The media frenzy this week is about the need to slow AI down — and about government stepping in to restrain AI development. The President was right that government restraint isn't the answer. The people urging AI development to slow down are also right.
What gives?
Do the President and AI leaders really disagree? Probably not. They're just coming at this from different angles.
This took me back a couple of months, to an article I read about the COO Summit in Scottsdale this past summer. The comments came from Ravi Kumar, CEO of Cognizant, who'd clearly already been thinking this through. He talked about the importance of understanding — and using — two ideas: Macro Delegation and Micro Steering.
What was he referring to?
He pointed out that AI gives all of us the ability to delegate the details — to macro delegate — most often to AI applications. That frees people up to spend more time on Micro Steering.
Macro Delegation is easy to understand: figure out the routine parts of a task and hand them to AI.
Micro Steering is getting a lot of attention in the media right now, amid all the talk about AI moving too fast without controls. So what is it?
Micro Steering means deeply and thoroughly understanding where the limits on AI should be placed — on a project-by-project basis.
(This is where the President is right. Major companies have carefully and thoroughly developed their AI products. It's end users modifying those original products that can throw the process off and generate entirely different results. )
What now?
Readers, I need your help here — I'm not sure I have all the answers.
People developing AI understand exactly what they're trying to do with it. The macro part comes easily. The hard part is identifying exactly what the outcome should be, what that outcome should look like, and — most important — which AI application to use, and when.
Every one of us who uses AI is an AI developer.
Let that sink in for a minute. If you're using AI in your daily life, you're an AI developer.
The slow-down advice coming from leaders of major AI companies?
It's meant for the users of AI — not so much the original developers.
I suspect the President's lack of concern extends to the companies originally developing major AI programs. Those companies follow strict design standards, with real checks and balances in place.
So why are AI leaders urging a slowdown? Because they're watching all the modifications end users — people like you and I — are making to their original products. A company can't stand behind a product if end users have unrestricted ability to get in and change it.
Remember the game we all played as kids? Kid 1 tells a simple story. Kid 2 repeats it as best they remember. Kid 3 repeats Kid 2's version and works back in some of what Kid 1 said that Kid 2 missed. Kid 4 does the same. By the time the story reaches Kid 5, it sounds nothing like the original.
That's the situation some AI leaders see happening.
The President is thinking about AI at its original point of development — where it's under control.
AI leaders are thinking further downstream: What happens if changes made to a cryptocurrency AI product — by well-meaning users or bad actors — invalidate transactions? Or simply stop the software from completing them? Or disrupt production cycles on commercial aircraft manufacturing? Or interfere with an air traffic controller's ability to direct planes landing and taking off?
What concerns these AI leaders is how easily and quickly the products they build can be changed by others. Changes originators don’t even know are happening.
Within a couple of months, no one will know for certain what a given AI product actually does anymore.
That, in turn, can lead to massive, unintentional misuse of AI — not because the application stops working, but because it starts spitting out large amounts of data built on changed assumptions no one is tracking, causing damage no one sees coming.
There's another layer to this. AI is getting very good at creatively redesigning itself, with no help from us. That means AI applications can drift so far off course that even their original creators won't know how to begin correcting them.
So where are the fallback procedures?
Bluntly: there aren't any. Not when the original creators have no idea who has modified their product, how many people have done it, or for what purpose or end.
The financial and operational damage from that could be enough to send the world economy into a sudden nosedive — with no warning to the public, companies, or governments.
AI developers don't know how their AI can be manipulated by others. Sometimes that's malicious.
More often, it's people who don't realize how the tweaks they add, change the overall design and output of the product they're using. Then the next person who sees it treats it as just another application to use — and tweaks it again.
The same thing is happening right now across thousands of AI applications, many being used in ways their original developers never intended. The people making the tweaks often don't understand how much those changes can alter how the original operates, or what it produces.
The original developer knows their own plan, use case, and goal — but has no visibility into how others are using the product downstream. They've moved on to new projects. They're not going back to check on something they built six months or a year ago.
So what's the 3-step solution?
1. Give original AI developers something like copyright protection — prohibiting changes to their programs without written, legal consent. There should be a formal, electronic channel for users to submit suggested modifications, with a requirement that developers respond within a set timeframe. Not everyone will love this, but it would:
Slow the process down.
Let original developers test suggested changes and see how they affect results before they go live.
Prevent unplanned consequences that could affect the entire existing user base.
2. Eliminate the need for government involvement. No one-size-fits-all mandates — the kind governments are famous for and that never actually work.
Even Microsoft should have learned this by now: trying to build one version of Word for everyone. Has cost companies and individuals countless hours and untold lost dollars and productivity losses. Just trying to figure out the latest changes, whether or not those changes even apply to them.
3. Create a formal channel for users to flag proposed "tweaks" and ask whether a given change would affect the program's results — positively, negatively, or not at all — before making it.
Together, these three steps would protect the reputation and integrity of original developers and their companies, while still allowing rapid development of tweaks that genuinely improve products.
It's a simple, planned process to protect AI development, users, and the broader market — and to help ensure AI actually delivers the results it's meant to.
How does unfettered management impact careers?
Every AI user needs to be sure the AI applications they're using deliver the intended results — clearly stated, in plain English.
Not once but every time application is used is used.Know the difference between Macro Delegation and Micro Steering on every project.
Understand when and how to Micro Steer — who's affected, how, and when.
Without above people in management are at the mercy of AI applications they are using. That directly impacts career performance. Promotions. Control how AI applications you are using are controlled.
By downloading AI applications separate from original program. Stop ability of users outside your organization to make changes to AI application you are using. Insure users of any AI application follow 3 step process above.
Model your leadership — and readiness for bigger assignments by assuming responsibility for AI under your influence.
Have ideas or suggestions to make this work better? Reach out to Tom Ferree on LinkedIn, by email, or by text.
Tom Ferree is the founder of Ferree & Associates and SecureEmploy, organizations focused on helping companies find exceptional talent and helping professionals advance their careers. Since founding Ferree & Associates in 1977, Tom has worked extensively with hospitality companies, executives, and rising leaders across the industry. Through SecureEmploy, he shares practical career strategies, leadership insights, and real-world advice to help professionals grow their careers and help organizations build stronger teams.
