Revisiting Human Potential in the Era of AI
Part 1. The Speedrunner
It was a sunny Saturday afternoon in New York City, and Mike Thompson was nervously pacing back and forth outside the Times Square Arcade.
Mike, better known to the gaming community as “The Sonic Master,” was preparing to attempt a world record run in Sonic the Hedgehog 3 at the annual New York Speedrunning Competition.
As a child, Mike had always been fascinated by video games. He grew up in a small town in rural Pennsylvania, where his parents had bought him a Super Nintendo for Christmas when he was eight years old. From that day on, Mike spent countless hours playing games like Super Mario World and The Legend of Zelda: A Link to the Past, becoming a master of the 16-bit era.
But it wasn’t until he discovered speedrunning that Mike truly found his passion. As a teenager, Mike would spend hours practicing his speedrunning skills, trying to beat his personal best times in various games. It was the game Sonic the Hedgehog 3 that truly captured Mike’s imagination. He was fascinated by the game’s intricate level design and the endless possibilities for optimization. And he was determined to become the best Sonic the Hedgehog 3 speedrunner in the world.
As Mike waited for his turn at the arcade, he fidgeted with his controller and avoided making eye contact with the other gamers who were gathered around to watch. Mike was not known for his social skills, and he was always a bit awkward around other people. But when it was finally his turn to play, his nervousness melted away. Mike sat down at the console, took a deep breath, and began the run.
For the next fifteen minutes, Mike’s fingers moved across the controller with incredible speed and precision. He expertly executed a series of jumps and spin dashes to navigate the levels, using his knowledge of the game’s glitches and shortcuts to gain an advantage.
On the notoriously difficult “Carnival Night” level, Mike employed a strategy known as the “Pogo Spring Skip” to bypass a section of the level that would normally require multiple attempts to pass. And on the final boss fight, Mike used a series of well-timed spin attacks to exploit a weakness in the boss’s AI, quickly draining its health bar.
When the game ended and the world record time flashed on the screen, Mike let out a triumphant shout. He had done it — he had set a new world record in Sonic the Hedgehog 3.
But just as he was about to celebrate his victory, Mike had a sudden realization. He realized that his incredible skills and expertise in the game had not come from hours of practice and hard work alone. They had come from a deep understanding of the game’s mechanics and a mastery of its possibility space.
Part 2. Possibility Spaces
At first glance, the concept of possibility spaces may seem like a novel and obscure idea. But in reality, it is something that is intuitively recognizable to anyone who has ever played a video game, or even engaged in a more mundane activity like driving to work. We all have a natural tendency to explore the full range of possibilities within any given situation, and to use our knowledge and expertise to manipulate those possibilities to achieve our goals.
This process is perfectly captured in the movie Groundhog Day, starring Bill Murray as a weatherman who finds himself living the same day over and over again.
As Bill Murray’s character, Phil Connors, experiences the same day over and over again, he begins to explore the full range of possibilities within his situation. At first, he tries to escape the repetition by attempting to die in various ways, from jumping off a building to driving his car into a tree. But each time, he finds himself waking up back on the morning of February 2nd, no closer to escaping the cycle.
Eventually, Phil realizes that he can use his knowledge and expertise to manipulate the situation to his advantage. He begins to experiment with different strategies, using his knowledge of the town and the people in it to try to achieve his goals. For example, he uses his knowledge of the town’s layout to quickly navigate to the places he needs to be, and his knowledge of the townspeople’s routines to anticipate their actions and avoid potential conflicts.
One of the key scenes in the movie that demonstrates this concept is when Phil tries to win the affections of the film’s female lead, Rita. In one iteration of the day, he uses his knowledge of her interests and preferences to impress her with a perfect date, complete with a romantic dinner at her favorite restaurant and a beautiful bouquet of flowers. In another iteration, he uses his knowledge of her daily routine to stage a series of coincidences that make it seem like they are meant to be together.
In both cases, Phil is using his knowledge of his surreal possibility space — in this case, the town of Punxsutawney and the people in it — to manipulate the situation and achieve his goal. As he continues to speedrun through each day, he becomes increasingly skilled at navigating the possibility space and achieving his goals. He learns to anticipate the actions of the townspeople and to use his knowledge of the game’s mechanics to his advantage. By the end of the movie, he has become a master of the art of exploring and manipulating possibility spaces, able to achieve anything he sets his mind to.
Part 3. Our Personal Groundhog Day
As we go about our daily lives, we too are constantly cycling through thoughts and memories, exploring the possibilities within our own situation. Just like Phil Connors in Groundhog Day, we use our knowledge and expertise to manipulate the world around us, anticipating others’ movements and using our understanding of the game’s mechanics to achieve our goals.
But what exactly is happening in our brains as we mentally rehearse different scenarios? Recent developments in neuroscience suggest that our brains are constantly running simulations of the future, using our past experiences to predict and prepare for what might happen next. These simulations allow us to navigate the possibility space of our lives, much like Phil Connors navigates the possibility space of Groundhog Day.
Consider the simple act of crossing the street. As we approach the intersection, our brains run simulations of the potential actions of drivers and pedestrians, using statistical models to predict their movements and choose the best course of action. This allows us to make quick, intuitive decisions without having to consciously analyze every possible outcome.
