As someone who's spent over 25 years in development, steering tech investments and scaling operations for startups and enterprises alike, I've seen my fair share of hype cycles. From blockchain to the metaverse, each wave promises to redefine everything—until reality sets in. Today, generative AI (GenAI) is the darling of the tech world, churning out poems, code, and cat memes with eerie proficiency. But in my work advising on AI integrations for business-critical systems, I'm doubling down on "classic AI." You know, the unsexy stuff: rule-based systems, decision trees, support vector machines, and traditional machine learning models that have been quietly powering industries for decades. Don't get me wrong—GenAI is a marvel for creativity. But for the predictable, explainable, and mission-critical services that keep businesses running, it's often too risky and fragile. Let me break down why classic AI remains the reproducible, scientific backbone we need, and why GenAI might need to grow up before it earns a seat in the boardroom. What Do I Mean by "Classic AI"? First, a quick distinction. Classic AI refers to the foundational approaches that predate the transformer revolution. It is the umbrella under which most machine learning falls under.  Think expert systems that encode domain knowledge through rules, or supervised learning models like random forests that learn from labeled data to make predictions. These are the tools behind fraud detection in banking, recommendation engines in e-commerce (pre-ChatGPT era), and predictive maintenance in manufacturing. GenAI, on the other hand, encompasses large language models (LLMs) like GPT variants, diffusion models for images, and other neural networks trained on massive datasets to generate new content. They're probabilistic powerhouses, excelling at pattern matching across sequences of words, pixels, or sounds. The key difference? Classic AI is built on explicit logic and verifiable math, while GenAI thrives on implicit correlations mined from oceans of data. The Scientific Bedrock of Classic AI: Reproducibility and Rigor I learned early in my career to prioritize investments (monetary and time) that deliver consistent ROI. Classic AI shines here because it's inherently scientific—rooted in methods that are testable, repeatable, and grounded in theory. Take reproducibility: Run a decision tree model on the same dataset twice, and you'll get the exact same output. No "stochastic parrot" surprises. This determinism stems from classic AI's reliance on structured algorithms with clear parameters. For instance, in a linear regression model predicting sales forecasts, you can trace every coefficient back to the data's statistical relationships. It's not magic; it's math. This scientific foundation extends to validation. Classic AI models undergo rigorous testing—cross-validation, error metrics like RMSE or AUC-ROC—that provide quantifiable confidence intervals. In my experience rolling out inventory optimization systems for retail clients, we could simulate scenarios and prove the model's accuracy down to the decimal point. Regulators love this; so do auditors. It's why classic AI powers FDA-approved diagnostic tools and SEC-compliant financial models. Contrast that with GenAI's black-box tendencies. Sure, you can fine-tune an LLM, but good luck reproducing outputs across versions or even runs due to temperature settings and randomness. In business, where decisions impact livelihoods, this fragility is a non-starter.  Remember that when an LLM doesn’t have any good answers, it picks from the pool of “best probabilistic option” (aka it guesses) which is why we end up with “hallucinations.” GenAI's Superpower: Creativity Unleashed To be fair, GenAI is transformative for creative use cases, and abundantly so. It's a brainstorming buddy on steroids. Writers use it for drafting articles, designers for ideation, and marketers for personalized campaigns. (spoiler: this article even went through an editorial pass with AI!) Even now, I see teams leverage tools like nano banana for rapid prototyping of ad visuals or Midjourney for conceptual art in product pitches. editorial The magic lies in its ability to synthesize novel combinations from learned patterns. Need a haiku about quantum computing? Done. Want to generate code snippets for a niche API? It'll spit out something workable in seconds. For non-critical, exploratory tasks, it's a productivity rocket. But here's the rub: Creativity often comes at the cost of control. The Risks of Depending on GenAI for Mission-Critical Services GenAI's Achilles' heel is its unreliability in high-stakes environments. It's too fragile for the predictable, explainable services most businesses demand. First, the risk factor: Hallucinations—those convincing fabricated factoids—are infamous. An LLM might "invent" legal precedents in a contract review or a new organ inside the human body when facing new situations to analyze. I often advise against deploying GenAI for compliance checks because a single error could trigger lawsuits or fines. Classic AI, with its bounded outputs, minimizes this by design (though you still need to monitor how it performs). Fragility amplifies the issue. GenAI is hypersensitive to inputs—tweak a prompt, and outputs swing wildly. Tools like Zapier and n8n are great examples.  Prompt engineering becomes an art form, but it's not scalable or sustainable for enterprise systems. Imagine relying on that for air traffic control routing or supply chain logistics. One adversarial input (intentional or not), and the system crumbles. Explainability is another gap. Businesses need to know why a decision was made, especially in regulated industries like finance or healthcare. Classic AI delivers this through interpretable features—e.g., a logistic regression highlighting "credit score" as the top predictor in loan approvals. GenAI? It's a neural soup. Techniques like SHAP or LIME help, but they're post-hoc bandaids, not inherent transparency. why For most use cases, this makes GenAI unsuitable for mission-critical ops. Sure, it's fine for chatbots handling low-stakes customer queries, but for fraud detection? Stick with ensemble models that have been battle-tested. When AI Models Grow Up: Toward Cognitive Architectures Looking ahead, GenAI might mature by incorporating "cognitive” tech—think neurosymbolic AI that blends deep learning with symbolic reasoning. Concepts relate to concepts, instead of words predict next words.  Future models could (should!) reason over concepts, rules, and causal relationships. Imagine an AI that verifies facts against knowledge graphs or simulates outcomes logically, not probabilistically. Projects like OpenAI's o1 model hint at this, with internal reasoning chains. But until we shift from sequence-based prediction to true cognitive processing, GenAI will remain better suited for the creative fringes than the business core.  Of course, we still need to navigate a rather large issue of defining truth, morality, and measuring the “truthiness” of content that our models scrape off our opinion-driven interwebs. In my view, this evolution could bridge the gap, making AI more scientific overall. Until then, I'm advising clients to hybridize: Use GenAI for augmentation, but anchor with classic AI for reliability. Why Businesses Should Double Down on the Classics In the rush to adopt GenAI, we're overlooking the proven power of classic AI. It's not flashy, but it's the reproducible, scientific foundation that drives real value—predictable outcomes, explainable decisions, and minimal risk. I’m not a spring chicken anymore – but I can still ninja code up a storm, but my experience has brought clarity around anticipating trouble before it happens.  I've learned that sustainable growth comes from betting on substance over hype. Classic AI isn't going anywhere; it's evolving quietly while GenAI grabs headlines (having stood on the shoulders of Classic AI’s progress). For entrepreneurs and execs building resilient services, it's time to double down on the classics. Your bottom line—and your sanity—will thank you.

This story contains new, firsthand information uncovered by the writer.

OpenAI

Stick

Doubling Down on Classic AI: Why I'm Betting on Science Over Spectacle in Business

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

Curate Your Own Ad-Free YouTube Experience: A Simple Guide for Watching Your Favorite Channels

3 Areas Where Businesses Should Use AI to Inform Their SEO Efforts

2023: What's Driving The Breakout Year for Generative AI?

Amidst COVID-19 Restrictions, How Can AI Help Retailers?

An In-Depth Look at Biden's National AI Strategy

Curate Your Own Ad-Free YouTube Experience: A Simple Guide for Watching Your Favorite Channels

3 Areas Where Businesses Should Use AI to Inform Their SEO Efforts

2023: What's Driving The Breakout Year for Generative AI?

Amidst COVID-19 Restrictions, How Can AI Help Retailers?

An In-Depth Look at Biden's National AI Strategy

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps