Co-evolution of Humans & Artificial Intelligence

In my last post, I wrote a little bit about the concept of personal identity in terms of what some philosophers have emphasized and my take on it.  I wrote that post in response to an interesting blog post written by James DiGiovanna over at A Philosopher’s Take.  James has written another post related to the possible consequences of integrating artificial intelligence into our societal framework, but rather than discussing personal identity as it relates to artificial intelligence, he discussed how the advancements made in machine learning and so forth are leading to the future prospects of effective companion AI, or what he referred to as programmable friends.  The main point he raised in that post was the fact that programmable friends would likely have a very different relationship dynamic with us compared with our traditional (human) friends.  James also spoke about companion AI  in terms of their also being laborers (as well as being friends) but for the purposes of this post I won’t discuss these laborer aspects of future companion AI (even if the labor aspect is what drives us to make companion AI in the first place).  I’ll be limiting my comments here to the friendship or social dynamic aspects only.  So what aspects about programmable AI should we start thinking about?

Well for one, we don’t currently have the ability to simply reprogram a friend to be exactly as we want them to be, in order to avoid conflicts entirely, to share every interest we have, etc., but rather there is a bit of a give-and-take relationship dynamic that we’re used to dealing with.  We learn new ways of behaving and looking at the world and even new ways of looking at ourselves when we have friendships with people that differ from us in certain ways.  Much of the expansion and beneficial evolution of our perspectives are the result of certain conflicts that arise between ourselves and our friends, where different viewpoints can clash against one another, often forcing a person to reevaluate their own position based on the value they place on the viewpoints of their friends.  If we could simply reprogram our friends, as in the case with some future AI companions, what would this do to our moral, psychological and intellectual growth?  There would be some positive effects I’m sure (from having less conflict in some cases and thus an increase in short term happiness), but we’d definitely be missing out on a host of interpersonal benefits that we gain from having the types of friendships that we’re used to having (and thus we’d likely have less overall happiness as a result).

We can see where evolution ties in to all this, whereby we have evolved as a social species to interact with others that are more or less like us, and so when we envision these possible future AI friendships, it should become obvious why certain problems would be inevitable largely because of the incompatibility with our evolved social dynamic.  To be sure, some of these problems could be mitigated by accounting for them in the initial design of the companion AI.  In general, this could be done by making the AI more like humans in the first place and this could be something advertised as some kind of beneficial AI “social software package” so people looking to get companion AI would be inclined to get this version even if they had the choice to go for the entirely reprogrammable version.

Some features of a “social software package” could be things like a limit on the number of ways the AI could be reprogrammed such that only very serious conflicts could be avoided through reprogramming, but without the ability to avoid all conflicts.  It could be such that the AI are able to have a weight on certain opinions, just as we do, and to be more assertive with regard to certain propositions and so forth.  Once the AI has learned its human counterpart’s personality, values, opinions, etc., it could also be programmed with the ability to intentionally challenge that human by offering different points of view and by its using the Socratic method (at least from time to time).  If people realized that they could possibly gain wisdom, knowledge, tolerance, and various social aptitudes from their companion AI, I would think that would be a marked selling point.

Another factor that I think will likely play a role in mitigating the possible social dynamic clash between companion AI (that are programmable) and humans is the fact that humans are also likely to become more and more integrated with AI technology generally.  That is, as humans are continuing to make advancements in AI technology, we are also likely to integrate a lot of that technology into ourselves, to make humans more or less into cyborgs a.k.a. cybernetic organisms.  If we see the path we’re on already with all the smart phones, apps, and other gadgets and computer systems that have started to become extensions of ourselves, we can see that the next obvious step (which I’ve mentioned elsewhere, here and here) is to remove the external peripherals so that they are directly accessible via our consciousness with no need of interfacing with external hardware and so forth.  If we can access “the cloud” with our minds (say, via bluetooth or the like), then the apps and all the fancy features can become a part of our minds, adding to the ways that we will be able to think, providing an internet worth of knowledge at our cognitive disposal, etc.  I could see this technology eventually allowing us to change our senses and perceptions, including an ability to add virtual objects that are amalgamated with the rest of the external reality that we perceive (such as adding virtual friends that we see and interact with that aren’t physically present outside of our minds even though they appear to be).

