Philosophy Essay: Against Representation

**This was for my ARTI/PHIL/PSYC 3550 class as my final essay and in summary, this essay argues against the classical models of cognitive science where representation is seen as essential to artificial intelligence. Final grade was a B+, so there are points where there might be thin spots, but overall, if y’all have any thoughts or points of disagreements, leave’em in the comments below. :)**

Introduction

Throughout much of our study and development of human cognition and its replication in the various forms of artificial intelligence, there has been an underlying assumption from which we have based our work on– that to replicate intelligent thought and intelligent behavior requires extensive representation. However, representation isn’t only unnecessary but that it would actually be detrimental to our efforts to create true artificial intelligence and our understanding of our own cognition if we keep the level of representation that we currently implement in our machines.

Embodied Cognition

“A machine can exist in an abstract plane where it can crunch numbers detached from the physical world. We, on the other hand, are enmeshed in it.”

A key difference between us and the machines we create is in our embodiment. A machine can exist in an abstract plane where it can crunch numbers detached from the physical world. We, on the other hand, are enmeshed in it. Our bodies constantly feed us information about the environment and we have no way of disconnecting from it. Our cognition is fundamentally embodied and our bodies affect the way we think and vice versa. 

This can be shown in our use of language. Language is a cognitive tool that both permits the expression of and the limitations of our cognition and how we use it tells us a lot about how we think. Since so much of our cognition is expressed and shaped within the constraints of our language, Language shows us just how much of our cognition is rooted in our bodies. For example, warmth is associated with affection (“warming up to someone”), weightiness is associated with value and importance, and exposure to immoral or unethical instances causes a feeling of uncleanliness [1]. In these cases, it shows that there is a reciprocal relationship between our cognition and our physiological form. 

This strong connection between body and mind means we often never have to completely hold our thought processes in our brains. For example, if we are working out a math problem, we can write it down and refer back to earlier steps to help complete later steps. This way, we only have to store the current step of the process in our brain, make the immediately relevant calculation and then the information can be stored and kept track of on an external medium. Since our brains are limited in its energy stores, it saves a lot of energy by processing information in small chunks and to externalise it like this. Rather than wasting time and energy replicating a model of the problem space internally to manipulate, we can just reference reality to inform us on what to do next. This ties back to cognitive technology [2] where external mediums can be used to bolster our cognitive processes and therefore, become an extension of our cognition. It makes sense, then, that since we can save energy, increase our cognitive abilities and make use of a body we were born with, a lower level of representation and a more embodied cognitive process will allow us to make better use of the body we possess and our energy resources and therefore, is a much more efficient way than the fully representational classical models. We will see how this need for efficiency means that, in our brains, just the minimal level of representation is used to allow us to “scrape by”.

Efficiency Argument

The human brain and its mysterious inner workings that we try to replicate in artificial intelligence have been shown to use far less representation than we thought. We can see this in all the ways that the brain fails to notice what should have been significant details like in the famous Gorilla Experiment [3] where participants, upon being asked to watch a clip of a basketball game, fail to spot when a person in a gorilla suit walks through the middle of the game. If the brain used the same level of representation as our machines, then all the participants should have made a complete internal model of the basketball game and noticed the gorilla. Instead, it seems that the brain is selective in its attention and by doing so, restricts the amount of information it needs to process at one time and is, therefore, more efficient.

” Rather, it seems that, for the most part, our brains can get by with as little representation as possible to be functional and be generally correct when it comes to problem-solving.”

Besides incomplete representation, the brain is also prone to a phenomenon called gist memory[4] where our memory can be, at times, approximative and at worst, unreliable. In a task where participants are asked to remember words with similar associations like “ice”, “snow”, and “winter” from a list, participants often say they remember a word, like “cold”, that wasn’t present on the list but also shared those associations. Other shortcomings like the notoriously unreliable eyewitness testimony have further exposed just how little our brain represents from the world. Rather, it seems that, for the most part, our brains can get by with as little representation as possible to be functional and be generally correct when it comes to problem-solving.

Evolution Argument

However, we are not the only ones capable of exhibiting intelligent behavior. In fact, much of intelligent behavior doesn’t need a brain but rather an interlocking system of simple operations that, when viewed gestalt, suggest intelligence. Roboticist Rodney Brooks coined the term “subsumption architecture”[5] to describe such a structure. We need to look no further than our own bodies for an example. Our immune system possesses remarkable adaptability and complexity[6] that, if we didn’t know any better, we might have thought that the individual cells within our body possess their own intelligence. From the outside, the operations of the cells that make up our immune system seem to be a conscious, coordinated effort but each cell is actually just responding to certain stimuli released by other cells like proteins or hormones. 

This sort of organisation can be seen in organisms like ants where they use very simple chemical markers that elicit simple behaviors that, when repeated within a colony of thousands or millions of members, can forage for food, wage war, maintain fungus farms and aphid herds, and build floating rafts to survive floods. Brooks himself built creatures based on crickets that showed how intelligent behavior like finding mates which require pathfinding and spatial reasoning can actually be the result of very simple physiochemical reactions within the body[5].

