EdCircuit Staff 
Blackboard founder Stephen Gilfus believes most of the world is building AI literacy the wrong way: by reading about artificial intelligence instead of using it. In a June essay on LinkedIn, “AI Is an Automobile,” he compared AI to the arrival of the car. People did not need a lesson on combustion engines to understand what a car could do; they watched one pass at a speed no runner, cyclist or horse could match. AI offers the same kind of jump in capability, he argues, with one difference that matters: it has no physical form. It does not drive past anyone’s house, so many people have heard about it, read about it or watched a demonstration without ever sitting in the driver’s seat. His conclusion is short: “No one truly understands the automobile from the sidewalk.”
Gilfus has watched a shift like this before. He co-founded Blackboard in 1997, when many educators doubted that teaching could happen online at all, and the learning management system it helped create is now ordinary in schools, universities and workplaces. Today he serves on the board of ISACA, the global association for IT audit, cybersecurity, risk and governance professionals, where he chairs its Innovation and Technology Committee. He also chairs edCircuit’s board.
The AI literacy gap, in numbers
The largest recent survey on the subject suggests Gilfus is describing two gaps, not one. In Trust, attitudes and use of artificial intelligence: A global study 2025, the University of Melbourne and KPMG surveyed more than 48,000 people in 47 countries between November 2024 and January 2025. Two in three said they intentionally use AI on a regular basis. Yet 61 percent had received no AI training, and half reported limited knowledge of AI.
The first gap is the one in the essay: about a third of people still are not using AI regularly, so they are judging it from the sidewalk. The second is less obvious and arguably more urgent. Many of the people who are driving taught themselves on the highway. At work, 58 percent of employees said they use AI regularly, and the study’s full report found that many rely on AI output without evaluating its accuracy (66 percent) and that 56 percent have made mistakes in their work because of AI.
Microsoft and LinkedIn found the same pattern in their 2024 Work Trend Index, a survey of 31,000 people in 31 countries. Three in four knowledge workers said they use generative AI at work, 78 percent of those users brought their own AI tools, and only 39 percent of people using AI at work had received AI training from their company.
How the world learned to drive
The car did not become safe and useful on the day it was invented. Over time, societies built the system that lets ordinary people drive: driving schools and instructors, practice in quiet places before busy roads, a learner’s permit before a license, a test of real skill, road signs and lane markings, traffic laws, and insurance for when things go wrong. Nobody suggested that people stop driving until all of it existed, and nobody suggested that people drive without any of it.
AI literacy needs the same system, and each piece has a counterpart:
- Instructors. People learn fastest with someone who has driven before: a colleague, a trainer or a teacher who shows how a tool is used on real work.
- Quiet streets first. Low-stakes tasks, such as drafting, summarizing or brainstorming, are better first drives than decisions about people, money or health.
- A permit before a license. Organizations can start with pilots for a defined group, a time limit and a review, before rolling a tool out to everyone.
- A test of real skill. Competence means checking output, spotting errors and knowing when not to use AI, not only writing prompts.
- Road signs. Rules work best where people meet them: inside the tools and workflows they already use.
- Insurance. Logs, human review and clear accountability exist for the day something goes wrong.
How to use AI: your first drives
For anyone still on the sidewalk, Gilfus’s advice is to get in the car, and the driving comparison suggests how:
- Pick a real task. Choose something you do every week, such as an email, a report outline or a meeting summary, so you can judge the result against work you know.
- Keep it low stakes. Start with tasks where a mistake is cheap, and keep confidential, personal or customer data out of tools your employer has not approved.
- Check your mirrors. Read every answer as a draft. Verify facts, figures and quotations against a source before you use them.
- Drive with a passenger. Use the tools and rules your organization provides, and compare notes with colleagues who use them too.
- Log your hours. Keep a short note of what worked and what did not. The time saved, and the errors caught, are what turn use into skill.
- Move to busier roads gradually. Take on harder tasks once you can reliably catch the tool’s mistakes.
Rules of the road around the world
Traffic law varies by country, but every driver meets some version of it. AI now has its rules of the road too, and several apply far beyond the place they were written.
The European Union’s AI Act makes AI literacy a legal duty. Article 4, in force since February 2, 2025, requires providers and deployers of AI systems to take measures that support the AI literacy of their staff and others using AI on their behalf. An amendment in July 2026 made clear that no specific level of literacy is mandated, and the European Commission’s AI literacy questions and answers says no certificate is required. The duty still has reach: it applies to organizations inside and outside the EU whose AI systems are used in the Union, and national authorities begin supervising and enforcing it in August 2026. The Commission’s own example is a company whose staff use ChatGPT to write advertising copy; those staff should be told about specific risks such as hallucination.