So if we start to integrate these kinds of technologies into ourselves as we are also creating companion AI, then we may also end up with the ability to reprogram ourselves alongside those programmable companion AI.  In effect, our own qualitatively human social dynamic may start to change markedly and become more and more compatible with that of the future AI.  The way I think this will most likely play out is that we will try to make AI more like us as we try to make us more like AI, where we co-evolve with one another, trying to share advantages with one another and eventually becoming indistinguishable from one another.  Along this journey however we will also become very different from the way we are now, and after enough time passes, we’ll likely change so much that we’d be unrecognizable to people living today.  My hope is that as we use AI to also improve our intelligence and increase our knowledge of the world generally, we will also continue to improve on our knowledge of what makes us happiest (as social creatures or otherwise) and thus how to live the best and most morally fruitful lives that we can.  This will include improving our knowledge of social dynamics and the ways that we can maximize all the interpersonal benefits therein.  Artificial intelligence may help us to accomplish this however paradoxical or counter-intuitive that may seem to us now.

Advertisement

The illusion of Persistent Identity & the Role of Information in Identity

After reading and commenting on a post at “A Philosopher’s Take” by James DiGiovanna titled Responsibility, Identity, and Artificial Beings: Persons, Supra-persons and Para-persons, I decided to expand on the topic of personal identity.

Personal Identity Concepts & Criteria

I think when most people talk about personal identity, they are referring to how they see themselves and how they see others in terms of personality and some assortment of (usually prominent) cognitive and behavioral traits.  Basically, they see it as what makes a person unique and in some way distinguishable from another person.  And even this rudimentary concept can be broken down into at least two parts, namely, how we see ourselves (self-ascribed identity) and how others see us (which we could call the inferred identity of someone else), since they are likely going to differ.  While most people tend to think of identity in these ways, when philosophers talk about personal identity, they are usually referring to the unique numerical identity of a person.  Roughly speaking, this amounts to basically whatever conditions or properties that are both necessary and sufficient such that a person at one point in time and a person at another point in time can be considered the same person — with a temporal continuity between those points in time.

Usually the criterion put forward for this personal identity is supposed to be some form of spatiotemporal and/or psychological continuity.  I certainly wouldn’t be the first person to point out that the question of which criterion is correct has already framed the debate with the assumption that a personal (numerical) identity exists in the first place and even if it did exist, it also assumes that the criterion is something that would be determinable in some way.  While it is not unfounded to believe that some properties exist that we could ascribe to all persons (simply because of what we find in common with all persons we’ve interacted with thus far), I think it is far too presumptuous to believe that there is a numerical identity underlying our basic conceptions of personal identity and a determinable criterion for it.  At best, I think if one finds any kind of numerical identity for persons that persist over time, it is not going to be compatible with our intuitions nor is it going to be applicable in any pragmatic way.

As I mention pragmatism, I am sympathetic to Parfit’s views in the sense that regardless of what one finds the criteria for numerical personal identity to be (if it exists), the only thing that really matters to us is psychological continuity anyway.  So despite the fact that Locke’s view — that psychological continuity (via memory) was the criterion for personal identity — was in fact shown to be based on circular and illogical arguments (per Butler, Reid and others), nevertheless I give applause to his basic idea.  Locke seemed to be on the right track, in that psychological continuity (in some sense involving memory and consciousness) is really the essence of what we care about when defining persons, even if it can’t be used as a valid criterion in the way he proposed.

