Daniel Dennett and His Intuition Pumps

Dennett's "Intuition Pumps" has inspired me, at last, to write a book. Maybe. I'd call it "Better Intuition Pumps" (BIP).

Dennett's book is a lovely "intuition pump" in its own right, but it cries out to be rewritten, refuted or at least challenged.

In other posts, I'll flesh out possible chapters in BIP, but I can outline some main points here.

Dennet champions as "Scientific" discipline for the study of mind, which he calls heterophenomenology "H". The idea is that anything important about the mind can be learned from the "outside", through experiments, interviews etc.  The data "H" works with is the set of observable, documented "observations" along these lines.

My problem with H is that it doesn't actually study the mind. Dennett is making the assumption that the mind is, for all intents and purposes, the brain. But one would think this is exactly what he needs to prove.

"Pumps" sets up a false alternative. Either the mind is the brain or the mind is something "inside" the brain or somehow associated with the body, like a "soul" or "homunculus". Why are those the only alternatives?

His mistake seems to be assuming his conclusion. To escape this mistake, we need to take a closer look at what a "mind" is. I happen to have one that we can examine as closely as time permits. Mine.

Here is the first "Better" Intuition pump. Consider the "mind", M, to be identical to the first-person experience of the Universe. This makes no assumption about the brain or, for that matter, a "soul". Moreover, we can use H techniques to investigate this mind in as much depth as we wish.

My discussion of M will show that it is not possible, even in principle, to determine anything of interest about M using H. In particular, H studies brains not minds.

A second BIP is the "Xombie". This is related to Dennet's "philosophical Zombie", a fascinating creature that is identical in all respects to a human being except he lacks the experience of being human. There is "nobody home". If you think that such a being is impossible, you are, according to Dennet, a victim of the "Zombie Hunch". You think that there must be something other than what shows "from the outside". Dennet challenges the reader by saying that this is "just a hunch" and needs proving. Otherwise, the default view is H, which assumes that we can never tell the difference between a Zombie and a person with "real" experience of the world.

I hold that Dennet's Zombie is interesting only in a rather trivial way, by proposing a way to "build" a Zombie in a particular way - a Xombie. The difference between a Xombie and a Zombie is that we specify in principle how the Xombie manages to "fake" consciousness. In summary, we imagine a robot that is exposed to he totality of human experience of the world from "birth". The Xombie can fake having a "mind" because we maintain that the mind itself is M, the totality of first person experience of the Universe. Thus, our Xombie is, by definition, "conscious".

Then, the question boils down to whether we could, in principle, actually build such a robot. This question veers of into what we might call "garden variety philosophy" as we look into things that make up an "M-like" mind but may be hard to implement, even in principle. Here are a few things in my mind that may be hard to implement:

• I know that fish don't wear hearing aids
• I may mistakenly believe that Napoleon was King of Spain
• I don't understand the math of General Relativity and I know I don't

My suspicion is that the "boundary" between the known Universe of M and the "known unknowns", the "unknown unknowns" and so forth may be hard to model.

IBM's "Deep Blue" tells us that such things may not be as hard as they look.

But of course, we are still left with the suspicion that our robot doesn't have a real experience of the world. He has no real "first person" - he just fakes it. But at least we have a better idea of what he's faking: a deep, intimate, long-term relationship with the Universe. In other words, the Universe itself plays a large (perhaps indispensable) role in a "mind". The brain is just a detail.

There are other things to say about H. What, exactly, are the "results" of H - the theories, generalizations and hypotheses that make H a "Science". A Science is not just a menagerie of curious phenomena. What is H about? If H leaves out what I regard as the most important aspect of M (the Universe), what is it studying? My position is that you can't make generalizations about "minds" because they are by definition unique. Any generalization you make will be about brains, "typical" human behaviour etc. Is it really meaningful to talk about what people on average claim to experience? Haven't you left the baby out with the bath water?

A simple example can suffice. Ask our subjects to describe the first time they made love. It's hard to imagine that that all the "H-data" in the world would come close to "investigating" the real, personal experience of the individual, which is, after all, what we are supposed to be investigating.

I'm sure Dennet is familiar with this objection, but I'm not sure he is aware that there are an infinite number of objections like this. In fact, the infinite "space" of M is precisely the most impressive feature of M itself. This "space" can be explored by H, but such an exploration leaves theories of the brain far behind.

