimately protects us.
Ali Al Mokdad
