A Unified Theory of Imagination
Google introduces Willow, conceivability vs. Possibility, and AI Imagination
On Monday, December 9th, Google’s CEO Sundar Pichai introduced the launch of Willow, Google’s new quantum computing chip. You can see the launch video below.
What is quantum computing?
Unlike traditional digital computers that operate by toggling bits between 0 and 1 (on or off), quantum computers harness the power of qubits—infinitesimally small units that can be on, off, or both simultaneously.
Qubits also leverage quantum entanglement, an enigmatic phenomenon that links the states of particles at most microscopic levels of the universe, no matter how far apart they are, creating a kind of synchronicity at the subatomic level. Quantum computers use such quantum mechanics to calculate highly complex problems that cannot currently be addressed with classical computers.
The blog post, written by Google Quantum AI Founder and Leader Hartmut Neven, who left the AI field to work on quantum computing, announced two major achievements:
The first is that Willow can reduce errors exponentially as we scale up using more qubits. This cracks a key challenge in quantum error correction that the field has pursued for almost 30 years.
Second, Willow performed a standard benchmark computation in under five minutes that would take one of today’s fastest supercomputers 10 septillion (that is, 1025) years — a number that vastly exceeds the age of the Universe.
Perhaps the most astonishing part of the announcement was this:
This mind-boggling number exceeds known timescales in physics and vastly exceeds the age of the universe. It lends credence to the notion that quantum computation occurs in many parallel universes, in line with the idea that we live in a multiverse, a prediction first made by David Deutsch.
In the context of quantum computing, the Many-Worlds Interpretation (MWI) implies that a quantum computer, like Google's Willow chip, performs calculations not just in our observable universe but potentially across multiple parallel universes.
It’s easy to discount such a statement as science fiction. It sounds inconceivable. It is impossible to imagine, or conceive, of 10 septillion years or quantum superposition. But what if the problem lies not in the validity of such a proposition, but the boundaries of our imagination?
This essay will seek to explore that question.
We will explore Chalmers thoughts on the different kinds of conceivability and Descartes meditation on the difference between understanding and imagination to create the on ramp by which we can step outside ourselves and step inside the inconceivable world of the future.
A Unified Theory of Imagination
We are approaching a world with artificial general intelligence, one where the human mind will no longer have a monopoly over what can be imagined and conceived. As such, the future will not bend on its current trajectory, because the current trajectory is a function constrained by a human-only imagination paradigm.
But what is imagination? How does it work?
Imagination draws heavily from memory, particularly semantic and episodic memory. Episodic memory, which stores personal experiences, acts as the "building blocks" for imagination by providing specific details and context, while semantic memory, which stores general knowledge and concepts, provides the framework and structure needed to construct imagined scenarios, essentially working together to create a comprehensive imaginative experience. Both are crucial for imagining future events or novel situations.
Imagination encompasses both imaging (the production of mental images) and creativity (the ability to generate new ideas and solutions). It is a dynamic synergy between these two aspects, involving both concrete, associative thinking and abstract, systematic reasoning.
In their paper, “On the interaction between episodic and semantic representations—Constructing a unified account of imagination” in The Cambridge handbook of the imagination, Irish and Abraham demonstrate:
The output from imagination varies in terms of its episodic and semantic constituents, the relative weighting of which depends upon a host of factors including task-driven variations in content, invocation of prior knowledge and experience, and the accessibility of that information. As such, it demonstrates that the outcome of imagination is best conceptualized as the convergence of episodic and semantic elements, moving toward an all-encompassing and unified theory of imagination.
Thus, to simplify:
The output of an individual’s imagination is the product of an individual’s (or agent’s) lived experience, general knowledge and recall ability.