Another BIP is "Object Oriented" design (OOD), which would replace Dennet's lame chapter that tries to explain how computers work.  OOD explains everything about computers from the microscopic structure of memory to the emergent structure of the group mind - the Internet - and everything in between. It is perhaps one of the most powerful set of memes ever invented. It as a subset of human language that has he advantage of being totally unambiguous and "mappable" to the "real world". With OOD in hand, one can ask interesting questions about how the Internet acquires something like an M-type mind. OOD can be used to explore complexity and emergent phenomena.

Another BIP comes from self-organizing structures, which would complement Bennett's chapter on the game of Life. Besides being built entirely out of OOD components, the Internet is also self-organizing. 

So, computers and the Internet are worth including as "imagination pumps" in their own right.

It would be worth discussing whether a living thing is "sorta" an OOD creation. I'd say it's not and the differences may turn out to be fundamental - in other words, it may be impossible to build a carbon-based living thing following only OOD principals. If true, this would be hugely interesting. There seem to be serious problems with any attempt to "model" living cells, organelles, synapses and tissues using OOD ideas. These also happen to reveal some objections to the idea that a living thing is, in some sense, a "machine". Does this mean that there is some fundamental obstacle to building a machine mind.

Finally, Denett's books tend to present a rather simplified view of the brain. His "neurons" and "genes" are almost mythical creatures, no much different from demons. They are philosophical constructions - "memes", with almost all real behavior of brain tissue left out. What do we learn when we introduce real brain structures into Dennet's discussions? Is there a simplified version of the real brain that could serve as an "imagination pump"? Would it be useful to keep in mind that Dennet's discussions that seem to talk about the brain, using words that seem to refer to real things may trick us in the same way that Astrology tricks us by talking about real stars?

There seem to be quite a few publications around that describe real brain function. Worth a look. At first glance, we see a lot of OOD-based concepts.

What, if anything, is the "self"?

What is the "self"?

Among proponents of mindfulness and Zen, it is common to say that the "self" is an illusion - that the "self" does not really exist. Such talk leads us immediately into paradox, but Zen loves paradox. Vast amounts of ink are spilled in discussions about how, if the self doesn't exist, who or what it is that knows it doesn't exist. The self seems to be a matter of daily experience. Saying it's an illusion is little more than an attention-getting opinion or a device to sell books.

To me, the problem seems to be a shortcoming of language.

I just read a treatment of this issue that purported to solve it all by claiming that the self is a verb not a noun. This poor individual was trying to tackle one of the hardest problems in the world using his tenuous grip on the grammar he learned in school. He dimly perceives that the "self" is not an ordinary "thing", so he concludes it must be the only other kind of grammatical entity he remembers: a verb. As an amateur philosopher, he imagined that the problem can be solved by re-defining a few terms and then solving the whole issue by talking about it.

If he had been awake in High School physics, he might have heard about other types of "thing" - such as force fields and processes. Since the "self" is obviously connected in some way with living human beings, and life itself is clearly a process, it makes sense that the mind (the self) must be come kind of process. The problem with that is that there are millions of types of process and probably billions more awaiting discovery. Calling the "self" a process doesn't do much for us, but it does free us from "thing think", such as the image that "we" are a "soul" that somehow drives our bodies around durning life and migrates to another plane of existence when our body dies.

My own personal experience is with the "process" described by a computer program. Since very few people are aware of the power of modern computer languages. Let me briefly digress to describe how they work:

• "things" are described as "objects" which can be totally abstract or perhaps an abstract representation of a "thing", such as a person or a bank account. To control an ICBM, the missile would be represented and controlled as a "thing".
• Objects have behaviours that can be as simple as adding numbers together or or as complex as driving down the road without hitting anything
• Objects have a "state", which can be thought of as a kind of memory that influences behaviour. For example, a bank account with zero balance will respond with an "insufficient funds" message if asked to transmit money to another account.
• Objects send messages to each other, which can be instructions, "inputs", "outputs" etc.

