The Science of Your Mind, Part 1
An Introduction to Cognitive Science
Cognitive science studies the human mind and conscious experience. If your human mind is anything like mine, consciously experiencing things is literally 100% of what you do all day. You’re a natural at it, and without even trying! Doesn’t it just grease your blood pump to know there are people who spend their entire lives working on making sense of that for you? Thanks, science!
The subject of my latest effort in cognitively scientific self-education is the book Mind, Body, World: Foundations of Cognitive Science by Michael Dawson. Dawson introduces the reader to important concepts in the field and explains the main approaches we’ve developed for the task of unraveling your twisted mind. I’m finding the book fascinating and full of insights that would be valuable even to non-practitioners, and I’ve therefore accepted a solemn duty to communicate this information to you. The hope is that these ideas will give you a new perspective on your own mind as you realize that “teetering bulb of dream and dread” is more than just a meat bag you need to lug around every time you want to go get some Chipotle. Also, it gives me a chance to write Medium articles with poetic quotes I stole from Hofstadter that convince you I’m smart and know things! Yay for both of us!
In this first of what will (attention span willing) be a series of articles, we’re going to get introduced to the three main approaches in the field: classical, connectionist, and embodied cognition. Fear not! We’re going to keep things pretty non-technical today. If you’re interested in the nitty-gritty, then stay tuned for the following articles. When links are available, I’ll put them right here.
Links
- Part 2: The Ogre Theory of Mind
- Part 3: Let’s Get Physical (Symbol System)
- Part 4: Neural Networks and Hostage Cats
- Part 5: Bodies Everywhere!
- [MORE TO FOLLOW AT SOME FUTURE TIME IN THE FUTURE]
What is cognitive science?
According to the Cognitive Science Society, the goal of cognitive science is “understanding the nature of the human mind.” As you might imagine, that’s a pretty tough nut to crack, and it’s a problem that lends itself to analysis from many different disciplines.
Research in cognitive science includes professionals from neuroscience, psychology, computer science, philosophy, artificial intelligence, and more.
As it turns out, having a mind is pretty important to many different types of science. Go figure! These are the types of questions that cognitive science is trying to answer:
- How is it that you can have a conversation with your friend in a noisy bar, picking out her words from the ocean of noise?
- How is it that you can undesrtand this sentence, even though I just misspelled the word understand?
- What’s going on in your brain when you take drugs? Why do you experience it the way you do? Why do you experience it at all? Why is it that you can smoke weed only occasionally but Jeff seems determined to convert the world supply of marijuana into smoke in his basement?
- What is an emotion? Can you have an emotional reaction without being aware of the emotion? How much of what you’re feeling on a daily basis are you actually aware of, and how much is just going on behind the scenes?
- How do we turn photons into vision? Do all animals have the same vision system? Why don’t we notice saccades or the blind spot in our retina?
- What kinds of conscious experiences are possible, why do you prefer some and not others, and how can you move toward having more of those you enjoy?
- What is addiction, and what are the biological, social, and environmental conditions that lead to it? What does it look like in the brain? When you’re quitting, are you “forgetting” the addiction or learning a new behavior?
- Does a computer think? Is artificial consciousness possible? Under what constraints?
- In the words of the Dalai Lama, “where is mind?”
For each of the above questions, you might feel that some particular discipline is best suited to its study. For example, you might give vision to the neuroanatomists, let the computer nerds tackle AI, and psychologists get the emotional stuff. How does cognitive science bring all these ideas together? Well, the first word in “cognitive science” is cognitive; let’s start there.
A founding principal of cognitive science is that cognition is information processing.
“Ah! Clear as mud,” I can hear you thinking. And if you were hoping for a clear-cut answer, I have some bad news for you:
“One accomplishment that has eluded cognitive science is a unified theory that explains the full range of psychological phenomena, in the way that evolutionary and genetic theory unify biological phenomena, and relativity and quantum theory unify physical phenomena.”
Thagard, 2005, p. 133
All hope is not lost, though! Granted, the exact definition of information processing differs between the classical, connectionist, and embodied camps. In trying to understand the position of each approach, though, we learn more about its history, the types of theories it has developed, and the methods by which those theories are tested. So let’s dive in and see if we can’t tease apart this problem.
The big three
The classical approach
Way back in the dark ages of the 1950s, a few really important things happened. Foremost among them is, of course, the birth of Rush drummer and lyricist Neil Pert. Perhaps slightly more relevant to this article was the world-shaking conversation painstakingly transcribed below:
Bro, these new digital computers are sweet! They can solve all these problems that only humans could do before. It’s almost like they can think or something. I wonder if the human brain works in a similar way…
(Ruminating in silence)
Hey! What if we used a computer to model, like, the mind?! I mean, information goes in, decision comes out, same as a computer. It’s just information processing, right?! And then — oh my God, Jeff, are you listening? I’m having the best idea right now — THEN we could use those cybernetics math proof things and apply them to OURSELVES!
