Artificial Intelligence, Evolution & Pizza

Very few words create as much controversy with their simple utterance as evolution. From theological arguments about human evolution to the scientific challenges of the inverse quest for more complexity in a universe that is moving toward more simplicity, the concept of evolution is dissected and contemplated almost incessantly. When that same concept shifts to the evolution of technology, the conversation can become even more contentious.

I’ve worked in tech for decades and built systems, from conception to production, dozens of times. I have worked during the dot com boom, during the cloud migrations, during the use of containerized code, and now, during the AI boom. But these are only the modern tech booms. We can trace the evolution of technology back to airplanes, railroads, cars, electricity, assembly lines, steam, and keep going back to farming, writing, and fire. At one point, fire was cutting edge technology, and it took humans hundreds of years to figure out how to contain it, control it, manipulate it, and eventually master it. Now, we just turn a knob without a second thought and watch the calm blue flame pop up on our stove tops.

There is no doubt that the pace of technological evolution is accelerating. With the advent of modern Generative Pre-trained Transformers (GPTs, as in ChatGPT), the public has been introduced to ongoing work in Artificial Intelligence in a very impactful way. This machine can type! This machine can talk! And just like that, the AI boom has completely obliterated every other technological innovation in the media and is all anyone can talk about. AI is going to solve (or destroy) absolutely every single thing, and it is going to do it in about five minutes.

We must ask ourselves some questions at this point, though. First, how does AI solve something? Can AI solve physics? Can AI solve Spanish? Can AI solve art? What does it mean to us to solve one of these things? This may seem like an overwhelming number of questions, but hang on - I’m going somewhere with this.

In technology, we must often ask what problem we are trying to address. Some problems can’t be solved, but that doesn’t mean we should not try and do something to improve a given situation. A product or solution must be needed, or it won’t be used absent the initial hype. This brings us back to biological evolution. Humans have changed over the eons, slowly and inexorably driven by the changing environments we’ve lived in. We evolved to be smart, focused, able to absorb information and create new ideas, and to use tools. We needed to be smarter to outlast other species. We needed to be focused to hunt prey or forage for nontoxic foods. We needed to be creative to outthink other predators who were bigger and stronger. We needed to use tools to augment our abilities and survive.

Technology is a tool. It extends our abilities beyond what we are gifted with by our innate biology. As we have evolved, so have the tools that we use. As I noted earlier, our tools have evolved right along with us, from fire to steel to the internet to AI, as we have discovered greater needs for these tools. Looking carefully at the tools we have now, the most recent innovation we can’t stop hearing about is artificial intelligence, and some believe this is the final tool that humanity will create, as it will now begin to evolve by itself.

This tool, AI, is presumed to be intelligent. Modern AI, even the best models, are not yet actually intelligent. They are devices, very big prediction engines that we have trained to respond with the statistically best prediction when we input patterns of information. This is a truly exceptional tool, able to conversationally respond to questions and inquiries, research through other written information, generate computer code, images, or video with some simple prompts, and accelerate work that humans do in remarkable ways. But it remains a tool. Something to be used by humans.

The question of the tool evolving itself is being hotly debated right now in technical spaces. Two camps have arisen in this debate. One group claims that if we as humans simply overpower natural intelligence and give these models more computational resources they will eventually ‘wake up’ and begin to operate intelligently and creatively. A second group believes that this current model structure is in no way capable of becoming truly intelligent and that new methods and technology would need to be invented for a model to become intelligent and creative. This entire debate is beginning to sound more religious than scientific, with large groups claiming that a god-like artificial intelligence is just over the horizon, and that this AI god will solve all the world’s major problems, wiping our concept of evolution out of existence and simply advancing intelligence to infinity.

It would take me an entire book, and in fact many have been written, on why accelerating returns is mathematically challenging, and that artificial intelligence returns are predicated on intelligence growing exponentially in a known pattern. An actual mountain of assumptions would need to become reality for AI to become generally aware in any meaningful way and begin to evolve itself or even be subject to a new norm of natural evolution. We as people tend to lose sight of the immense time it took for natural evolution to create biological intelligence, and it takes supreme hubris to think that we can circumvent this system in just a few years. I believe we will achieve artificial general intelligence much faster than it took nature to achieve general intelligence, but it will not happen in the way that we are currently trying. We can’t manhandle the elegance of evolutionary progress, no matter how many processing cores we throw at the problem.

I have worked with dozens of data scientists over the last decade, trying to understand patterns in information, categorize these patterns as data information, and eventually pinpoint knowledge. A very smart colleague of mine and I were at lunch last year, talking about how challenging it was to get time on Graphics Processing Units (GPUs, ideal chips for artificial intelligence work). We mused about the idea of a new type of processing unit and new cores that could handle the challenging math associated with the predictions that LLMs make. As we were eating, he said something profound:

“Our brain is very interesting. It takes time seeing our environment, hearing our environment, and then making decisions, and interacting with our environment. Given enough time it can do that math that GPUs do, it can predict proper outcomes, and although much slower, it can do what these models can do. And it’s smaller and much more efficient. And it runs on pizza.”

The efficiency and simplicity with which our brains solve problems might never be matched by a machine. It will undoubtedly require a completely different perspective on how humans’ natural intelligence evolved - looking at how evolution created intelligence in us - before we get artificial general intelligence that can evolve itself.

In the meantime, let’s have more pizza.