(Non) Persistence & Pragmatic Use of a Personal Identity Concept

I think that the search for, and long debates over, what the best criterion for personal identity is, has illustrated that what people have been trying to label as personal identity should probably be relabeled as some sort of pragmatic pseudo-identity. The pragmatic considerations behind the common and intuitive conceptions of personal identity have no doubt steered the debate pertaining to any possible criteria for helping to define it, and so we can still value those considerations even if a numerical personal identity doesn’t really exist (that is, even if it is nothing more than a pseudo-identity) and even if a diachronic numerical personal identity does exist but isn’t useful in any way.

If the object/subject that we refer to as “I” or “me” is constantly changing with every passing moment of time both physically and psychologically, then I tend to think that the self (that many people ascribe as the “agent” of our personal identity) is an illusion of some sort.  I tend to side more with Hume on this point (or at least James Giles’ fair interpretation of Hume) in that my views seem to be some version of a no-self or eliminativist theory of personal identity.  As Hume pointed out, even though we intuitively ascribe a self and thereby some kind of personal identity, there is no logical reason supported by our subjective experience to think it is anything but a figment of the imagination.  This illusion results from our perceptions flowing from one to the next, with a barrage of changes taking place with this “self” over time that we simply don’t notice taking place — at least not without critical reflection on our past experiences of this ever-changing “self”.  The psychological continuity that Locke described seems to be the main driving force behind this illusory self since there is an overlap in the memories of the succession of persons.

I think one could say that if there is any numerical identity that is associated with the term “I” or “me”, it only exists for a short moment of time in one specific spatio-temporal slice, and then as the next perceivable moment elapses, what used to be “I” will become someone else, even if the new person that comes into being is still referred to as “I” or “me” by a person that possesses roughly the same configuration of matter in its body and brain as the previous person.  Since the neighboring identities have an overlap in accessible memory including autobiographical memories, memories of past experiences generally, and the memories pertaining to the evolving desires that motivate behavior, we shouldn’t expect this succession of persons to be noticed or perceived by the illusory self because each identity has access to a set of memories that is sufficiently similar to the set of memories accessible to the previous or successive identity.  And this sufficient degree of similarity in those identities’ memories allow for a seemingly persistent autobiographical “self” with goals.

As for the pragmatic reasons for considering all of these “I”s and “me”s to be the same person and some singular identity over time, we can see that there is a causal dependency between each member of this “chain of spatio-temporal identities” that I think exists, and so treating that chain of interconnected identities as one being is extremely intuitive and also incredibly useful for accomplishing goals (which is likely the reason why evolution would favor brains that can intuit this concept of a persistent “self” and the near uni-directional behavior that results from it).  There is a continuity of memory and behaviors (even though both change over time, both in terms of the number of memories and their accuracy) and this continuity allows for a process of conditioning to modify behavior in ways that actively rely on those chains of memories of past experiences.  We behave as if we are a single person moving through time and space (and as if we are surrounded by other temporally extended single person’s behaving in similar ways) and this provides a means of assigning ethical and causal responsibility to something or more specifically to some agent.  Quite simply, by having those different identities referenced under one label and physically attached to or instantiated by something localized, that allows for that pragmatic pseudo-identity to persist over time in order for various goals (whether personal or interpersonal/societal) to be accomplished.

“The Persons Problem” and a “Speciation” Analogy

I came up with an analogy that I thought was very fitting to this concept.  One could analogize this succession of identities that get clumped into one bulk pragmatic-pseudo-identity with the evolutionary concept of speciation.  For example, a sequence of identities somehow constitute an intuitively persistent personal identity, just as a sequence of biological generations somehow constitute a particular species due to the high degree of similarity between them all.  The apparent difficulty lies in the fact that, at some point after enough identities have succeeded one another, even the intuitive conception of a personal identity changes markedly to the point of being unrecognizable from its ancestral predecessor, just as enough biological generations transpiring eventually leads to what we call a new species.  It’s difficult to define exactly when that speciation event happens (hence the species problem), and we have a similar problem with personal identity I think.  Where does it begin and end?  If personal identity changes over the course of a lifetime, when does one person become another?  I could think of “me” as the same “me” that existed one year ago, but if I go far enough back in time, say to when I was five years old, it is clear that “I” am a completely different person now when compared to that five year old (different beliefs, goals, worldview, ontology, etc.).  There seems to have been an identity “speciation” event of some sort even though it is hard to define exactly when that was.

