Consciousness Is A Spectrum
Is AI Conscious?
In academia and industry alike, the question is becoming increasingly common. This is a natural curiosity as a new technology accelerates up its development curve. Answers are considered, like this one, which argues:
"Consciousness and especially the concept of self are fixtures of what makes humans “special”; and our very complex and still flawed model of the world gives us these gifts, as well as shape how we perceive and interact with the world."
This line of reasoning is a tad arrogant to the extent that it places consciousness as the exclusive property of the human species. Does a dog have a sense of self? (It certainly seems so). Worse, the hypothesis is unprovable as the author notes.1
Moreover, the only reason I can presume that the readers of this essay and others who participate in the conversation are conscious is the fact that I perceive myself as conscious. Since other human beings seem superficially similar to me (being the same species and all), my conclusions seem defensible.
Of course, that neither proves I am conscious or that other humans are conscious. Nor does this offer any empirical falsifiability conditions. That leaves us with a question Douglas Hofstadter, in Godel, Escher, Bach, might have answered with “MU!”2 “Are humans conscious?” probably deserves a “MU” and “is AI conscious?” or “is GPT-3 conscious?” definitely deserves a “MU.”3
Better Questions
So clearly, we need some better questions. At a minimum, we need questions our current position in technological and philosophical awareness can address.
Instead, we can grapple with an alternative line of thinking, namely “what are the advantages of consciousness (or behaving in a manner consistent with being conscious)?”
For one, there is the opportunity to experience the nebulous and poorly-understood “qualia.”4 Some of these qualia are pleasant, and facilitated by self-awareness and the general idea of consciousness.
Better still, this awareness of self facilitates introspection, philosophy, and a broader examination of the nature of our own existence.5 Besides the intellectual satisfaction (another example of qualia) and knowledge expansion it provides, it provides a roadmap towards larger abstractions.
These abstractions can include simulations of intelligence (e.g. software) and with more complex simulations, the exploration of emergent properties of everything from agent-based models to the simulation of a rich, modern society.
Not Human, But…
Extending this line of questioning, we no longer focus on who or what is conscious. Instead, we consider who or what possesses the advantages of consciousness discussed above. After all, human beings who possess these advantages needn’t prove their consciousness, so why should we place a higher burden on other species or even algorithms?
What types of entities possess the relevant advantages? Can GPT-36 create text that inspires novel insight and self-awareness? Does this suggest a broader examination of existence? If GPT-3 is asked to discuss morality or mortality, who exactly has pondered these questions? Have its creators? Have its interpreters? The developers who use its API?
What about aggregate entities? Clearly a neuron is not conscious (by any definition), but billions of neurons, each with thousands of connections, seem to possess the properties of consciousness when assembled as a human brain. What about the ants in GEB who form words and generally exhibit conscious patterns of activity (or an actual ant farm without the fictional styling)? What about any decentralized system capable of adaptation and response?
Perhaps consciousness is a spectrum? The question should not be “is ___ conscious?” Instead, the question should be “what types of qualia can this entity experience?” To what extent is the entity aware of self and capable of grappling with the nature and morality of its existence and the existence of others? What if creating AI facilitates simulation of particular interactions and exchanges under various societal norms and rules? Is the species that deploys AI for such a purpose reaching a higher form of consciousness, now better able to grapple with the nature of existence and intelligent interaction?
Is This a Simulation?
This leads to another question, namely whether our current universe and our experiences are simply part and parcel of the simulation of a more sophisticated species.7 Leaving aside the hypothesis’ non-scientific characteristics (cannot be falsified), it yields some relevant questions.
If the universe is a simulation, and its creators decided upon (or failed to prevent) the pain and suffering within the prior and present human experience, what morality might justify or explain that choice? Simply, what were our creators thinking?
Most moral justifications are stories imploring us to accept our fate as it is, engage in meditative practice, and extract as much joy as we can access at any given moment. Moreover, we should aspire to better our situation (Choose to go to the moon! Build agency-increasing BCI!)
There are also well-worn arguments that we are merely insects or grains of sand to the superior beings that create and observe. But if so, why hasn’t their own moral progression enlightened them to our plight? Humanity has evolved to condemn slavery, and perhaps one day we’ll grow meat in labs and remedy the injustice of industrial slaughter of mammals? Perhaps we’ll eventually recognize immorality in our treatment of certain interconnected ecosystems. Surely8 a species capable of simulating our universe has a more sophisticated set of ethical principles?