Therefore, not only is representation not needed at all for certain intelligent behaviors but even for the human brain, representation is only very minimally used. However, there is more to our intelligence than merely problem-solving and so far, it doesn’t seem like they’ve been accounted for. 

Rebuttals

Imagination, for one, hasn’t been accounted for. Imagination, according to Merriam-Webster, is “the act or power of forming a mental image of something not present to the senses or never before wholly perceived in reality”. This means that, necessarily, to be able to imagine means that the things we imagine must not already exist and therefore unable to be represented. It can be argued that our imagination is actually just a composite of previous experiences (representations) put together in novel ways to make something technically “new”. However, it must be considered that just because representations of something exists doesn’t mean that it’s necessarily real. For example, illusions and hallucinations causes a person to represent something that isn’t actually there. I am of the thought that the false representation of a non-actual object cannot be truly considered to be represented the same way we represent what’s real[see 7]. In the same vein, although I cannot yet explain the presence of these false representations, I argue that imagination, due to its focus on the non-actual, does not in fact rely on representations or at least, the type of representation as we currently understand it to be.

“We don’t know why it is so, it just feels right.”

Second, our memory, it might seem, also depends on representations. After all, memory is formed through the encoding of past experiences. However, I argue for limited representation, not the total elimination of representation. It is true that a certain type of memory (fact-based, non-phenomenal memory) may be represented but it still doesn’t account for other types. Muscle memory, for example, is formed through repetition of a movement and once formed, is accessed without conscious effort. This is different from the explicit content that makes up represented memories. Similarly, ability-based memory also seems to lack representation. For example, when teaching another to drive, it is hard to articulate the feeling of the car, how to know how far to turn or how far the hood extends past the steering wheel. Perhaps a better example can be found in our use of language. We can usually go about our day and communicate with no problem but once we are forced to slow down and explain the finer details of grammar or convention, we are often stumped. We don’t know why it is so, it just feels right. Efficiency once again plays a role here. By cutting out the thinking part of certain operations like driving a car or figuring out grammar before speaking, the brain doesn’t need to waste resources “reinventing the wheel” and rather, just knows that certain stimuli should entail certain reactions. You don’t need to think, “the light is green and green means go” before pressing the gas pedal. You do it automatically. This way, the process cuts out thinking and representing entirely and can go straight from sensing to reacting.

It was given for a long time that the representational theory of mind would be the basis on which we can produce higher cognition in our own creations but now we know better than to let that be the end-all-be-all. With a better understanding of the reciprocal relationship between our bodies and our cognition, studies that reveal just how little our own brains rely on complete representation, and the fact that not everything that looks intelligent is intelligent, it seems that representation is actually a rather inefficient and insufficient explanation for all the miracles we are capable of.

References

[1] McNerney, Samuel. “A Brief Guide to Embodied Cognition: Why You Are Not Your Brain.” Scientific American Blog Network, Scientific American, 4 Nov. 2011, blogs.scientificamerican.com/guest-blog/a-brief-guide-to-embodied-cognition-why-you-are-not-your-brain/.

[2] “Chapter 9: Extended Minds?” Mindware: an Introduction to the Philosophy of Cognitive Science, by Andy Clark, Oxford University Press, 2014, pp. 192–211.

[3] Simons, Daniel, director. Selective Attention Test. YouTube, YouTube, 10 Mar. 2010, www.youtube.com/watch?v=vJG698U2Mvo.

[4 ]Makin, Simon. “What Happens in the Brain When We Misremember.” Scientific American, Scientific American, 9 Sept. 2016, www.scientificamerican.com/article/what-happens-in-the-brain-when-we-misremember/

[5] Brooks, Rodney A. “Intelligence Without Reason.” The Artificial Life Route to Artificial Intelligence, 2018, pp. 25–81., doi:10.4324/9781351001885-2. 

[6]Chaplin, David D. “Overview of the Immune Response.” The Journal of Allergy and Clinical Immunology, U.S. National Library of Medicine, Feb. 2010, www.ncbi.nlm.nih.gov/pmc/articles/PMC2923430/

[7] Loar, Brian. “Transparent Experience and the Availability of Qualia.” Consciousness and Meaning, 2017, pp. 273–290., doi:10.1093/acprof:oso/9780199673353.003.0016.

March 2019 Quote of the Month

“You asked me to teach you chess and I’ve done that. It’s a useful mental exercise. And through the years, many thinkers have been fascinated by it. But I don’t enjoy playing. Because it was a game that was born during a brutal age, when life counted for little, and everyone believed that some people were worth more than others. Kings and pawns. […]Chess is just a game. Real people aren’t pieces, and you can’t assign more value to some of them than to others. Not to me, not to anyone. People are not a thing that you can sacrifice. The lesson is that anyone who looks on the world as if it was a game of chess deserves to lose.”

– Harold Finch, Person of Interest