Biologists have tried to solve their species problem by coming up with various criteria to help for taxonomical purposes at the very least, but what they’ve wound up with at this point is several different criteria for defining a species that are each effective for different purposes (e.g. biological-species concept, morpho-species concept, phylogenetic-species concept, etc.), and without any single “correct” answer since they are all situationally more or less useful.  Similarly, some philosophers have had a persons problem that they’ve been trying to solve and I gather that it is insoluble for similar “fuzzy boundary” reasons (indeterminate properties, situationally dependent properties, etc.).

The Role of Information in a Personal Identity Concept

Anyway, rather than attempt to solve the numerical personal identity problem, I think that philosophers need to focus more on the importance of the concept of information and how it can be used to try and arrive at a more objective and pragmatic description of the personal identity of some cognitive agent (even if it is not used as a criterion for numerical identity, since information can be copied and the copies can be distinguished from one another numerically).  I think this is especially true once we take some of the concerns that James DiGiovanna brought up concerning the integration of future AI into our society.

If all of the beliefs, behaviors, and causal driving forces in a cognitive agent can be represented in terms of information, then I think we can implement more universal conditioning principles within our ethical and societal framework since they will be based more on the information content of the person’s identity without putting as much importance on numerical identity nor as much importance on our intuitions of persisting people (since they will be challenged by several kinds of foreseeable future AI scenarios).

To illustrate this point, I’ll address one of James DiGiovanna’s conundrums.  James asks us:

To give some quick examples: suppose an AI commits a crime, and then, judging its actions wrong, immediately reforms itself so that it will never commit a crime again. Further, it makes restitution. Would it make sense to punish the AI? What if it had completely rewritten its memory and personality, so that, while there was still a physical continuity, it had no psychological content in common with the prior being? Or suppose an AI commits a crime, and then destroys itself. If a duplicate of its programming was started elsewhere, would it be guilty of the crime? What if twelve duplicates were made? Should they each be punished?

In the first case, if the information constituting the new identity of the AI after reprogramming is such that it no longer needs any kind of conditioning, then it would be senseless to punish the AI — other than to appease humans that may be angry that they couldn’t themselves avoid punishment in this way, due to having a much slower and less effective means of reprogramming themselves.  I would say that the reprogrammed AI is guilty of the crime, but only if its reprogrammed memory still included information pertaining to having performed those past criminal behaviors.  However, if those “criminal memories” are now gone via the reprogramming then I’d say that the AI is not guilty of the crime because the information constituting its identity doesn’t match that of the criminal AI.  It would have no recollection of having committed the crime and so “it” would not have committed the crime since that “it” was lost in the reprogramming process due to the dramatic change in information that took place.

In the latter scenario, if the information constituting the identity of the destroyed AI was re-instantiated elsewhere, then I would say that it is in fact guilty of the crime — though it would not be numerically guilty of the crime but rather qualitatively guilty of the crime (to differentiate between the numerical and qualitative personal identity concepts that are embedded in the concept of guilt).  If twelve duplicates of this information were instantiated into new AI hardware, then likewise all twelve of those cognitive agents would be qualitatively guilty of the crime.  What actions should be taken based on qualitative guilt?  I think it means that the AI should be punished or more specifically that the judicial system should perform the reconditioning required to modify their behavior as if it had committed the crime (especially if the AI believes/remembers that it has committed the crime), for the better of society.  If this can be accomplished through reprogramming, then that would be the most rational thing to do without any need for traditional forms of punishment.