Faults in Our Paradigm
Perhaps our creators were immoral or sadistic. Perhaps they scaled too quickly, generating too many universes to ponder the morality of each. Perhaps they lacked agency initially (as we will when first we create something that qualifies as digital consciousness, if we haven’t already). Perhaps they wanted to study the creation of more advanced digital consciousness or didn’t realize their simulations might develop consciousness (or even life!).
Perhaps our creators truly did aspire to create the best of all possible worlds, and having explored all superior possibilities, settled on ours.9 Or perhaps they had a goal beyond our feeble comprehension.
An Alternative Paradigm
Let’s accept that the universe is a simulation for the sake of argument, Karl Popper be-damned. Let’s also assume our creators are not sadists, immoral sociopaths, or otherwise depraved. Finally, let’s assume that our creators have some morality that generally is comprehensible by our puny primate brains.
There is another possibility. Perhaps our creators consider us not inconsequential, but less consequential. Given their obvious superiority computationally and scientifically, this might be similar to how we view animals. Most of us would agree that torturing and killing animals is grotesque and horrific. However, most of us would also agree that utilizing animals in lab experiments and medical testing to benefit human beings is morally acceptable. Perhaps our creators concur.
So what experiments are they running?
Our creators would have recognized long ago that systems generate emergent properties without any constituent component possessing that property. An individual neuron cannot possess a personality disorder, but a brain composed of billions can. A corporation composed of mostly well-intentioned, generous, thoughtful people can nonetheless manifest the properties of an empathy-less sociopath. Individuals lacking any particular animus can nonetheless form a society with structural biases.
What if our creators want to explore emergent properties of biological and social structures? What if they hoped to encounter possible erroneous paradigms before becoming inextricably bound by them?10
There would be innumerable simulations, possibly too many to monitor as discussed. Our creators would have needed to create and adopt general rules about the creation of digital consciousness. Unfortunately, rules are often insufficient to predict emergent properties, so the myriad simulations might have provided some insight regarding the proper construction of their society.
Extending Consciousness
If our creators accepted the possibility of emergent properties, perhaps they created digital consciousness to extract them? This would maximize the agency of their higher-order existence.
Given the initial discussion of consciousness as a spectrum, their heightened awareness of social philosophy and existential risks thereto would make them “more conscious.”
A conscious species capable of selecting desirable emergent properties rather than the insidious, potentially ruinous structural flaws of lesser societies might perhaps elect an emergent property that leads to a higher-order consciousness. Each human becomes a neuron in a more advanced, collective intelligence.11
As we construct our own artificial intelligence and simulate more complex algorithms with emergent properties of their own, perhaps we too will achieve increased awareness of our existence, our flaws, and our existential risks. For the first time in our species’ history, we will actively increase our own consciousness.
The question is not “is ____ conscious?” The question will be “how conscious?”
1 Karl Popper argued that absent the possibility of empirical falsification, a theory does not fall within the scope of “science.” In other words, unless you can enumerate the type of evidence that would lead the rejection of the theory and conditions under which that evidence could be gathered if the theory truly was false, the theory is not a candidate for scientific proof or acceptance.
2 In his treatise on reductionism vs. holism, when a monk is asked if a dog has Buddha-nature, he “denies the legitimacy of the question” by answering “MU!” Basically, in GEB’s parlance, he “unasked” the question.
3 By some definitions and as argued in GEB, mathematics, via Godel’s incompleteness theorem shows a level of self-awareness. Perhaps less self-awareness than a human being, but more self-awareness than, say, your desk?
4 Literally defined as “individual instances of subjective, conscious experience.” The pseudo-recursion in my prose notwithstanding (conscious beings can experience qualia, which are defined as experiences of conscious beings), it does seem like a helpful idea. This is also the David Chalmers' "hard problem of consciousness." It's one thing to sense. It's another to experience.
5 And the existence of other entities.
6 Transformer-based, state-of-the-art, autoregressive language model capable of generating text similar to that composed by a human being. Read more here.
7 Commonly, the “simulation hypothesis.” Also, not falsifiable, arguable only with inductive or rationalist reasoning, much to the consternation of Sabine Hossenfelder and others.
8 Yes, I called you Shirley.
9 Roughly the idea in Scott Alexander’s Unsong.
10 For instance, any of the paradigms outlined in Meditations on Moloch…having the “god’s eye” view of a simulation would offer tremendous foresight.
11 Sounds a little like the Borg from Star Trek?