“Generative AI is dumb. It’s dangerous. And it’s making people lazier by the day.”
That’s what a close friend declared recently — and he meant every word.
(Background: He is an investor and strategist at a Deterministic AI company; his views should not be taken lightly.)
In his view, only deterministic (predictive) AI matters — the kind that follows strict logic, offers clear answers, and never tries to ‘think’. To him, large language models (LLMs) like ChatGPT are just fancy autocomplete engines — impressive sounding, but ultimately hollow.
Me? I disagree.
Strongly.
I believe generative AI is one of the most valuable tools we’ve ever had — not for everything, but for a lot of things. Yes, it’s probabilistic and imperfect. But it’s also fast, scalable, and surprisingly capable across creative, research, and development tasks. The key is knowing when to use it — and when not to.
What Is Predictive AI?
My friend and I agree on this part: predictive AI, also called deterministic or rule-based AI, is the gold standard when you want reliability.
- It powers credit scoring systems, diagnostic models, recommendation engines, and traditional machine learning.
- You give it structured inputs. It gives you predictable, repeatable outputs.
- It’s auditable, explainable, and safe for mission-critical use cases — from finance to flight control.
It’s like a surgeon: precise, trained, and responsible for life-or-death decisions.
What Is Generative AI?
Where we diverge is here.
Generative AI — such as LLMs like ChatGPT or image models like DALL·E — doesn’t follow fixed rules. It learns from patterns in data, predicting the most likely next word, token, or pixel.
My friend sees that as a weakness.
I see it as a possibility.
Generative AI:
- Draft emails, articles, and code
- Summarizes 100-page documents in seconds
- Translates ideas across languages, domains, and tones
It may not be deterministic, but it’s insanely productive. That’s where I find its strength.
Coding: Where Generative Proves Its Worth
My friend challenged me with this: “How can a model that doesn’t even know code write good programs?”
Fair question.
I explained that generative AI works because of exposure. When I ask ChatGPT to write a Python script, it doesn’t logically ‘understand’ Python like a compiler. Instead, it draws from millions of code examples, bug fixes, tutorials, and real-world projects.
That’s why even when I ask something like:
“Write a function to generate a haiku poem using only rhyming words.”
It returns syntactically correct code with rhyme logic and poetic structure. Not perfect, but shockingly close.
A deterministic model would choke unless it had rhyme rules, syllable logic, poetic constraints, and a text generation engine hard-coded into it.
Where I Agree with My Friend:
My friend isn’t entirely wrong.
When it comes to:
- Medical diagnostics
- Nuclear control systems
- Autonomous vehicles
- Flight navigation
I absolutely want deterministic, explainable models.
I want to know why the decision was made, and I want logs, logic, and guarantees.
Generative AI shouldn’t be flying planes or diagnosing cancer, not without a deterministic layer vetting its output.
Where I Push Back
But I push back — hard — on the idea that generative AI is useless or destructive.
To me, it’s like saying the printing press made people lazy because they stopped memorizing books.
Or that Google ruined intelligence because now we “just search”.
Every technological leap shifts what “skill” means.
Generative AI is not replacing critical thinking; it’s amplifying it, especially for people who know how to ask the right questions.
You don’t stop thinking because you use an AI assistant.
You stop wasting time on the basics and focus on higher-level strategy, insight, and innovation.
The Middle Path: Use the Right Tool
I don’t believe in blind adoption or blind rejection.
Instead, I believe:
- Use deterministic AI where accuracy and traceability matter
- Use generative AI where flexibility, creativity, and speed matter
In fact, the most exciting path forward is hybrid:
Imagine a deterministic AI detects a health issue… and a generative AI explains it to the patient in a clear, compassionate way.
Now that’s intelligence at work.
Final Thought: Two Tools. One Brain.
My friend believes generative AI is a shortcut to mediocrity. I believe it’s a shortcut to momentum.
He wants proof before performance.
I want performance that gets validated later.
Neither of us is wrong. But we’re solving for different things.
In a world that needs both precision and possibility, maybe we should stop arguing about which model is better and instead ask:
“Which model fits the moment?”
That, I think, is the smarter question.