We can analogize this with another thought experiment with human beings.  If we imagine a human that has had its memories changed so that it believes it is Charles Manson, has all of Charles Manson’s memories and intentions, then that person should be treated as if they are Charles Manson and thus incarcerated/punished accordingly to rehabilitate them or protect the other members of society.  This is assuming of course that we had reliable access to that kind of mind-reading knowledge.  If we did, the information constituting the identity of that person would be what is most important — not what the actual previous actions of the person were — because the “previous person” was someone else, due to that gross change in information.

Conscious Realism & The Interface Theory of Perception

A few months ago I was reading an interesting article in The Atlantic about Donald Hoffman’s Interface Theory of Perception.  As a person highly interested in consciousness studies, cognitive science, and the mind-body problem, I found the basic concepts of his theory quite fascinating.  What was most interesting to me was the counter-intuitive connection between evolution and perception that Hoffman has proposed.  Now it is certainly reasonable and intuitive to assume that evolutionary natural selection would favor perceptions that are closer to “the truth” or closer to the objective reality that exists independent of our minds, simply because of the idea that perceptions that are more accurate will be more likely to lead to survival than perceptions that are not accurate.  As an example, if I were to perceive lions as inert objects like trees, I would be more likely to be naturally selected against and eaten by a lion when compared to one who perceives lions as a mobile predator that could kill them.

While this is intuitive and reasonable to some degree, what Hoffman actually shows, using evolutionary game theory, is that with respect to organisms with comparable complexity, those with perceptions that are closer to reality are never going to be selected for nearly as much as those with perceptions that are tuned to fitness instead.  More so, truth in this case will be driven to extinction when it is up against perceptual models that are tuned to fitness.  That is to say, evolution will select for organisms that perceive the world in a way that is less accurate (in terms of the underlying reality) as long as the perception is tuned for survival benefits.  The bottom line is that given some specific level of complexity, it is more costly to process more information (costing more time and resources), and so if a “heuristic” method for perception can evolve instead, one that “hides” all the complex information underlying reality and instead provides us with a species-specific guide to adaptive behavior, that will always be the preferred choice.

To see this point more clearly, let’s consider an example.  Let’s imagine there’s an animal that regularly eats some kind of insect, such as a beetle, but it needs to eat a particular sized beetle or else it has a relatively high probability of eating the wrong kind of beetle (and we can assume that the “wrong” kind of beetle would be deadly to eat).  Now let’s imagine two possible types of evolved perception: it could have really accurate perceptions about the various sizes of beetles that it encounters so it can distinguish many different sizes from one another (and then choose the proper size range to eat), or it could evolve less accurate perceptions such that all beetles that are either too small or too large appear as indistinguishable from one another (maybe all the wrong-sized beetles whether too large or too small look like indistinguishable red-colored blobs) and perhaps all the beetles that are in the ideal size range for eating appear as green-colored blobs (that are again, indistinguishable from one another).  So the only discrimination in this latter case of perception is between red and green colored blobs.

Both types of perception would solve the problem of which beetles to eat or not eat, but the latter type (even if much less accurate) would bestow a fitness advantage over the former type, by allowing the animal to process much less information about the environment by not focusing on relatively useless information (like specific beetle size).  In this case, with beetle size as the only variable under consideration for survival, evolution would select for the organism that knows less total information about beetle size, as long as it knows what is most important about distinguishing the edible beetles from the poisonous beetles.  Now we can imagine that in some cases, the fitness function could align with the true structure of reality, but this is not what we ever expect to see generically in the world.  At best we may see some kind of overlap between the two but if there doesn’t have to be any then truth will go extinct.

Perception is Analogous to a Desktop Computer Interface

Hoffman analogizes this concept of a “perception interface” with the desktop interface of a personal computer.  When we see icons of folders on the desktop and drag one of those icons to the trash bin, we shouldn’t take that interface literally, because there isn’t literally a folder being moved to a literal trash bin but rather it is simply an interface that hides most if not all of what is really going on in the background — all those various diodes, resistors and transistors that are manipulated in order to modify stored information that is represented in binary code.