IF
Imagination of an Agent = Episodic Memory * Semantic Memory * Memory Recall
AND
Episodic Memory = Lived Experience
Semantic Memory = General Knowledge
THEN
Imagination of an Agent = Lived Experience * General Knowledge * RecallImagination exists as intangible mental imaging and creativity in the spatial sprawl of our mind. Imagination forges new ideas, new ideas create goods and services that solve problems, novel solutions spur innovation. Thus, imagination is the basis of innovation. As a result, the future is shaped by what we imagine today. We, in this context, refers to beings with the ability to voluntarily imagine. For the past 70,000 years, humans were the only being with such a faculty.
To increase innovation, you can:
Increase imagination output
Reduce the friction of materializing imagination into real goods and services
A society un-traps imagination by providing the necessary infrastructure to make it easy to materialize imagination into tangible goods and services. This is what Fagerberg and Srholec call an “innovation system” — e.g. patenting, scientific publications, ICT infrastructure, ISO 9000 certifications and availability of finance — in their 2006 paper “The role of “capabilities” in development: Why some countries manage to catch up while others stay poor”. In their analysis, they found a very close correlation between GDP per capita (proxy measure of innovation) and quality of innovation system.
For humans, the best way to increase imagination output is to improve general knowledge. As we can see from the same study, there is a high correlation between education system and GDP per capita.
Pulling these two levers has been the driving force of innovation for the past 300 years. But this paradigm fundamentally operates under the assumption that voluntarily imagination output is solely a product of humans. How do we orient ourselves when that changes?
Conceiving the Inconceivable
If imagination output is a product of general knowledge, lived experience and recall, and we know that artificial intelligence will have exponentially greater general knowledge and recall ability, then it follows that this new being will have exponentially more imagination output than humans.
To understand what will be possible, we have to familiarize ourselves in the space of possibility outside of what we can conceive, but what yet is metaphysically possible.
David Chalmers, in his work on the philosophy of mind and consciousness distinguishes between various types of conceivability to address the complex relationship between what we can conceive and what is metaphysically possible.
In his book Does Conceivability Entail Possibility?, Chalmers defines conceivability as the following:
Conceivability is a property of statements, and the conceivability of a statement is in many cases relative to a speaker or thinker.
Thus, conception (used interchangeable with imagination) is a function of the thinker that is doing the conceiving. From a human perspective, our conceivability is bounded by our imagination. And our imagination is bounded by our semantic and episodic memories, which are in turn bounded by our knowledge and lived experiences and recall.
Yet, we also accept as true that, quite quickly, the human perspective will no longer be the sole perspective capable of conceiving and imaging and materializing such imagination into real innovation.
To understand what this could look like, let’s explore Chalmers thoughts on two types of conceivability: prima facie and ideal conceivability.
Prima facie conceivability refers to what an individual, with their current knowledge and cognitive abilities, can imagine or understand. Ideal conceivability is a philosophical concept that distinguishes between what an average person can conceive and what an ideal, perfectly knowledgeable and rational intelligence can conceive.
For example, If you are not familiar with the concept of cardinality, the idea that the set of real numbers (which includes all decimal numbers, including fractions and irrational numbers) is larger than the set of integers (whole numbers) might seem incomprehensible. This is because you lack the requisite knowledge to understand this mathematical concept.
An Ideal Intelligence, on the other hand, is a hypothetical agent with perfect knowledge, rationality, and cognitive abilities. This agent would possess all necessary information and understanding to comprehend any concept, no matter how complex. This new Ideal Intelligence will be able to imagine and conceive truths that we find inconceivable.
Additionally, there are concepts that we can understand but simply cannot imagine.
In his Sixth Meditation, Descartes muses on the distinction between imagination and pure understanding. He offers a helpful thought experiment to draw the distinction:
When I imagine a triangle, for example, I don’t merely understand that it is a three-sided figure, but I also see the three lines with my mind’s eye as if they were present to me; that is what imagining is. But if I think of a chiliagon [thousand-sided figure], although I understand quite well that it is a figure with a thousand sides, I don’t imagine the thousand sides or see them as if they were present to me. When I think of a body, I usually form some kind of image; so in thinking of a chiliagon I may construct in my mind—·strictly speaking, in my imagination·—a confused representation of some figure. But obviously it won’t be a chiliagon, for it is the very same image that I would form if I were thinking of, say, a figure with ten thousand sides.
