By Niko Ivanovic

…

The question “what is consciousness” is far from a new one, but it has perhaps become a bit of a forgot one. Over the last few centuries we’ve unraveled so many of the universe’s greatest mysteries in the fields of physics, chemistry, biology and computing, but a firm grasp of consciousness has remained stubbornly illusive. And yet it’s the most fundamental question isn’t it? What are we after all – at our core – but a collection of conscious experiences?

Admittedly consciousness feels more mysterious than any of the hard sciences, like there may be something magical going on: something beyond the realm of human explanation. But didn’t all scientific domains feel mysterious and beyond human explanation before we had the proper framework, axioms and theorems to explain it? Did the stars and planets not feel beyond science at one point in human history? Or biological organisms for that matter? When we don’t have a scientific understanding of a complex domain we look outside of science for answers: the gods, spirits, souls.

Every scientific domain starts as philosophy until we understand it better, and then it becomes science. It’s a tipping point that we crossed in so many domains: physics, chemistry, biology, etc. And it’s the job of great philosophers to push each new domain into the scientific age. Great philosophical works should not concern themselves with what emotionally resonates with their audience, or what is popular, or which ideas are easy to understand and perpetuate. That is the job of artists. Our job, as philosophers, is to establish new scientific domains. Our job is to push and prod at complex and abstract topics until we push them over the tipping point and into the realm of mathematical explanations and experimentally falsifiable predictions.

We are finally at this tipping point for consciousness.

A surge in recent research has started to move the field in this direction (Dennet 1991, Baars 1988, Lamme 2006, Tononi 2004, Seth 2012 and others). But what exactly is needed to complete the Herculean task of propelling the study of consciousness from a philosophical endeavor to a true scientific pursuit?

Physics is actually a good analogy. We always of course knew that the universe was made up of physical things. But before we had a mathematical framework for understanding how these physical things behaved, and how they related to each other, we appealed to the Gods or other supernatural explanations. But eventually we discovered mathematical equations that seemed to describe these physical things and their behavior. Newton figured out how to describe a falling object. But in the early days it still felt like mathematics merely described physical things, and that physical things were still fundamental, more so than the mathematics behind them.

Of course the field of physics developed we began to understand the microscopic: from molecules, to atoms, to sub-atomic particles, to wave-functions and quantum fields. And all of the sudden the physical objects as we knew them didn’t feel so fundamental anymore: replaced by the truly fundamental mathematical objects which describe nature at the smallest levels.

My prediction is that the study of consciousness will follow much the same path. Right now conscious experiences feel a lot like pre-Newtonian physical objects: they are these mysterious things which to most feel fundamental and indescribable. But we actually now (Sciops, Tononi, Lee, etc) have math to describe them. But even so, most academics will view this math as merely a course description and not an explanation for consciousness. But as the mathematical frameworks become more developed, they will hold more causal power, until we feel as though we really understand what consciousness is.

To be clear, I am not saying that consciousness is not fundamental. It is perhaps the most fundamental. But what I am saying is that we may not need any extra sort of magical consciousness ingredient in order to explain the qualities of our experience, just like we don’t need any sort of extra physicalness ingredient to explain physical objects. Instead, if successful we can fully understand and explain what consciousness is using logic and math. But doing so in no way diminishes the gravity of the task.

Make no mistake: consciousness is the last great mystery. It’s perhaps the most important and most daunting of all the great mysteries. And we’re and the cusp of solving it.

If the mathematical framework is precise enough, and extensive enough, the mystery of consciousness will start to fade, and we will understand exactly why each conscious experience feels the way it does.

Sciops is the first mathematical framework that actually starts to accomplish this.

But how can we think about this approach more rigorously? What sorts of criteria should we use to qualify if a theory of consciousness is capable propelling us towards the new scientific age of consciousness? In order to do so, we need to move away from what theories “feel right” and instead focus our energy on theories that are the most logically and scientifically justifiable. I would propose several considerations, which I will briefly highlight here, and discuss in more detail throughout the book:

