AI: The Revolution That Isn't – Exposing the Hype and the Old Wine in New Bottles

For many years, artificial intelligence has been surrounded by a narrative of brash claims of a coming revolution that would upend entire industries overnight. But strip away the gloss that is linked to generative deep learning and large language models, and it is apparent that one is observing not a revolutionary paradigm shift, but a refined iteration of processes that have been in place for many decades.

Simply put, advanced deep learning models display remarkable capabilities. They can process huge quantities of data at unprecedented velocities, reshape existing patterns, and generate "novel" material that is perceived to be novel. One should not be deceived, though; this is not magic—it is a hyper-efficient refinement of the same rule-based engines that power complex event processing, which have been used to support various industries such as finance, drug discovery, and digital media for many years. Today's AI systems do not produce concepts ex nihilo; they interpolate between existing data points at a speed and scale that makes older methods seem plodding in comparison.

An instructive example of this is algorithmic trading. Historical quantitative trading systems used statistical models and rule-based engines to recognize signals in the market and place orders. By comparison, today's AI systems process orders at astronomically high velocities using exponentially more quantities of information. The potential advantages include faster decision-making and more efficiency; yet, the underlying economic principles stay in place. These so-called "novel" orders are not a product of epiphanies of inspiration; more a refined, amplified iteration of established methods. In the end, it is a function of applying a tried process to a much higher order of scale that is perceived to be revolutionary.

In a related vein, in the realm of pharmaceuticals development, firms today can point to dramatically smaller time frames and costs. Generative models allow for new molecular architectures to be created through the combinatorial exploration of large chemical spaces that are endowed with abundant historical and experimental data. Generative models do not work independently, though; instead, they augment more time-consuming screening approaches that have been a cornerstone of drug discovery and development for many decades. Artificial intelligence allows for a previously unmatched throughput in keeping with the fundamental rule of hitting on candidates that meet specified efficacy requirements. The process is not new; instead, it is a dramatically accelerated variant of established methods.

Virtual companionship services provide a related example. Sites like Replika provide increasingly personalized and interactive experiences using deep learning methods; nevertheless, at their heart, these systems continue to use pre-trained response models that have been tuned to perfection using multiple rounds of training datasets. The systems learn from user interactions by recognizing patterns in statistics—essentially a sophisticated evolution of earlier chatbots and virtual assistants. In many ways, the key innovation in this realm is not in applying a new economic model but in dramatically decreasing friction around commonplace tasks, thus making it possible for individuals to assign drudgework to their digital counterparts.

The enticing promise of AI copilots is undeniable. One is obliged to note that few enjoy spending lengthy hours composing boilerplate emails, crafting lengthy presentations, or laboriously crafting reports. Tools like Microsoft 365 Copilot have become a lifeline for beleaguered professionals, freeing up time to move beyond drudgework to higher-order strategic work. Such copilots promise to be more than just better templates; they work as sophisticated assistants that tailor their output to personal tastes and needs. Far from revolutionizing our ways of communicating, they augment our capabilities of communication, hence freeing us to pursue more imaginative, more added-value work.

Moving to such new tendencies as “synthetic data as a service,” one would logically suppose that licensing of synthetic data for machine learning is a revolutionary move. In reality, it is a new way of dealing with a classic problem: acquiring high-quality, compliant data to meet business needs related to market analysis. Historically, companies spent capital accumulating, processing, and curating data to sell to analysts. Synthetic data is a privacy-friendly, scalable substitute; it is, though, a next-generation data brokerage business in disguise rather than a genuinely new business model.

What does this mean for our conception of AI? The description of AI as a revolutionary, unprecedented force is largely a marketing narrative—one that overstates a redescription of a mere acceleration of old algorithmic practices. Such contemporary AI technologies deliver large increases in efficiency, scalability, and speed that are truly transformative in applied contexts. Nevertheless, when one peels away hyperbole, it is apparent that such systems simply reinterpret and interpolate from the same reservoirs of data that have been feeding our systems for many decades.

This viewpoint is not to be taken to mean that we should overlook the capabilities of artificial intelligence; it is more of a call to continue our investigation in this field. The true potential in AI of today lies in its capacity to build on established processes, making systems more efficient, productive, and scalable. The vision of new business models—one that upsets existing economic paradigms—is a vision of the distant future. The most notable effects that can be observed today take their form in incremental improvements: existing sectors are being refined and maximized using sophisticated calculation methods.

There is a great benefit in this realm. Imagine having AI copilots to take up monotonous work; freeing our time and mental capabilities to focus on imaginative and strategic work that actually stimulates innovation. Imagine the freedom of having a skilled sidekick to craft letters, produce presentations, or author preliminary reports—able to work on heavy concepts and key decisions. This is not a substitution of human creativity, but more of a boost to it, enabling mastery of our areas of proficiency while machine work is done on a day-to-day basis.

Overall, our story is one of AI-driven revolution, yet it is a much more complex reality. The AI of today is more of an evolutionary step—an unprecedented booster that maximizes existing processes without creating new paradigms. The potential for new business models is in the promise of distant futures. The advances of today lie in existing methods that deliver exponential gains in efficacy.

As leaders in business and innovators, it is crucial to move beyond mere hyperbole. One must recognize the true benefits of artificial intelligence—such as its quickness, efficiency, and scalability—and be cognizant of its limitations. Focus on incorporating these new tools in existing frameworks to enable incremental, sustainable improvement. In addition, it is crucial to remember that, even though repetitive work is typically undesirable, AI copilots help enable us to recapture our time, enabling us to focus on what is truly important.

I challenge you to ask yourselves this question: Is your company after a mythical vision of revolutionary change, or are you willing to use the true capabilities of AI to build on existing winning strategies? Let's cut through the hype, take advantage of the potential for more efficient use of our time, and build on the solid foundations that we've developed over the decades—since sometimes the most groundbreaking advances come not from wholly new concepts but from making existing ones work better. #AI #DontBelieveTheHype #TheGrumpyArchitect #TheGrumpySoothSayer