Over the course of the last decade, the use of artificial intelligence (AI) and machine learning (ML) have become commonplace. Businesses today must invest in AI tools if they wish to stay competitive. This is particularly true in marketing and customer service. But these represent the more basic applications of AI that hardly reflect the tremendous made in the field as of late. Today, AI foundational models are entering a new era of development that is nothing short of mind-blowing. This is why investors and researchers alike see the future of AI as the real game-changer of the century.
In the last five years, there have been some pretty significant developments that ushered in current AI models. Certainly, AI and ML were already making an impact on daily lives prior to this. But quietly, some interesting discoveries led to a paradigm shift in how AI systems operated. These new systems are referred to as AI foundational models because they are significantly more flexible. And by adding more and more parameters to these systems, their capabilities are expanding at an exponential rate. Because of this, it’s worth taking a look at what the future of AI will look like in the coming years.
“AI foundation models are an emerging paradigm for building AI systems that lead to an unprecedented level of homogenization: a single model serving as the basis for a wide range of downstream applications.” – Percy Liang, Computer Science Professor, Stanford University
What Are AI Foundational Models?
In order to take a peek at the future of AI, we must first go back a bit to examine the historical landscape. In the early days, initial systems were best described as neural networks. All machine learning systems are based on neural networks, which mimic how brain cells interact. Using tiny chips called graphic processing units, or GPUs, virtual “neurons” could be connected into networks. And based on weights given different neural connections, machine learning could occur. By the late 2000s, computing power and GPUs had advanced enough that large neural networks could be formed. And the magnitude of these networks is what created the groundwork for the future of AI today.
Early AI models began to perform some pretty amazing tasks. Text translation, voice recognition, and facial recognition are some of the initial innovations these systems performed. These systems were designed to perform sequential tasks that allowed trail-and-error through structured pathways. But in 2017, computer scientists at Google and the University of Toronto made quite the discovery. By taking away these sequential restraints, AI systems could learn through pattern recognition. No longer did AI models simply deal with text, but now they engaged in deep learning through images, video, and more. These new models, which are termed AI foundational models, are highly flexible, adaptable and capable of self-learning.
“AI models used to be very speculative and artisanal, but now they have become predictable to develop. AI is moving into its industrial age.” – Jack Clark, Co-founder of Anthropic
The Industrialized Future of AI
With the introduction of AI foundational models, things have moved at a pretty fast pace since. The design of these new AI systems has expanded their capabilities with some now writing programming code and even poetry. But pattern recognition designs weren’t the only major discovery as of late. AI system designers have also learned that parameters an AI system has, the more robust it becomes. In essence, parameters are additional coefficients that AI systems can apply to their calculations. It used to be believed that there was little to gain after a certain number of parameters were added. But as it turns out, this is not the case. For this reason, the most advanced AI foundational models today have trillions of parameters.
These developments have been rapid. Four years ago, the most advanced AI system, Google’s BERT, had 110 million parameters. Today, the more advanced AI foundational models have roughly a trillion parameters. And currently, Graphcore is actively developing its Good AI model, named after Jack Good, fellow WWII codebreaker of Alan Turing. It will boast around 500 trillion parameters and currently represents the future of AI. Without a doubt, such AI systems will be able to markedly accelerate their learning. But at the same time, these developments have standardized AI system development. In other words, system designs have become much less creative and much more industrialized. This is why 80% of today’s AI development is being invested in the creation of ever-more-powerful AI foundational models.
“Covid has taught us that exponentials move very quickly. Imagine if someone at Google builds an AI that can build better AI’s, and then that better AI builds an even better AI—and it can go really quickly.” – Connor Leahy, Lead executive at Eleuther
Forging Into the Unknown
Today, AI foundational models are being used for a variety of activities. Recent developments have witnessed the use of AI in healthcare and in business mergers and acquisitions. More advanced applications are occurring as well at companies like Meta, Alphabet, and Facebook. And there’s little doubt that states and national security departments are tapping int the future of AI as well. The computing power and potential of these AI foundational models are highly attractive, which is why many recommend greater caution. Equipped with massive data and enormous learning capacities, such systems pose a variety of threats ranging from economic to political. And of course, some still believe the potential for sentient conversion of AI exists.
Despite these potential risks, investments are pouring into the future of AI. Last year alone, venture capitalists invested $115 billion in various AI companies. At the same time, Microsoft, Alphabet, Meta and even Tesla have made AI foundational models a priority. Even China have stated the future of AI is a priority for them as well. All of these developments and the rapid increase in AI system capabilities paint a highly dynamic picture. It appears that the race for advanced AI systems is on. But it remains unclear if the future of AI with these systems will ultimately be better or worse for humanity.