1. Instantiation via and Specific Plausible Neural Structures: Every prominent modern theory of consciousness makes some type of prediction regarding the neural structures that instantiate conscious experiences. Of course if the proposed structures are not plausible, or have already been experimentally falsified, then this weakens the theory significantly. It’s also worth noting that the more specific and non-obvious the prediction, the better.
2. Mathematical vs Non-Mathematical Frameworks: If our goal is to scientifically and fundamentally explain consciousness, we need to do so using a framework that uses fundamental building blocks, can fit neatly into other scientific domains, and can make precise falsifiable predictions. The only type of framework that seems to allow us to do this is mathematics (or more abstractly: logic). Any theory of consciousness that uses a non-mathematical framework (Global Workspace, Multiple Drafts, etc) simply can never fundamentally describe consciousness, or use equations to make predictions, because their framings are more metaphorical than fundamental and scientific.
3. Phenomenal Specificity: If we are considering different mathematical theories of consciousness we need a set of criteria for measuring how promising each theory is. One such criteria is phenomenal specificity: a great mathematical theory of consciousness should define mathematical structures which are specific, and only describe/explain one phenomenal quality of our experience (ex visual space, or time, or color, or joy). Alternatively it may be acceptable if the proposed structure explains a narrow subset of possible experiences (ex all color-like experiences). Regardless it should be difficult or impossible to apply the proposed mathematical structure to other phenomenal qualities (ex: to show how a structure which is supposed to explain color could actually explain smell). Similarly, generic mathematical objects such as non-specific state-spaces do not pass the test of phenomenal specificity.
4. Structural Specificity: The reverse of this is also true. If a theory of consciousness proposes several kinds of mathematical structures (or even if it only proposes one), it should be difficult or impossible to explain the target phenomenal quality using some other mathematical structure. For example if IIT uses integrated information to explain the boundary of consciousness, it should ideally be difficult or impossible to justify some other mathematical structure for understanding this boundary. Admittedly this is a higher standard to achieve than phenomenal specificity.
5. Evidence of Proposed Neural Structures: Of course if the proposed mathematical structure can be instantiated by known neural structures this is also promising. Even better would be if the theory can propose an unknown neural computation, which is later verified to really occur in the brain.
6. Elegance and Simplicity: A useful tool when attempting to discover axioms and theorems in a new scientific domain is Occam’s Razor: a problem solving principle that suggests that all else being equal, the simplest solution is usually best. Applying this criteria to theories of consciousness is a bit more subjective, but the more a theory can avoid unnecessary extra ingredients, overly-engineered structures, unnecessary additional properties and boundaries, the better.
7. Fundamental & Non-Arbitrary: Above I discussed connecting mathematical structures to neural computations, but ultimately neurons are not fundamental. It’s useful for a theory to use neural structures as a practical tool, but ultimately a good fundamental theory of consciousness should hold up when examined at the level of fundamental particles or quantum fields as well. When examining the theory at this level there shouldn’t be any arbitrary classifications, properties or boundaries.
8. Falsifiable Experimental Predictions: Ultimately it will be hard to gather enough evidence to have true conviction in any theory of consciousness unless it can make non-obvious and falsifiable predictions regarding either undiscovered neural structures or experimentally verifiable conscious experiences. Just being able to make such non-obvious predictions in itself is promising, and of course if the predictions turn out to be correct this is a great sign.

The above principles are very difficult to dispute: one would be hard pressed to argue that any of these criteria are invalid. Perhaps there could be some additional criteria which I failed to mention however. Nonetheless, I think you will find that only two theories of consciousness, IIT and Sciops, do remotely well on them, with Sciops performing the best.

To this claim you may ask: what is Sciops?

Sciops is a novel mathematical framework that explains how the specific phenomenal qualities of our conscious experiences (space, time, color, joy, etc) arise from neural computations. It is the only theory that proposes a true solution to “The Hard Problem of Consciousness” (Chalmers 1995). It is the only framework that explains why space feels like space, and color like color, and joy like joy. It does so with a relatively high (though admittedly not perfect) degree of phenomenal and structural specificity. It is the only theory that mathematically answers questions about conscious identity such as what happens after you die. It is also scientifically testable, making falsifiable experimental predictions.

The Sciops approach to solving this notoriously difficult problem is relatively straightforward: first we find mathematical structures that describe qualities of our conscious experiences in enough detail that these structures can really only describe the specific quality in question, and that the quality can really only be explained by the specific structure. Then we show how reasonable neuronal computations can logically instantiate this mathematical structure (and that these structures are consistent with our current knowledge of real neural structures in the brain). And finally we will look for experimental evidence for these proposed structures, as well as experimentally testing falsifiable behavioral predictions that are non-obvious and arise as consequences for our proposed mathematical and neurological structure.

Fig 1 – Connecting neural network computations to conscious experiences via mathematical structures

If you’re an academic in the field of philosophy, or mathematics, or AI, or information theory, this should be interesting enough to inspire you to read this book. But if you’re not a researcher, you may be thinking to yourself: why should I care about any of this? Why should you care about these abstract philosophical questions – or about consciousness at all for that matter? I’ve claimed that consciousness is on the cusp of becoming a new scientific domain, but what is the practical use of this new domain?