- 100% authentic quote from 1950s historical science man to his roommate, Jeff
Luckily for us, Jeff was listening, and cognitive science was born.
OK, some of my history might be of the alternative type, but you can read the book if you want journalistic accuracy.
In a time of rising enthusiasm about the digital computer and its widespread application, cognitive science drafted its first interpretation of the phrase information processing: symbol manipulation through well-defined rules.
Let’s unpack that idea a bit. Early cognitive scientists imagined the mind as interacting with the world by first translating input from the senses (sight, sound, etc.) into “symbols” or mental models. These symbols are mental placeholders for the state of the world, including oneself. Given symbols in some initial state and a desired goal state, we would essentially go through a checklist of rules and apply those that meet the current conditions. Applying a rule can change the symbols to a new state, and this process repeats sequentially until the model is in the goal state. At that point, our master plan is ready, and we pass it on to the body to carry out post-haste (we need that Chipotle, body!).
The algorithmic process described above is what’s known as the sense-think-act cycle or the “classical sandwich.” According to this paradigm, the mind does not interact directly with the world. There is a thinking step that separates sensing and acting, and the role of the brain is to plan the action.
Game-playing and problem-solving are examples of where this view of information processing works well. Imaging yourself playing a game of chess. You might indeed have a checklist of rules that you run through before you make a move. Which of my opponent’s pieces poses the most risk to me? What moves can I make to defuse that risk? Of those moves, which costs me the least? Should I just flip over the board and go home because Jeff is DEFINITELY cheating? When a move occurs, it changes the state of your mental model of the game. And what you’re interacting with is really the mental model, not the board in front of you. Your mental model may not include all of the pieces. You might simplify by grouping together low-risk pieces and only individually track each of the pieces that could mess you up. Your model might have layered onto it moves that haven’t happened yet. You look at the board, you break it down into useful symbols, you manipulate those symbols to make a plan, you tell your hand to move a piece, repeat.
The connectionist approach
Some cognitive scientists saw a major flaw in the classical approach in its reliance on what are called well-posed problems. A well-posed problem is characterized by states of knowledge, goal states, and operations for converting states of knowledge that are all clearly defined. The benefit of this approach is that problems can often be solved by simply searching the space of possible solutions. There were many early successes in using this approach to build systems to understand natural language and help with expert decision-making, and these early wins persuaded us that we might be on to something. The thing is, a lot of the things that we meat-bag-luggers are good at (speaking, seeing, walking, etc.) are just a bit too squidgy to fit into such neat boxes. Squidgy is definitely a technical term from the book, you don’t need to check.
There was this group of cognitive scientists that just wouldn’t shut up about the squidginess. Not content to just throw shade at the state of the art, though, these overachievers actually went out and developed a whole new theory of cognition! Taking inspiration from the biological neuron instead of the digital computer, they created the architecture that gave them the name connectionist: the neural network.
“Wait a second,” I hear you ask, “neural networks? My friend Ruth told me about those! Doesn’t Facebook use neural networks to sell Coca Cola election ads to the Russians and try to clone my dog from our Hawaii vacation photos?” Well, no, because that’s… that’s just not a thing and Ruth definitely needs help. But also yes, those are the same neural networks; they’re just a lot bigger and shinier now. So, partial credit to Ruth for making the connection.
Also, screw Facebook.
The neural network is a parallel, decentralized system that learns from example. Where the classical approach relied on hand-crafted rules that are logically and physically separate from memory, the neural network both stores and manipulates information in the same structure. Instead of performing logical operations on symbols, the neural network finds patterns in data without necessarily developing any symbolic representation of that data.
To understand this view of information processing, let’s return to our chess example. It’s worth noting that the best chess-playing computer programs in the world right now use neural networks, not a checklist of rules. Like us, these artificial neural networks learn how to play chess by playing it (just not against Jeff, that damn cheater!). The insights they develop might not deal with specific pieces, but rather with higher-level dynamics involving configurations of multiple pieces (e.g. two-pronged attacks) or even sub-piece characteristics (e.g. pieces that can move diagonally).
Maybe this is something like the intuition that a master chess player develops over a career, which would be hard to put into words and would therefore not be cleanly expressed by if-then rules of the type that classical cognitive science deals in. Such a player might look at the board and — information working its way through untold matrices of historical experience — have some intuitive feeling of the possible moves and evolution of the game without any conscious problem-solving.
However, it’s worth noting that these computer chess players are trained on millions of games. They have practically infinite memory, perfect recall, and can check thousands of possible moves per second. So even though computers can kick our collective asses at chess, it’s not necessarily clear that they play the game the same way we do.
