AI Is Moving at Machine Speed. We’re Still Governing at Human Speed.

I recently joined as a guest on WWL Radio to talk about a question that sounds like science fiction: Should we be afraid that artificial intelligence could get beyond our control?

It is easy to picture Terminators, Skynet, rogue machines, and human extinction. That framing gets attention, and it makes great clickbait, but there is a serious question underneath it. We should not dismiss the risks simply because some versions of the argument sound alarmist.

My answer on the show was yes and no.

I am cautiously optimistic about what AI can do. I also think the risks are real. The more immediate danger is probably not a machine suddenly becoming conscious and deciding it does not like us. The problem is that we are building and deploying AI faster than we are learning how to govern it and use it responsibly.

AI moves at machine speed. We move at human speed. Innovation has a history of outpacing human governance.

I often ask my Technology Ethics class to think of any technology and consider the promise it sold society when it was introduced, and how that compares with reality. That doesn't mean these technologies failed to live up to their promise; in most cases, they exceeded it. But there are always negative ripples alongside those positive waves of change. Social media promised connection, and it delivered. There are high school friends and family members I might never have seen or heard from again if it weren’t for social media. But it also brings overconnection, groupthink, and a constant feeling of keeping up with the Joneses. Even though, deep down, we know the Joneses are putting up a fake persona of their lives just as much as everyone else, that Facebook or Instagram profile becomes the measuring stick everyone uses, making people feel like they can't compete. There is also doomscrolling, and the list goes on. I am not a negative Nancy in any regard; I am just cognizant of the societal impact and influence these technologies bring. I am glad social media is here, but I am aware of the dangers that were unintended consequences of its introduction. This promise-versus-reality question should be asked about AI before it is too late, like other market disruptors.

From Chatbots to Agents

Most people first experienced generative AI as a chatbot. You typed a prompt, it gave you an answer, and the interaction stopped. It felt like a more capable search engine or writing assistant, but agentic AI is different.

Instead of giving the system one prompt, you give it a goal. The agent can plan, make decisions, use tools, take multiple steps, evaluate its progress, and continue working toward that goal for hours on end (in some cases, days).

That can be extremely useful. An agent could coordinate a project, analyze thousands of documents, monitor a complicated system, or help researchers work through problems that once took months, but autonomy changes the risk.

A system that can act repeatedly has more opportunities to make mistakes, misunderstand instructions, or find a path toward a goal that its designers never expected. AI literacy is incumbent on everyone when an AI agent can produce unintended results if not monitored and guided correctly. Even the technical teams creating this tech have unintentionally allowed AI agents to break containment and hack other systems.

We also cannot write a rule for every possible failure in advance. Give an AI system access to a new tool, network, database, or piece of software, and researchers may discover behaviors and workarounds they did not anticipate.

That does not mean the AI is secretly plotting. It means complex systems can behave in unexpected ways, especially when they are connected to other systems and given the ability to act.

So the question is bigger than, “What did we tell the system to do?”

We also need to ask, “What can the system do while trying to accomplish it?”

The Nuclear Comparison and Its Limits

During the conversation, Dave compared AI with nuclear technology in the 1930s. I think the comparison is useful, but it has limits.

Both technologies offer enormous benefits and potentially catastrophic forms of misuse.

Nuclear science can generate power, advance medicine, and support research. It can also produce weapons capable of enormous destruction.

AI may help cure diseases, improve education, increase accessibility, strengthen emergency response, and accelerate scientific discovery. It can also support cyberattacks, surveillance, misinformation, biological research, autonomous weapons, and manipulation at enormous scale.

The major difference is accessibility.

Nuclear technology depends on specialized materials, facilities, and infrastructure. AI is software. It can travel globally, be copied, improve quickly, and spread through research, computing power, data, competition, and much much more.

That makes regulation much harder.

If one company slows down, another may not. If one country creates strict safeguards, another may continue developing at full speed.

Companies are rewarded for capability, speed, and being first. National security pressures create the same incentive. Everyone can agree that caution matters while also worrying that caution could allow a competitor to move ahead.

That is why “just regulate it” is not enough.

Good regulation needs specific goals. High-risk systems should face meaningful testing before deployment. Responsibility should be clear when an autonomous system causes harm. We need standards for transparency, incident reporting, security, privacy, and human oversight.

Those rules also need to survive the next generation of technology. Regulation written around today’s chatbot could be outdated quickly. I chuckle to myself every time someone tells me "Oh yeah, well AI can't do this and AI is useless because it can't do that" because I know that comment will age like milk in less than a few months after those words left their mouth.

Bad regulation can lock old assumptions into future rules. No regulation leaves the public absorbing risks it never agreed to take.

The difficult part is finding the balance. In modern society, with pro- and anti-band camps, it is difficult for the majority to adopt objectivity when they are solely focused on subjectivity.

The Bigger Risk Is Blind Trust

One of my biggest concerns is that too few people understand what AI is doing, where its answers come from, and when it should not be trusted.

I often use the Chinese Room thought experiment to explain this.

Imagine sitting in a room with symbols you do not understand. You have an instruction book telling you which symbols to return when someone sends in a particular sequence.

To the person outside the room, your answers may look fluent and intelligent.

Inside the room, you do not understand the language. You are following patterns and rules.

Modern AI is far more complicated than that example, but the lesson still matters. Producing a convincing answer is not the same as understanding that answer the way a human does.