While these ideas are almost childishly simple, the elements of object oriented architecture can be used to describe a stunning range of "things" in the world, especially the "things" we build. There are architectural rules and "best practices" that describe how to create design objects that are fairly robust descriptions of "things" that are likely to exist in the "real world" or at least ideas that are "orthogonal" (describing the world with a minimum of non-overlapping concepts).  The idea is to extend the language in a way that the "words" have a meaning outside of the context that lead to them being invented in the first place. This creates such things as the vast zoo of objects underlying the Java language - the language of the Internet. But I digress ...

When you send a line of text (object) to Google, thousands of messages are sent between objects in Google computers, chatting back and forth, to assemble a big long list of links (another type of object) that are displayed on your computer (another object). You can click on one of them (a message), to trigger a server somewhere (another object) to send a message back to you (a web page object).

The underlying language and architecture we are dealing with is incredibly powerful - far more powerful in some ways that any human language. It can, for example, completely describe the operation of a machine that will land on Mars and proceed to explore the place.

On the other hand, we must not slip into thinking that "object oriented" computer languages can describe anything that the human mind can conceive. But we can go a long way to describe the operation of the brain in object oriented terms. For example, neurons are "stacked" in a pattern endlessly repeated in the grey matter of the brain. At least to the first approximation, these "stacks" can be simulated using object-oriented architecture. 

Think of it this way. Suppose the "self" is somehow "like" a giant conventional machine, like a jet aircraft. Now imagine that someone wants to explain this machine in a giant essay. It's easy to see that words will fall short of the task - you need blue prints, graphs, engineering diagrams, demonstrations and a lot of training just to understand any of it. Words will just not cut it. In fact, the more words you use, the less likely it is that the description will make any sense. The human mind is, by definition, exactly as complex as anything anyone can ever possibly know. Any non-trivial description of the operation of one mind, let alone the interactions between multiple minds can be expected require a few powerful descriptive conceptual tools that require specialized knowledge to use.

Chimpanzees can make tools out of twigs and have probably done so for a million years. They will continue to do so for another million years. But they will never make a telescope out of twigs. Philosophers pretend to discuss the most difficult problems of existence using ordinary words, the bluntest of descriptive tools and the oldest tools available to human beings. In fact, to me, their entire dialogue seems to be about re-defining and mis-using words. They are paid by the word and the more they add to the store of words, the less they describe anything in a useful way. But I digress ...

From the examples of twigs and jet transports, it should be noted that even sophisticated technical tools of description don't cover all we need to know or describe about our subject. For example, many of the aircraft components are made from special alloys using sophisticated manufacturing techniques. Awkwardly, we are not always sure why some alloys have the properties they do. They just do. We are quite sure that the same is true of the mind - it's not just a machine processing information. It's also a chemical reaction of nasty complexity and, of course, it's alive, meaning that its biological nature must be factored in to any useful description of its operation.

Having said that, allow me, for a moment, to imagine that the "self" is an "object" in the sense that a running computer program is an "object". I imagine the "self" as a "program" running in the brain, just like the operating system runs (most of the time) in your computer. It's a process, but a special kind of process. This is a way of thinking of the self as a kind of machine, but a very sophisticated machine. We can then ask a few things about what's going on a "self".

• It's easy to see that sensations are messages received by the senses of the body
• Memories are part of the "state" of the self
• Actions are output messages

That's all pretty obvious, but it gets interesting when you start to ask about what kinds of "things" are "ideas", "beliefs", "intentions", "regrets"  and "thoughts". Those are obviously coming into the mind and popping out of it and running around inside it without obvious external signs. There are very good reasons to suspect that "ideas" cannot be reduced to the kind of formalism that we'd like. The closest we have come so far is the meme, an interesting analogue of the gene. Since meme theory originated in 1976, progress in "memetics" has resembled philosophy more than physics.

In this little note, I don't intend to nail down exactly what it is that the mind "contains" or works on: only to suggest that this metaphor of the mind (the self) is much more productive and probably closer to reality than saying the self is an illusion or a "verb" or a "soul".

We are also free to ask about what aspects of the mind are not covered by this metaphor. For me, the most glaring omission is the fact that human minds work in huge assemblies called "societies" and all the "thoughts" running through individual human minds run through the entire society. It seems to me that interactions that add up to culture are not so easy to think of in object oriented terms. To begin with, machines don't hate, mate, have children, die and fear death.  They are not honest or dishonest, good or evil.

Come to think of it, they don't have theories about their own programming ...