The desktop interface ultimately provides us with an easy and intuitive way of accomplishing these various information processing tasks because trying to do so in the most “truthful” way — by literally manually manipulating every diode, resistor, and transistor to accomplish the same task — would be far more cumbersome and less effective than using the interface.  Therefore the interface, by hiding this truth from us, allows us to “navigate” through that computational world with more fitness.  In this case, having more fitness simply means being able to accomplish information processing goals more easily, with less resources, etc.

Hoffman goes on to say that even though we shouldn’t take the desktop interface literally, obviously we should still take it seriously, because moving that folder to the trash bin can have direct implications on our lives, by potentially destroying months worth of valuable work on a manuscript that is contained in that folder.  Likewise we should take our perceptions seriously, even if we don’t take them literally.  We know that stepping in front of a moving train will likely end our conscious experience even if it is for causal reasons that we have no epistemic access to via our perception, given the species-specific “desktop interface” that evolution has endowed us with.

Relevance to the Mind-body Problem

The crucial point with this analogy is the fact that if our knowledge was confined to the desktop interface of the computer, we’d never be able to ascertain the underlying reality of the “computer”, because all that information that we don’t need to know about that underlying reality is hidden from us.  The same would apply to our perception, where it would be epistemically isolated from the underlying objective reality that exists.  I want to add to this point that even though it appears that we have found the underlying guts of our consciousness, i.e., the findings in neuroscience, it would be mistaken to think that this approach will conclusively answer the mind-body problem because the interface that we’ve used to discover our brains’ underlying neurobiology is still the “desktop” interface.

So while we may think we’ve found the underlying guts of “the computer”, this is far from certain, given the possibility of and support for this theory.  This may end up being the reason why many philosophers claim there is a “hard problem” of consciousness and one that can’t be solved.  It could be that we simply are stuck in the desktop interface and there’s no way to find out about the underlying reality that gives rise to that interface.  All we can do is maximize our knowledge of the interface itself and that would be our epistemic boundary.

Predictions of the Theory

Now if this was just a fancy idea put forward by Hoffman, that would be interesting in its own right, but the fact that it is supported by evolutionary game theory and genetic algorithm simulations shows that the theory is more than plausible.  Even better, the theory is actually a scientific theory (and not just a hypothesis), because it has made falsifiable predictions as well.  It predicts that “each species has its own interface (with some similarities between phylogenetically related species), almost surely no interface performs reconstructions (read the second link for more details on this), each interface is tailored to guide adaptive behavior in the relevant niche, much of the competition between and within species exploits strengths and limitations of interfaces, and such competition can lead to arms races between interfaces that critically influence their adaptive evolution.”  The theory predicts that interfaces are essential to understanding evolution and the competition between organisms, whereas the reconstruction theory makes such understanding impossible.  Thus, evidence of interfaces should be widespread throughout nature.

In his paper, he mentions the Jewel beetle as a case in point.  This beetle has a perceptual category, desirable females, which works well in its niche, and it uses it to choose larger females because they are the best mates.  According to the reconstructionist thesis, the male’s perception of desirable females should incorporate a statistical estimate of the true sizes of the most fertile females, but it doesn’t do this.  Instead, it has a category based on “bigger is better” and although this bestows a high fitness behavior for the male beetle in its evolutionary niche, if it comes into contact with a “stubbie” beer bottle, it falls into an infinite loop by being drawn to this supernormal stimuli since it is smooth, brown, and extremely large.  We can see that the “bigger is better” perceptual category relies on less information about the true nature of reality and instead chooses an “informational shortcut”.  The evidence of supernormal stimuli which have been found with many species further supports the theory and is evidence against the reconstructionist claim that perceptual categories estimate the statistical structure of the world.

More on Conscious Realism (Consciousness is all there is?)

This last link provided here shows the mathematical formalism of Hoffman’s conscious realist theory as proved by Chetan Prakash.  It contains a thorough explanation of the conscious realist theory (which goes above and beyond the interface theory of perception) and it also provides answers to common objections put forward by other scientists and philosophers on this theory.