Outside the law, two widely used frameworks set shared expectations. The U.S. National Institute of Standards and Technology released its AI Risk Management Framework on January 26, 2023, for voluntary use by any organization; it is organized around four functions: govern, map, measure and manage. ISO/IEC 42001, published in December 2023, is what ISO calls the world’s first AI management system standard, and organizations can be certified against it. ISACA, the association on whose board Gilfus serves, has added AI credentials for professionals, including Advanced in AI Audit, Advanced in AI Security Management and Advanced in AI Risk.
Gilfus’s own test for any of these rules is whether people meet them at the moment they decide. In a field note on his website, “The governance gap in AI-assisted learning,” he writes that policies sitting in documents do not govern behavior: “If the rule isn’t in the workflow, it isn’t real.” A seatbelt chime works because it is in the car, not in the owner’s manual.
Drivers and the traffic system
The comparison also helps separate two kinds of AI that are often sold as one. In another field note, “AI Tutors vs. AI Orchestrators,” Gilfus distinguishes AI that works with one person, such as a tutor or an assistant, from AI that coordinates the systems around many people: identity, permissions, data, policy filters and a record of what happened. The first is the car and its driver. The second is closer to the traffic system of signals, lanes and records that lets many drivers share the road.
His summary is written for education, “Tutors answer to teachers; orchestrators answer to institutions,” but the point travels to any organization. The person using an assistant is the right judge of whether it helps them work. The organization’s technology, legal and security staff are the right judges of whether the coordinating layer protects data and keeps an audit trail. Buying the two as one product tends to measure each against the wrong standard.
Schools already run driver’s education
Education has a head start on this problem, because schools have taught people to drive for generations. The same global study found that 83 percent of students regularly use AI in their studies, while only half said their school or university has policies, resources or training to support responsible use. The driver’s-education model suggests what closes that gap: AI training for teachers that includes hands-on time with approved tools, supervised first use inside lessons, and assessment that rewards students for showing how they worked. edCircuit’s guide to AI in education covers the rest.
Where the comparison stops
No analogy is complete. A car’s failures are usually visible; an AI tool’s errors can be fluent and wrong at the same time, which is why checking output is the core driving skill. A car does not keep a record of what its driver says, while an AI tool may hold personal, customer or student data, so organizations need to decide what may be entered and where it may be processed.
Access is uneven, too. The 2025 study found higher AI use, training and knowledge in emerging economies than in advanced ones: 80 percent regular use against 58 percent, and 50 percent with AI training against 32 percent. AI literacy is not following the usual map of wealth, and organizations in advanced economies cannot assume their people are ahead.
Gilfus does not argue that experience should replace caution. In his view, AI is neither something to fear nor something to embrace blindly, and he pairs experimentation with what he calls pragmatic governance. Driver’s education makes the same bargain: it gets new drivers on the road sooner because it takes the risks seriously.
A one-year plan for any organization
Gilfus’s governance note sets out a one-year sequence written for schools and colleges. It translates to any organization that wants its people driving with skill:
- First three months: take inventory and set guardrails. List every AI use already in place, including tools people brought themselves, and the data each touches. Pause new uses in high-stakes areas until vendors can explain their models and keep logs, while low-stakes experiments continue.
- Months three to six: run guided pilots. Choose a few teams with different kinds of work and people who volunteer. State the role AI plays in each task, and report results to leadership monthly on one page.
- Months six to nine: build the rules into the tools. Turn interim guidance into templates, settings and contract terms, and expand only the uses the pilots supported.
- Months nine to twelve: make it routine. Move proven uses into regular service, review data flows and model changes each quarter, and record the training people have had.
None of this asks anyone to move faster than they are comfortable with. It asks that the people using AI, and the people making rules about it, do so with their hands on the wheel, not from the sidewalk.
Stephen Gilfus is chairman of edCircuit’s board and a director of ISACA. His field notes on AI governance and learning are at stephengilfus.com, and his original essay appears on LinkedIn.
Related reading
Subscribe to edCircuit to stay up to date on all of our shows, podcasts, news, and thought leadership articles.
More on Artificial Intelligence