The embodied approach
Connectionist thinking did away with a good deal of what was core to classical cognitive science. The embodied folks did away with the rest.
“In particular I have advocated situatedness, embodiment, and highly reactive architectures with no reasoning systems, no manipulable representations, no symbols, and totally decentralized computation.”
Brooks (1999, p. 170)
Well, bully for you, Brooks!
The embodied approach rejects the idea that the mind can be understood separate from its environment. The two are inextricably linked, and both contribute to cognition. This is what’s knows as the extended mind hypothesis. You might have also heard this called distributed cognition. From this perspective, your notepad isn’t just a tool, it’s external memory. The pilot doesn’t fly the plane; rather, the pilot, the checklist she runs through before takeoff, and all the other tools in the cockpit are what fly the plane.
Oh, and about that “classical sandwich,” sense-think-act and what have you. Why the think? The embodied camp goes straight from sense to act, serving the kind of boring-ass sandwich that only the Dutch could love. You don’t need representations or planning, you’re just acting directly on your perceptions, and the brain is just controlling that flow. That’s how an embodied cognitive scientist would describe information processing.
Let’s return to our game of chess one more time. From the perspective of embodied cognition, the who that is playing chess expands. The chess board is your working memory. Captured pieces lay beside the board, reminders of past victories and mistakes, of the strategy you’re employing. The chess clock is… well, it’s a clock. But the stationary hands on your side of the clock indicate that its not your turn so you can have some Doritos. The ritual of pushing the button on the clock might be an important step in organizing your thoughts. The book on chess you were reading yesterday contains insights from thousands of games you never played, yet you can now bring those insights to bear in in finally beating Jeff at chess NO WAIT GOD DAMN IT JEFF YOU CHEATER!
Recap
To summarize, here’s a table that breaks down what we discussed above.
Flame wars
Perhaps the best source of information about the main approaches are the researchers who developed them. Because we all seem to love a good trash-talking these days, here are some quotes from cognitive scientists of each camp destroying the views proposed by their intellectual opponents.
Well, more like politely explaining their difference of opinion. What? They’re scientists, not news pundits.
Classical
“The problem with connectionist models is that all the reasons for thinking that they might be true are reasons for thinking that they couldn’t be psychology.” (Fodor & Pylyshyn, 1988, p. 66)
“People are fascinated by the prospect of getting intelligence by mysterious Frankenstein-like means — by voodoo! And there have been few attempts to do this as successful as neural nets.” (Stix, 1994, p. 44).
Connectionist
“No serious study of mind (including philosophical ones) can, I believe, be conducted in the kind of biological vacuum to which [classical] cognitive scientists have become accustomed.” (Clark, 1989, p. 61).
“The idea that human activity is determined by rules is not very plausible when one considers that most of what we do is not naturally thought of as problem solving.” (Horgan & Tienson, 1996, p. 31).
“Good old-fashioned artificial intelligence was a failure. The contribution of standard architectures and standard programming artificial intelligence was a disappointment.” (Baumgartner & Payr, 1995, p. 36).
Embodied
“Models of the world simply get in the way. It turns out to be better to use the world as its own model.” (Brooks, 1991, p. 139).
“The realization was that the so-called central systems of intelligence — or core AI as it has been referred to more recently — was perhaps an unnecessary illusion, and that all the power of intelligence arose from the coupling of perception and actuation systems.” (Brooks, 1999, p. viii).
I conclude this article
And there you have it! A brief review of some important concepts in cognitive science. Join me next week for another dose of science and comedy. As a teaser for next week’s topic, let me leave you with a story that also serves as an introduction to one last key concept: the thought experiment.
Thought experiment: the Chinese room
Imagine an empty room. Into that room walks an elderly woman. She crosses the room to sit in front of the opposite wall. Beside her are a pen and a stack of blank sheets of paper. She takes a page and writes down a question. She writes in Chinese. She folds the page in half and slides it through a small slot in the wall she’s facing. She waits a short while, and a page is returned through the slot. She unfolds it and reads an answer to the question she asked. She smiles, stands, and leaves the room. After a moment, a young man enters. He sits down in front of the slot and reaches for a blank page.
Imagine a young man sitting at a desk against a wall. On his right sits a tome, hardbound in leather. Blank sheets of papers are stacked in front of him. A note, folded in half, slides out of a slot in the wall to his left and falls to the desk. He opens it, scans it, then starts flipping through the pages of the tome. After some searching, he stops. On the left page of the tome are the same symbols from the folded note, on the right are a different set. The man slowly copies the symbols from the right page to one of his blank sheets, folds it, and passes it back through the slot.
The man does not know how to read Chinese.
Takeaway: Once a problem is decomposed into primitives, the intelligence disappears! Intelligence can emerge through the complex interaction of simple components.
