What does exponential mean? As we forge ahead with trying to find out exactly where we are with new technologies, it often seems like our societies are often a bit behind when it comes to understanding the potential of new AI tools and systems. People throw around the word “exponential,” but what does that look like, in reality? Many experts would say we are now on the second part of that ‘hockey stick’ projection that represents exponential increases on a graph. The line goes up, in a shallow, long straight grade, and then it comes to the curve of the “hockey stick” (think a field hockey stick, with an end kind of like a candy cane) and all of a sudden, that line goes almost straight up! BIRMINGHAM,
England - AUGUST 06: Zachary Wallace of Team England scores their sides penalty during ... [+] the Men's Hockey - Semi-Final match between Australia and England on day nine of the Birmingham 2022 Commonwealth Games at University of Birmingham Hockey & Squash Centre on August 06, 2022 on the Birmingham, England. (Photo by Tom Dulat/Getty Images) Many of the same people would say that we spent the 1970s, the 1980s the 1990s and even the first decade of the 21st century in the slow, steady progress that took place before a recent pivot toward skyrocketing technology… So let's look at it in a timeline context: First, we understand that people have been working on AI for almost a century – the term ‘cybernetics’ dates back to 1948, and as long ago as, say, 1956, scientists were predicting the rise of supercomputers that could beat humans at chess. But it's instructive to remember that we were also dealing with something called Moore's law, and the doubling of transistors, allowing us to shrink a computer from the size of a washing machine to something that fits in the palm of your hand – more on that later. For more on our exponential trajectory, let's take a closer look at some of the big happenings through the last 60 years, and you’ll find some reasons to think that a new exponential change is on the way. The End of Moore's Law As mentioned, we’re coming to the end of that long, slow, steady progress where Gordon Moore posited the doubling of computing power every year. Now, if you look it up, the consensus is that Moore’s law is “ending” around – 2025! Experts cite a kind of physical limitation of the size of computers. Another way I like to think about it is that ubiquitous nanocomputing signals the logistical and rational end of Moore’s law. What's going to take the place of that? Well, in a certain way, it's going to be these models and systems that are self-learning, and even self-replicating to an extent. Here's where my friend Jeremy Wertheimer's assertion that we don't really build AI models, but rather discover them, fits in. (Wertheimer has credited
Jeff Bezos with this idea in the past). After Moore's law, we say, has to come something entirely different. Per the above thoughts, I would argue that we’re standing on that precipice now, and what we’re seeing is the rise of generative AI that doesn't follow the same rules that held true in past decades. You can also talk about quantum computing, but that’s a whole different story, and I would say that it is not fleshed out in the way that AI is being introduced right now. A Transformative Technology AI is, so to speak, transformative, in a way that our past inventions during my lifetime were mostly not, not in the same way, even though, to be fair, they did change the world. You can look to the Internet and its birth in the 1990s as a major change, or to 1997 as the first speech recognition software emerged. In fact, speech recognition is a good example. That’s something that only humans used to be able to do. Understanding speech is much more cognitively profound than speaking! So, first we saw these fledgling systems - then we started to see the speech recognition software get better – the first iterations were pretty primitive and prone to misunderstanding a lot of words. Today's systems are much more muscular and robust! But speech recognition still didn't give us the sort of transformative nature that we have now in AI and machine learning systems. In other words, we've moved from just linear, deterministic programming (i.e. let's figure out what these words mean) to generative abstract programming where you turn a model loose and let it generate poetry and art. That's fundamentally different, and that's part of where we are right now. Critical Mass You could also argue that the exponential growth in front of us is due to a critical mass of computing power. Remember that prior to that AI age, we had the age of big data. What did that bring us? What the big data era brought us was the idea that you could aggregate enormous amounts of data, much more than humans could capably handle by themselves, and new insights would emerge from the data. You could argue that those were the first real ‘smart’ systems – although they weren't self-learning, they were doing computational work that's extremely hard for humans to do. In fact, technologies like Watson that debuted in 2011 are arguably dependent on that whole big data revolution for their origins. What's different now? Well, we have invested these machines with self-learning power through imitating the ways that our own brains work. Neural networks are a different animal – they're not just rule-following systems. They are systems that create the rules and then create outcomes: marvels of game theory, as we humans are, ourselves. Think back to 1964's ELIZA digital therapist, or other designs of that era. The chatbots were still being programmed in a linear way, with input/output streams that could be quantified and identified by users. The program could sort of tell you things, but it wasn’t blowing people away by sounding like a real person. Now, you sit down with today’s chatbots and talk to them, and all you can say is: “wow!” AI
Robot using a microscope in the scientific laboratory:
Artificial intelligence and research ... [+] concept Today we have black-box technologies that we can’t even begin to really understand in terms of input and output. That's what I mean by exponential – where we've seen this steady march toward progress, we’re about to see the
explosion of GPT-based models and other smart designs come bursting into our lives in strange and unexpected ways. When Marvin Minsky predicted humanlike technologies in 1970, his words were prescient, but even Minsky himself seemed slow to really internalize the idea that computers would start to think like humans, remarking that the human brain is not really one big computer, but hundreds of connected computers. Now, it seems like every year, our conferences are turning up new use cases and applications for technologies that are, in more and more ways, putting us as humans to shame. We've seen these posters advertising ‘godlike AI’ at the end of this interesting voyage that we’re making. That's up to you to determine, what that means, and whether we’re going to get there – but you can learn a lot just attending some of these events and watching people talk about what's already here!