It’s not just our actions that are influenced by these mental simulations. One of the key insights from recent work by the neuroscientist Anil K Seth is that our brains are constantly running simulations of the world around us in order to generate our experience of reality. This means that our perception of the world is not simply a passive reflection of sensory information, but is actively constructed by our brains using statistical models.
This idea of consciousness as a controlled hallucination has significant implications for our understanding of the brain and its role in creating our experience of the world. It suggests that the brain is not simply a static organ that processes sensory information, but is a dynamic system that is constantly working to generate a coherent experience of the world.
One way to understand this is to think of the brain as a kind of speedrunner, constantly running simulations of the world in order to create a seamless experience of reality. Just as a speedrunner uses their knowledge of a game to quickly and efficiently navigate through levels, our brains use their knowledge of the world to quickly and efficiently process sensory information and create a coherent experience of reality.
But this idea of the brain as a speedrunner also highlights the inherent ambiguity of our sensory experience. Just as a game can have multiple possible paths and outcomes, our experience of the world is also influenced by a multitude of possible interpretations of the same sensory information.
Just like Phil Connors, we are all speedrunning through the possibility space of our own minds. We may not be trapped in the same day, but we are all trapped in our own thoughts and memories, constantly exploring and manipulating the possibilities within our situation. And just like Phil, we can use our knowledge and expertise to achieve our goals and navigate the world around us.
Part 4. AI, the Ultimate Speedrunner
The world of speedrunning is a fascinating one, where players attempt to complete video games as quickly as possible, using a combination of skill, strategy, and sheer tenacity. But what if, instead of navigating the virtual worlds of video games, these same speedrunners were tasked with exploring the vast landscape of human language?
Enter the newest generation of large language models and AIs, such as GPT-3, which are capable of generating human-like text and completing a wide range of tasks, from translation to answering questions. These models work by analyzing vast amounts of text data and using statistical methods to make predictions about what words or phrases are most likely to follow a given input.
Imagine a speedrunner tackling a complex game like Super Mario Bros. They might start by trying out different routes and strategies, testing which ones work and which ones don’t. In a similar way, a language model uses statistical methods to explore the vast “possibility space” of possible sequences of words and phrases, looking for the most likely sequence given a particular input.
But unlike a human speedrunner, who is limited by the constraints of their physical body and the speed at which they can process information, a large language model like GPT-3 has the computational power of a supercomputer at its disposal. This allows it to make predictions with a level of precision and speed that would be impossible for a human to achieve.
So how does a model like GPT-3 actually make these predictions? At the heart of these models is a technique called “statistical language modeling,” which involves analyzing large amounts of text data to identify patterns and relationships between words and phrases. This allows the model to make predictions about what words or phrases are most likely to follow a given input, based on the statistical patterns it has learned from the data.
Part 5 — Speedrunning the Future
For most of human history, the act of imagining, planning, and executing a series of actions towards a goal — what we now call speedrunning — was a formidable challenge. Doing something once was hard enough, and iterating meaningfully hundreds or thousands of times simply wasn’t feasible. But with the advent of video games, neuroscience, and AI, we are starting to see palpable manifestations of the power of speedrunning in action.
Video games have given rise to a vibrant community of speedrunners who compete to complete games as quickly as possible, using a combination of skill, strategy, and sheer tenacity. Meanwhile, advances in neuroscience are allowing us to model and simulate human cognitive processes in powerful new ways, leading to the development of AI systems that can perform a wide range of tasks using the same principles that underlie human cognition.
But the potential applications of speedrunning go far beyond the world of video games. As the world becomes more complex, governments and businesses alike struggle to anticipate the unforeseen consequences of their actions. The high stakes of policy making can make it difficult to make well-informed decisions, but what if we could apply the principles of speedrunning to the world of policy making?
With the help of technologies like AI, AR, and VR, we can now use the power of speedrunning to explore and test alternative solutions to a wide range of complex and high-stakes challenges. For example, we could use these tools to anticipate the impact of a community development on the natural ecosystem, or to anticipate the infrastructure needed to support increasing flows of climate refugees.
By leveraging the power of speedrunning, we can explore the long-term consequences of different courses of action, and make more informed decisions about how to tackle the toughest challenges facing humanity. This is where the true potential of spatial computing technologies like AR and VR comes into focus as a medium to bring these scenarios to life, allowing stakeholders to experience the potential impacts of different decisions in a more immersive and intuitive way.
These developments are exciting, and they offer us a glimpse into the incredible potential of speedrunning as a tool for tackling complex and high-stakes challenges. Communities, governments, and businesses are starting to use these new AI tools and spatial computing affordances to explore and test ideas in ways that were once unimaginable. But we must also approach this new frontier with humility and curiosity, recognizing that we are still at the beginning of our journey.
As a final note, I will reveal one final demonstration of this argument, in the form of this essay’s provenance and authorship.
Because of course, this entire essay was written in collaboration with ChatGPT over the course of about two hours, with all of the text above this paragraph being produced entirely by the AI and very lightly edited by me, mostly for minor stylistic preferences. This was made possible by speedrunning the different arguments and themes within the essay with ChatGPT over many iterations, Groundhog Day-ing may way through the complex web of concepts and arguments that I wanted to surface and then connect together. For the patient and curious, I have made the unedited prompts and responses that produced this essay available on Google Docs: https://docs.google.com/document/d/1MM37CvWEpk56J2eYKA7vV6JvkQMAaD0UypD_DLoXVbg/edit?usp=sharing