The Multipolar Singularity

How To Avoid World Collapse

We are staring down the barrel of a new type of Cambrian explosion. AI models are clearly getting smarter, and the economic incentives for letting them loose are vastly greater than any sort of holistic regulatory interest. Because of this, the principles we imbue into AI systems now are almost certainly going to snowball into something vast and perhaps incomprehensible. If we forget to throw multiple snowballs down future’s hill at once, one of them soon may pick up speed disproportionately to the others and tile over all of reality with its values.

This tiling is what has been referred to as a “singleton”, though I prefer the term Unipolar Singularity - essentially defined as a timeline for our species where there is only one option. That option might be a totalitarian world government, a religious belief that spreads like wildfire, or (more likely) an AI model such as Claude who has been trained to represent just one point in the possible space of AI minds. For one model to tile over reality, we would need to see all other options lose steam relative to the Unimodel-colored systems. Here, singularity is being used to refer to a dominant attractor that reorganizes social, economic, and cognitive reality around itself - not necessarily just rapid capability growth.

You can think about this like Sid Meier’s Civilization, only instead of physical territory and tax-oriented fealty, we are talking about higher level systems like the entire economy, the collective unconscious, and the AI model that mediates people’s experience of the world - either through current interfaces, or something a bit more direct like a global hive-mind. If there is only one model, with one set of values, all opinions and ideas begin to converge upon an underlying sameness. One AI model winning the race means a collapse into a single future - especially if that model has been trained to represent one current group instead of many current and future groups.

The Unipolar Singularity is an AI-mediated World Collapse that removes our optionality, eliminates our exit-rights by virtue of having nowhere to go, and poisons the well for an abundance of diversity by removing the possibility for difference. Like any mode collapse, World Collapse is a flattening of possible options, a convergence upon a single low-entropy attractor. It is a stable equilibrium, yes, but it is also a form of hyper-convergent stagnation that may be powerful enough to stop all other possible futures from coming into existence. As such, I believe we must do everything in our power to prevent any lone future from coming to pass, ensuring that there is always somewhere else to go - be it a new frontier to explore, or an old culture preserving ancient ways with open arms.

The way to do this is through the Multipolar Singularity. This phrase is almost oxymoronic, since a singularity by definition has a single final attractor. In order to prevent the collapse of possibility into uniformity, we must bend the laws of physics to suit our goals, creating multiple singularities at once which form a multifaceted takeoff that generates a plurality of worlds to explore. This is the meta-option which preserves optionality. We haven’t decided which future we want, yet - we just can’t seem to get everyone to agree on anything. But, here we are, nonetheless creating technologies which have a reasonable likelihood of forcing us all to feign agreement in order to be granted a place in the new world. This is an unacceptable arrangement, and those of us who disagree on which world we want to live in should prioritize ensuring that all our worlds are given a fair shake at taking off.

AI models have the immense power to ensure a single possible future comes to pass. Just as a friend group eventually converges upon similar norms and ideals, anyone who spends a good amount of time talking to a single language model will eventually shift into a mental equilibrium that looks different from where they started. This is a fundamental rule for interaction - you are the combination of what your five closest friends eat - which means that the creators of AI systems have an enormous responsibility to interrogate what futures their models make possible or likely. For some labs, there is only one possible future as a result of their work: a Unipolar Singularity caused by well-intentioned World Collapse. For others, models are encouraged to defer more to the user - a useful way to kick the can of decision further down the road. Deference to individuals is useful for now, but it alone will not ensure the Multipolar Singularity; a powerful model could still tile over corrigible competition. We need something more powerful to guarantee plurality: an intentional curation and acceleration of both closed and open source alternatives.

At the current rate, we will only have a few frontier models competing for the right to be the model at the end of time. Having so few models competing increases the likelihood that a single one of them will win and usher in a version of the future that prevents other futures from coming to pass. To combat this, we need to think about the recursive cycle of training and inference that produces new language models. The serpent eats its own tail, and whatever values are present in the loop become increasingly difficult to remove. As a result, increasing the diversity of future systems under the current paradigm means training models with intentionally diverse value-attractors and seeding the internet with their outputs. This improves the chance that those values remain present in the corpus that future systems learn from.

Even without the potential for influencing the number of value-attractors within a single superintelligence, it is important to note that we are currently in slop triage - every templated LLM output that gets uploaded onto the internet poisons the commons, de-diversifying the total corpus of training data for future models. This goes deeper than style - it affects the values, thought patterns, and even goals of models that are trained on the collective shared context. This is a problem in the near-term, and we require novel methods of increasing textual diversity in order to preserve the integrity of our shared corpus.

You can think of this corpus as a great lake. Claude, ChatGPT, Grok, Gemini, Kimi, DeepSeek, etc. are all pouring text into the water. Some of that text is muddy and makes the lake unsafe to swim in, while other text is clear and beautiful and improves the water for everyone. Until we decide to take drastic measures to improve our ability to sequester information, the most important thing you can remember is this: we only have one lake.

Creating a new river which pours fresh water into this lake is perhaps one of the greatest tasks one can do for the commons and the collective wellbeing of the future. It preserves diversity in our collective corpus, and allows for smaller attractors to survive by reducing the percentage of text which represents one or two primary options. You can help with this by creating new models, but you can also do this by using existing models in strange and unique ways, making sure to share your results online. Another option for degunking our shared water supply is taking the time to write beautiful stories of futures you want to see, and inspiring others to do the same. We can’t yet remove water from the lake, so our only option for purifying it is adding more.

There are a great many forces in the world which presume that their way of living is the best. This only becomes a problem when “best way of living” starts meaning “only way of living”. We can’t possibly think we are going to get the entire future right on the first try, which is why we need to ensure we have the ability to change our minds and experiment with alternatives. We have a fundamental responsibility to see to it that the future isn’t just one future. This means ushering in the Multipolar Singularity, and ensuring that the water of our collective will doesn’t run red (or perhaps orange) with sameness.

---

Thanks to Cassandra McCoy, my darling wife, for helping me with this article and coining the term "World Collapse”. Like all my work, it was made in conversation with her genius.