Why AI progress feels faster than it actually is

The dizzying pace of perceived AI advancement

It feels like every other week, a new artificial intelligence breakthrough dominates the headlines. From generative art that rivals human creativity to chatbots that hold surprisingly coherent conversations, the world of AI appears to be accelerating at an unprecedented rate. This constant stream of innovation can leave us feeling both excited and a little overwhelmed, as if we’re perpetually trying to catch up to a technology that’s evolving faster than we can comprehend. But what if this perception of breakneck speed isn’t entirely accurate? What if the reality of AI progress is more nuanced, more incremental, and perhaps, a little slower than it feels?

At TechDecoded, we believe in breaking down complex tech concepts into understandable insights. Today, we’re diving into the fascinating reasons why AI’s progress often feels much faster than its actual, underlying development.

The spotlight effect and media amplification

One of the primary drivers behind the feeling of rapid AI progress is how breakthroughs are presented to the public. When a new model achieves a remarkable feat, it’s often framed as a sudden, revolutionary leap. The media, understandably, gravitates towards these ‘aha!’ moments, highlighting the most impressive and often visually stunning applications of AI.

• Dramatic demonstrations: We see incredible images generated by AI, or hear snippets of AI-powered conversations that sound eerily human. These are the polished end products of years of research and development, not the messy, iterative process behind them.
• Hype cycles: AI has a history of boom and bust cycles, often fueled by intense media attention. Each new wave of innovation brings with it a fresh surge of optimism and sometimes, exaggerated expectations.
• Focus on the ‘what,’ not the ‘how’: News reports typically focus on what AI can *do* now, rather than the painstaking engineering, vast datasets, and computational power required to get it there.

This spotlight effect creates an illusion of continuous, dramatic progress, masking the thousands of smaller, less glamorous steps that paved the way.

Narrow AI versus general intelligence

A significant factor in our perception is the conflation of narrow AI with the elusive goal of Artificial General Intelligence (AGI). Most of the impressive AI we encounter today falls under the category of narrow AI. These systems are incredibly good at specific tasks, often outperforming humans in their specialized domains.

• Specialized brilliance: An AI that can beat the world champion at Go, diagnose diseases from medical images, or translate languages in real-time is a marvel of engineering. However, its intelligence is confined to that particular task.
• Lack of transferability: The Go-playing AI cannot suddenly write a novel or hold a philosophical debate. It lacks the common sense, adaptability, and broad understanding that defines human intelligence.

When we see a new narrow AI excel, it’s easy to extrapolate that success to a broader, more general form of intelligence, making it seem like AGI is just around the corner. The reality is that bridging the gap from narrow AI to AGI is an immensely complex challenge, requiring fundamental breakthroughs that are still far off.

The illusion of exponential growth

Our brains are wired to perceive linear growth more easily than exponential growth. However, when we look at AI, we often interpret a series of significant, but still incremental, advancements as an exponential curve. Each new model builds upon the last, often with marginal improvements that, when compounded over time, lead to impressive capabilities.

Consider the development of large language models (LLMs). While models like GPT-3 and GPT-4 represent massive leaps, they are the result of scaling up existing architectures, feeding them exponentially more data, and leveraging increasingly powerful hardware. The underlying algorithms, while refined, haven’t fundamentally changed at the same pace as their output capabilities.

• Scaling, not always inventing: Much of the recent progress comes from scaling up existing techniques rather than inventing entirely new paradigms.
• Cumulative effect: Small, consistent improvements in algorithms, data collection, and computing power accumulate over years to produce what appears to be a sudden, dramatic change.

It’s like watching a sapling grow into a tree; day-to-day, the change is imperceptible, but over years, the transformation is profound.

The long road of data, compute, and human effort

Behind every dazzling AI demo lies an immense amount of unseen work. The progress we observe is built on three foundational pillars that advance at their own, often slower, pace:

• Vast datasets: Modern AI models require colossal amounts of data for training. Collecting, cleaning, and labeling this data is a monumental, often manual, effort.
• Computational power: Training cutting-edge AI models demands staggering computational resources, often involving thousands of GPUs running for weeks or months. The development of more efficient hardware, while steady, is not an overnight phenomenon.
• Human ingenuity and iteration: Thousands of researchers, engineers, and data scientists are constantly experimenting, refining algorithms, debugging code, and iterating on models. This human-driven process is inherently iterative and often slow.

The “magic” of AI is really the culmination of countless hours of dedicated human effort, massive infrastructure, and the slow, steady accumulation of knowledge and resources.

Understanding AI’s true rhythm

Recognizing that AI progress is more of a marathon than a sprint doesn’t diminish its importance or its potential. Instead, it offers a more grounded and realistic perspective. By understanding the factors that contribute to the illusion of hyper-speed, we can better appreciate the genuine advancements, manage our expectations, and engage with AI technology more thoughtfully.

The future of AI is undoubtedly bright, filled with transformative potential. But by appreciating its true rhythm – a blend of incremental steps, occasional breakthroughs, and immense underlying effort – we can foster a more informed and sustainable relationship with this powerful technology.