An epic battle has erupted between AI glitz and AI value. Theatricality is challenging utility. It’s chic versus geek.

What started it? In the less than three years since I presented the bizML playbook for running predictive AI projects, a newer kind of AI has taken the world by storm: generative AI, or GenAI. It commands center stage in an AI craze that has absolutely exploded in this short time. GenAI offers new capabilities and value, and man, is it sexy as hell.

Of course, the sheer usefulness of predictive AI holds a seductive appeal of its own, but is that enough for it to survive? Will its proven serviceability and great untapped potential keep it afloat even as tides turn? Or will GenAI’s fashionability crush predictive AI, relegating it to obscurity and extinguishing most of its value across sectors?

In one corner, we have predictive AI, which learns from data to predict the outcome for each customer, patient, machine, or transaction. These predictions target marketing, fraud detection, risk management, maintenance, healthcare, and pretty much any other primary function. It’s the technology you turn to for improving existing large-scale operations.

In the other corner, we have GenAI, which generates new content: writing, computer code, graphics, and other media. Amazingly, its output is largely coherent. The content it synthesizes proves valuable, at least as a solid first draft.

Since GenAI responds to human-language prompts, it can converse interactively to answer questions, retrieve information, or even deliberate on decisions and logical arguments.

But the most valuable enterprise applications may not require organizations to choose between the two. Predictive AI and GenAI can work together, with each addressing the other’s limitations and strengthening the overall system.

Watching GenAI’s rise made me wonder: What would its mammoth hype mean for predictive AI? Were my efforts to improve predictive AI’s already-suffering deployment record for naught?

These two AIs should unite, but they compete

GenAI and predictive AI ought to live together in harmony. They solve different problems and present distinct value propositions, so they should compete no more than a camera and a telephone. In fact, they work best together: Hybridizing the two, using one to strengthen the other, delivers the greatest value for many projects.

But these two flavors of AI compete indeed: for resources, budgets, and attention. As this competition plays out on an international stage, GenAI appears to be destroying predictive AI.

Which is more valuable depends on the organization and project. But here’s my rule of thumb: Most companies should invest at least as much in predictive AI as in GenAI.

GenAI threatens to upend this balance. To many, it appears to be getting us closer to machines that are as smart as humans. It comes across as more humanlike than computers have ever seemed before, so it’s often construed as a step toward the AI of science fiction.

As a result, beyond its enormous media attention, GenAI is also attracting far more R&D and venture investment than predictive AI. It has become so dominant in the press that the term “AI,” in its general usage, has come to mean GenAI in particular.

But a knockout blow would be no good for anyone. If GenAI were to dominate to the point of virtually shutting down predictive AI, flashiness would have defeated merit.

Good news: Insider trends bode well for predictive AI

“Generative AI is a seductive distraction from the type of AI that is most likely to make your life better, or even save it: predictive AI.”

— Margaret Mitchell, PhD, Chief Ethics Scientist, Hugging Face

Fortunately, there’s a very different story inside the industry. When it comes to the number of enterprise projects in play, predictive AI is holding its own against GenAI.

Data scientists at large still widely adopt predictive AI. For one, in my work chairing industry conferences, I don’t see GenAI dominating the way it does in the press. Instead, I see a roughly equal division between predictive AI and GenAI projects.

After extensive calls for speakers that cast a wide net, the submissions reveal an even split. This balanced spread represents a rough gauge of the industry.

What’s more, novel predictive AI projects cross my desk as often as ever: for predicting things like electrical grid malfunctions, insurance claim denials, lawsuit settlements, lease terminations, dirty solar panels, no-show dental patients, abandoned shopping carts, successful startups, and willing blood donors.

Back-channel buzz tells the same story. I repeatedly hear that, even as data science teams feel pressure to experiment with GenAI and work to capture some of its promised potential, they see predictive AI projects as the ones realizing the most value.

Predictive AI’s singular value keeps it alive

Amid GenAI’s astronomical hype and undeniable allure, why is predictive AI still thriving?

Predictive AI delivers unique, critical value. Its role in this world will endure because uncertainty is an indelible aspect of life and business. Although GenAI is built on leading technology, it does not replace predictive AI. Predictive AI represents a different kind of endeavor: explicitly managing uncertainties across millions of outcomes.

It uses machine learning to play a scientific “be wrong less often” numbers game inherent to most large-scale operations. For that undertaking, GenAI is not well suited. Although it’s made with machine learning, it is not natively capable of performing machine learning algorithms. Instead, it is better to just run such algorithms intentionally, as needed.

Even when you use GenAI’s core methods to improve a predictive AI project, it remains a predictive AI project in form and function: It delivers value by systematically playing the odds over many cases. These two types of AI projects are intrinsically destined to remain distinct.

What’s more, GenAI’s popularity promises to actually bolster predictive AI’s longevity more than threaten it. This is because GenAI projects need predictive AI. Crucially, predictive AI addresses GenAI’s stubborn reliability challenge: GenAI hallucinates and exhibits other unacceptable behaviors that preclude its deployment.

This is particularly so for its more ambitious intended uses, such as assuming the role of customer service agent, analyst, educator, or virtual assistant. For GenAI to realize a meaningful portion of its bold, often audacious, promise of autonomy, its reliability must improve.

Predictive AI can help. It acts as a reliability layer that tames large language models by monitoring their behavior and targeting human attention toward situations most likely to go wrong.

I am witnessing the emergence of this kind of hybrid approach: Enterprises are actively adopting it, and I believe it represents the next killer app for predictive AI. For example, when a customer service chatbot interaction risks failure, it is held and escalated for human review. Once this becomes common practice, most GenAI projects will also employ predictive AI.

Mastering the rare art of predictive AI deployment

Predictive AI may be older, but it’s not “old school.” This is the original AI — established enterprise uses of machine learning that have accumulated decades of proven results. 

It’s destined to live long and prosper.

Yet despite its age, predictive AI still has some desperately needed maturing to do. Even after decades of usage, predictive AI initiatives routinely fail to deploy, never realizing value.

Since each project endeavors to change ops based on odds, the organization often faces a new challenge when attempting to sell the project’s culminating launch to business stakeholders. After all, even though predictive AI deployment is not terribly complex, it’s not yet widely understood.

If we can address this problem, the realized value stands to multiply many times over.

So, what’s needed?

A specialized business practice suitable for wide adoption. In The AI Playbook, I lay out the best-practice framework for ushering predictive AI initiatives from conception to deployment. This disciplined approach serves both sides: It empowers business professionals, and it establishes a sorely needed strategic framework for data professionals.

Predictive AI sustains great value alongside its younger sibling, GenAI. Find the greatest opportunities for your organization and determine which tech applies. Often, it will be predictive AI or, increasingly often, a combination of the two. In such cases, follow best practices to defy the odds, avert common pitfalls, and realize the potential value. Happy predicting!

*Note: This article is adapted from the new preface to The AI Playbook: Mastering the Rare Art of Machine Learning Deployment with permission from the publisher, MIT Press.