We interface with our imagination through our mind’s eye, and mostly do so through two and three dimensional imaging that uses semantic memory (stored as words and natural language) and episodic memory (stored as images) to create a new internally immersive experience in our imagination. Descartes discusses this as well —
In the case of a pentagon, the situation is different. I can of course understand this figure without the help of the imagination (just as I can understand a chiliagon); but I can also imagine a pentagon, by applying my mind’s eye to its five sides and the area they enclose. This imagining, I find, takes more mental effort than understanding does; and that is enough to show clearly that imagination is different from pure understanding.
It is human nature to bias against things we cannot imagine. We know and understand that a fourth dimension to reality exists, but we simply cannot imagine it because our imagination is limited to 2 and 3 dimensional cognition.
AGI and computers won’t be beholden to such constraints. They will be able to imagine a chiliagon just as easily as they will a pentagon, which is simply inconceivable for humans to do.
Conclusion
Our species has developed imagination through millions of years of evolution. While most nonhuman mammals can involuntarily imagine things that don't exist or haven't happened during REM sleep, only humans can intentionally create new objects and scenarios in their minds through prefrontal synthesis.
When did our species develop the ability for prefrontal synthesis? Artifacts dated earlier than 70,000 years ago could have been created by individuals without this capability. However, starting around 70,000 years ago, we see clear evidence of it: composite figurative objects like the lion-man, bone needles with eyes, bows and arrows, musical instruments, constructed shelters, and adorned burials that imply beliefs in an afterlife.
AI is still in the involuntarily imagination phase. The difference between voluntary imagination and involuntary imagination is analogous to the difference between voluntary muscle control and muscle spasm.
AI models do not imagine on purpose, rather require an external trigger (prompted instruction), which they understand as a series of tokens that they react to based on an incredible corpus of data that helps models pattern-match and predict which tokens come next. Natural Language Processing translates these tokens one-by-one to and from natural language to form a coherent input and output to these models.
This is not imagination in the sense as we know it. It is a reflex. It is programmed to its nature, much like how humans involuntarily kick out our leg when we’re hit in the patellar (knee) reflex.
But what happens when AI transitions from reflexive, involuntary responses to voluntary imagination? The trajectory of innovation will fundamentally shift.
For 70,000 years, human imagination has been the engine behind progress — from the creation of tools and art to complex societal systems. Our ability to voluntarily synthesize experiences, ideas, and knowledge has allowed us to conceive and shape the future. Yet, this imaginative capacity is constrained by the limits of our cognition, memory, and lived experience.
AI, on the other hand, is poised to break these constraints. As AI systems advance in their ability to store memory, synthesize vast datasets, and develop contextual "lived" experience, they will move beyond reactive outputs and towards genuine imaginative synthesis. In doing so, AI could achieve something akin to prefrontal synthesis, generating novel concepts, innovations, and possibilities far beyond human capability.
This transition will not only redefine what is conceivable but will also expand the boundaries of innovation. No longer limited by the confines of human memory, AI-driven imagination could create solutions and ideas that are currently inconceivable to us. The future will be shaped not by the limits of human thought, but by an unprecedented collaboration between human and artificial imaginations.
In this new paradigm, our role shifts. Rather than solely driving innovation, we may become curators, collaborators, and guides, helping steer AI-generated possibilities toward meaningful outcomes. To prepare for this future, we must remain open to the inconceivable, to ideas like MWI that sound like science fiction, and embrace a world where the boundaries of imagination are no longer confined to human minds.
In the end, the next frontier of innovation may not be what we can imagine, but what we can imagine with AI.