Just as computers, machinery, and agriculture, AI is poised to change the world and impact every tech movement that follows.
By Maeve Sentner
If you haven’t heard a single thing about AI, I’d love to know more about the going rate of scenic, spacious under-rock homes these days. Seriously, though, AI seems to be on everyone’s lips these days. It’s at once startling, confusing, and thrilling to know that we’re living during the era of a major technological boom.
Just as computers, machinery, and agriculture before it, AI is the new technology poised to change the world and impact every tech movement that follows it. As we watch modern processes grow, develop, and incorporate ever-more AI into every piece of infrastructure around us, it’s natural to wonder just how we got here and what might be next.
Today, we’ll be exploring the evolution of AI infrastructure, from its early beginnings to the advanced systems we see today.
AI (which, for those of you who are enjoying your under-rock lives stands for “artificial intelligence”) has transformed industries, enhanced technologies, and reshaped the future of innovation. Central to this transformation is AI infrastructure, the backbone that supports and enables the development, deployment, and scaling of AI applications.
But where did it come from?
AI may seem like a fully-formed concept when you plug a query into Google and see the AI answer box. Or when you ask ChatGPT to help you improve your cover letter. Or even when you make a new computer wallpaper using MidJourney. But what we’re really seeing right now is a tool in its infancy. And it didn’t spring forth fully-formed.
Holding the full history of computing—which can be traced as far back as the 1800s (or even longer, up to 2700 BCE, if you’d like to count tools like the abacus)—AI really took off as a viable concept in the mid-1900s. (Though it was conceptualized in its modern form even before that. We’ll get to that later.)
In the 1950s, students at Dartmouth College hosted a computer science workshop, and it was there that the concept of machines simulating human intelligence was first dreamed up.
During these early days, AI infrastructure was rudimentary. Computing power was limited, storage capabilities were minimal, and algorithms were basic. These were the days of computers taking up entire rooms and yet not matching a modern calculator in terms of processing power. Researchers relied on simple hardware configurations and software frameworks, often facing significant challenges in processing power and data handling.
Early AI research was conducted on mainframe computers, which were large, costly, and accessible to only a few institutions. These systems provided the necessary computational resources but were restricted by their size, speed, and lack of flexibility. Programming languages like LISP and Prolog were developed during this period, specifically to handle symbolic processing, which was central to AI research.
AI itself was still a very long way off, but that didn’t stop people from dreaming of the future. Remember when I mentioned that AI had been thought of long before it was treated as a project? The earliest instances of machines with human-like intelligence can actually be found in 1872, when an author named Samuel Butler wrote Erewhon: or, Over the Range, a novel which satirized Victorian society. In it, Butler wrote about a society of machines with human levels of intelligence, influenced largely by the recently-published On the Origin of Species by Charles Darwin.
With this early imagining of something like AI came the standard fears we still see around AI today: that machines would become too smart, self-governing and self-replicating, perhaps resenting humanity in being fully self-aware.
When those students at Dartmouth realized that computer technology was still too basic to ever hope to generate AI in their lifetimes, as they’d originally promised, academic enthusiasm for the subject cooled. For decades, AI continued in the vein of Butler’s works: science fiction.
The 1980s and 1990s marked a significant shift in AI infrastructure with the advent of machine learning (ML). This period saw the development of more sophisticated algorithms and an increased focus on data-driven approaches. The demand for computational power grew exponentially, leading to advancements in both hardware and software.
Finally, the technological bases to begin developing AI were starting to take hold.
The introduction of Graphics Processing Units (GPUs) revolutionized AI infrastructure. Originally designed only for rendering graphics, GPUs proved to be highly efficient at handling the parallel processing (or “thinking” of multiple things at the same time) required for ML tasks. This development led to significant improvements in training times and model accuracy.
Google's development of Tensor Processing Units (TPUs) further accelerated potential AI capabilities. TPUs are specialized hardware designed to handle tensor computations, which are fundamental to many AI models. These innovations allowed for more complex neural networks and large-scale data processing.
The 2000s brought about a paradigm shift with the rise of cloud computing. Cloud platforms like Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft Azure provided scalable, on-demand computing resources. These platforms brought democratized access to high-performance computing, enabling more organizations to develop and deploy AI applications.
Cloud-based AI infrastructure offers numerous advantages:
These elements were crucial to AI systems and AI models as we’d come to know them. In order to even come close to replicating human thought, AI had to be complex, fast, and useful, without breaking the bank to run. If any of these elements were missing, AI systems wouldn’t be able to work for anything.
Now that these singular boxes could be checked, AI systems needed the information necessary to truly “learn.”
That’s where data comes in. Big data.
Big data has been a driving force behind the evolution of AI infrastructure. The ability to collect, store, and analyze vast amounts of data has enabled more accurate and robust AI models. Modern AI infrastructure is designed to handle the three V's of big data: volume, velocity, and variety.
Technologies such as Hadoop and Apache Spark have played crucial roles in managing and processing large datasets. These frameworks allow for distributed computing, which breaks down large tasks into smaller, more manageable ones, significantly speeding up processing times.
AI models still rely on tons upon tons of unique data in order to grow machine learning models and develop “original” content. AI cannot exist without big data and ever-growing datasets. This process is time consuming and expensive, and brings up new challenges in the form of there needing to be ever-more data to keep improving and developing the existing AI models. So what’s next?
The need for ever-more information, and the challenges of acquiring it, brings us to the current status of AI systems development: edge computing.
The future of AI infrastructure is poised to be shaped by edge computing. This approach involves processing data closer to its source to reduce latency and bandwidth usage. Edge computing is particularly beneficial for applications requiring real-time processing, such as autonomous vehicles and IoT devices.
Edge computing offers several advantages:
All three of these elements unlock the doors standing between the current state of AI and what might come next. What will AI be able to do once these three paths are fully cleared? That’s exactly what had scientists, entrepreneurs, engineers, and hobbyists so excited.
AI is truly changing everything.
The evolution of AI infrastructure has been marked by significant advancements in computing power, data handling, and deployment strategies. From the early days of mainframe computers to the era of cloud and edge computing, each phase has brought about innovations that have propelled AI forward. As technology continues to evolve, AI infrastructure will undoubtedly become even more sophisticated, enabling new possibilities and applications that we have yet to imagine.
Related articles