In the mid-1990s, a mathematician named Andrew Wiles stood before the world and proved Fermat’s Last Theorem, a 350-year-old puzzle that had stumped the greatest minds. It wasn’t AI that cracked it. It was human persistence, creativity and decades of collective effort.
Now contrast that with what today’s generative AI can and can’t do. It can summarize 1,000 pages in seconds, predict customer churn, classify security threats, or route support tickets before your team finishes their morning coffee.
But ask it to replicate Wiles’ proof, it can’t. In fact, it fails spectacularly. And that’s exactly the point.
We’ve spent too long trying to measure AI by what it can’t do. The truth is, AI doesn’t need to crack ancient theorems to revolutionize business. It just needs to help us work smarter, faster and more securely.
That’s the generative AI shift. We’re moving from innovation theater to real business impact.
Don’t wait for AI brilliance
A recent study from Apple researchers, The Illusion of Thinking, got a lot of buzz, but it actually missed the point. It found, quite obviously in hindsight, that large language models collapse in accuracy when solving logic puzzles like the Tower of Hanoi, even when provided instructions. That’s because generative AI models excel at recognizing patterns, not applying recursive logic.
So what? Most business problems aren’t logic puzzles. They’re about detecting anomalies, automating workflows, correlating signals and reducing wasted effort – the kind of high-frequency, low-glory work that keeps companies moving.
Waiting for generative AI to “get smart enough” before you deploy it in your organization is like refusing to fly commercial jets because they don’t reach the moon. You’re missing the overwhelming utility already in front of you.
The real business value of today’s AI
AI’s true strengths, such as natural language processing, pattern recognition, and trend prediction, aren’t just academic capabilities. They’re profit drivers.
Consider a few impact areas:
- Customer service: AI summarizes support tickets, triages them by tone and urgency, and drafts first responses – reducing time-to-resolution and agent burnout.
- Finance and fraud: It flags outliers in transactions, reconciles mismatched ledgers, and forecasts spending with uncanny accuracy.
- Sales acceleration: AI pulls key insights from call notes, auto-updates CRM fields, and drafts personalized follow-ups.
- IT and security: It compresses alerts into digestible insights, surfaces early indicators of compromise, and suggests policy improvements.
This isn’t about eliminating human experts. It’s about removing the friction around them so they can focus on what matters.
AI will continue to elevate people in the future. For example, augmented human intelligence is where machines handle heavy lifting and humans apply judgment, creativity and context. Here’s how it can work in practice: A cybersecurity AI flags an unusual pattern of access attempts on the laptop that belongs to the CFO. The analyst knows that the company is engaged in a confidential acquisition, and sees the deal room with an intentionally obscured URL. AI spotted the anomaly. The human understood the stakes.
Augmented human intelligence doesn’t just speed things up. It makes outcomes smarter.
Why pattern recognition wins vs. symbolic logic
Let’s be clear on what most AI today actually does. It doesn’t reason like a mathematician; it predicts based on probability. That makes it incredibly powerful for use cases like time-series forecasting (capacity, spend, behavior), text classification and summarization, anomaly detection in logs or telemetry, correlation of seemingly unrelated events, and policy and workflow optimization.
The messy, semi-structured data that enterprises generate daily? That’s where AI thrives.
What it’s not good at is symbolic logic. Things like trigonometry proofs, recursive puzzles, or reasoning over abstract math. That’s fine because symbolic reasoning is fragile, domain-specific and not scalable across noisy business environments.
The messy, semi-structured data that enterprises generate daily? That’s where AI thrives.
Avoiding the WALL-E trap
Here’s what really worries me – not that AI can’t prove Fermat’s Theorem, but that we stop trying. We mistake convenience for capability, and we trust AI implicitly and stop thinking critically.
Remember the Disney movie WALL-E? Humans let machines do everything for them until people reached the point that they could no longer walk. It’s a metaphor for what happens when we outsource too much.
In business, this can manifest in a few ways. It could be marketing teams blindly accepting generative content without review and strategy alignment. Maybe it’s IT teams relying on AI-generated fixes without understanding the root causes. Or it could be security teams trusting black-box alerts without verifying the threat context.
The illusion isn’t that AI is thinking. It’s that we stop doing so.
Be like Wiles, not Wall-E
Andrew Wiles didn’t have an AI to help him prove Fermat. What he had was something more valuable: the drive to think hard, question deeply, and build new knowledge on old foundations. That’s our challenge too.
AI can get us 80% of the way. But it’s still up to us to go the last mile, to connect insight to impact, to ask the tough questions. Today’s AI won’t solve the hardest puzzles. But it will give you the time, tools, and clarity to do so.