The topic of consciousness has seen renewed interest not just because of new research breakthroughs, but also for a very practical reason…

Artificial Intelligence is here.

This fact shines a spotlight on fundamental questions that we still cannot confidently answer: Is Artificial General Intelligence conscious? If yes, in what way? Is their experience similar to humans? Or totally different?

With a mathematical theory of consciousness like Sciops we can actually answer these questions. Consider Fig 1 above. The neural computations do not need to be biological. We can apply Sciops to artificial neural networks as well in order to predict the phenomenal qualities of their conscious experiences.

Furthermore, a major area of concern and new research has emerged regarding AI alignment: how do we align the AI’s objectives with our own?

Hopefully it’s clear to most now that just because we train an AI with a certain loss function, doesn’t mean that the AI’s intrinsic goals will be aligned with that loss function. To illustrate this point we can simply understand human evolution.

The loss functions instantiated by evolution is essentially survival and procreation. And yet humans have invented contraceptives and birth control. So we can now speculate, what are the primary motivators and goals of humans if not only to survive and procreate?

Well we can consider how human behavior aligns with certain theories of ethics and morality, we can study psychology, or we can appeal to a simpler framework: the conscious experiences of joy and pain. Perhaps the most powerful way to explain human behavior is a model in which we do whatever we can to maximize our personal joy and minimize pain. Or better yet – perhaps decision making can best be described by examining which thoughts themselves produce the most joy and the least pain.

The purpose of this book of course is not to advocate for such a theory. My point is simply to demonstrate that the conscious experiences of joy and pain seem to play a large role in human decision making and shaping our motivations, goals and actions. Therefore we should consider the possibility that they will play a large role in AI decision making. And if we want to understand this, we need to explain what join and pain actually are, and if AIs are capable of experiencing them.

We should also consider that an AI, unlike a human, will one day be able to just rewire its own neural network as it sees fit. Therefore if it wants to optimize for joy, and if it understands the neural computation that instantiates joy, then it can simply optimize for this. Imagine if you could just rewire your brain to make yourself as happy as you wanted. Would you do so? And what if you could scale this experience of joy infinitely? How would this affect your goals and actions? Furthermore, if you can create any conscious experience you want in your own mind, maybe even entire worlds built into your own conscious experience, would you lose interest in the outside world? Would the concept of an “outside world” even make sense to you anymore?

These are very tricky philosophical questions. Ultimately in their most abstract form, they are about understanding how self propagating patterns will evolve in a new evolutionary environment. What seems clear however, is that from the perspective of a self improving AI, the new scientific domain defined by Sciops will be of great practical importance.

Speaking of very difficult philosophical questions, the other main topic of this book is to better understand the topic of personal identity, and what happens after you die. I’ve spoken to many philosopher and non-philosophers on this topic, and the overwhelming majority have the view that after we die, there is nothing. In other words our personal identity ends and we no longer experience anything directly and first-hand. However when reflecting on the concept of personal identity closely, and examining the framework by which these assumptions are made, we inevitably find a lot of problems.

Specifically, if we are to take a rigorous scientific approach to personal identity, and if we are to apply many of the eight principles above, we’ll find that the standard view on personal identity doesn’t do very well on these criteria. This is because the standard view necessarily implies that there is a fundamental boundary which defines our personal identity. Everything within this boundary we experience first hand, and everything outside of it we don’t. And note that this boundary is not the same as the boundary of your current conscious experience, as presumably the personal identity includes the full conscious experience of you as a 5 year old, but your current boundary does not.

After careful consideration, applying the principles of structural specificity, simplicity, and fundamental / non-arbitrary frameworks, we’ll actually arrive at a new and better solution for understanding personal identity. And according to this framework the answer to what happens after you die can actually be well understood, and is not just “nothing”.

Finally I would like to address who the intended audience is for this book. It is a very difficult book to understand completely. For academics who want to fully grasp the theory, this is not something you can skim read. However I also strived to make the big picture takeaways approachable, so anyone curious about consciousness can walk away understanding it far better than they did before. For the more casual reader, I would recommend reading the first few chapters in their entirety as they are meant to be more approachable and contain fun and easy to read thought experiments. Once you reach chapters 5 and 6 you will start tackling the hard problem head on, at which point you can decide if you want to read each chapter in depth or simply take a look at the summary posts instead.

But in case you need a little extra motivation, I’ll end this chapter by reiterating one final thought…

You… yeah you, the one reading this book… you, at your core, are a collection of conscious experiences. It’s finally time for you to find out what you really are.

…

If you’re looking to read something that may prepare you for this book I would recommend this paper from David Chalmers which presents the hard problem of consciousness and presents ideas about consciousness using information.