Friday, November 24, 2017

Materialism Sucks or Why Jeff Hawkins and Geoffrey Hinton Are Out to Lunch

Cortical Column (credit: Cajal Blue Brain Project)

Two Materialist Peas in a Pod

It is not surprising that both Jeff Hawkins and Geoffrey Hinton would hit on pretty much the same solution to the problem of invariant object recognition. After all, they are both materialists. They reasoned that, since we see a 3-dimensional world around us, the brain must have some internal mechanism to represent this 3D geometry. They believe that the brain has somehow evolved (all materialists believe in Darwinist pseudoscience by default) special neural mechanisms to emulate the geometry of the world. Hawkins proposes the existence of special neural circuitry that generates a "location signal" and Hinton believes that a "pose matrix" is required.

Materialist Representationalism Is Just GOFAI Redux

We can go right ahead and dismiss both approaches as representationalist nonsense since the brain's neurons are way too slow to model the 3D geometry of the world. Fast computation requires a lot of energy which the brain does not have. Besides, why go through the considerable trouble of computing a model of the world when the world is its own model and already performs all of its own computations? Isn't it much more plausible that the brain is designed to simply learn to sense the world? (For more on this topic, please read this article: The World Is Its Own Model or Why Hubert Dreyfus Is Still Right About AI.)

What I find even more interesting is that, without realizing it, both Hawkins and Hinton are practicing GOFAI, aka symbolic AI, aka representationalism. That is to say, they want to construct special internal mechanisms to represent some aspect in the world, in this case, its 3D geometry. This is no different than using a special internal symbol to represent a cat or a tree. I must say that Hinton is the bigger GOFAI artist here because, loud denials to the contrary notwithstanding, a deep neural net (which he helped pioneer) is just an old-fashioned rule-based expert system.

The Universal Principle of Perception

The beauty and power of human intelligence is its universality or generality. The moment one chooses to use any kind of special mechanism to process or represent a certain type of knowledge, universality goes out the window and with it the ability to generalize and make analogies. It would be a nightmare of integration if the brain used one way to process visual signals and another to process gustatory, auditory or olfactory signals. The only way to avoid this nightmare is to have a single perceptual mechanism based on a single principle to process every type of sensory input.

How does the brain do it? It is simple, really. The brain converts all sensory inputs into spikes. A spike is a precisely timed discrete signal that indicates that a minute change or event just occurred. Millions of sensors generate millions of spike streams. To the brain, every spike looks the same as any other. The brain must somehow find order in the spikes. Here is the clincher: the only order that can be found in multiple streams of discrete signals is temporal order: signals can be either concurrent or sequential.

It does not matter whether or not a spike is generated by a visual, auditory or tactile sensor. The brain processes them the same way. It discovers their temporal correlations and create millions of internal sensors to detect these correlations as they happen. Distances are converted into temporal intervals. In other words, the interval between two notes in a song and the distance between point A and B on the retina are not processed differently by the brain. They are both temporal intervals. At its core, the brain is a massive timing mechanism. The brain generates oscillatory pulses to accomplish its task. This is where brain waves originate.

The Curious Case of Jeff Hawkins

While I am not surprised by Hinton's promotion of his representationalist capsule theory, I am somewhat taken aback by Hawkins' location hypothesis. He used to know better. At least, I thought he did. In his 2004 book On Intelligence (pdf), Hawkins wrote the following regarding visual processing in the brain (emphasis added):
People tend to think that there's a little upside-down picture of the world going into your visual areas, but that's not how it works. There is no picture. It's not an image anymore. Fundamentally, it is just electrical activity firing in patterns. Its imagelike qualities get lost very rapidly as your cortex handles the information, passing components of the pattern up and down between different areas, sifting them, filtering them.
Natural vision, experienced as patterns entering the brain, flows like a river. Vision is more like a song than a painting.

Many vision researchers ignore saccades and the rapidly changing patterns of vision. Working with anesthetized animals, they study how vision occurs when an unconscious animal fixates on a point. In doing so, they're taking away the time dimension. There's nothing wrong with that in principle; eliminating variables is a core element of the scientific method. But they're throwing away a central component of vision, what it actually consists of. Time needs a central place in a neuroscientific account of vision.
I remember being blown away when I first read the above excerpt. I was amazed for two reasons. First, these were ideas that I had understood years before I read Hawkins' book. Like Hawkins, I got my first understanding of intelligence and the brain by reading articles and papers on neurobiology, especially on the organization and operation of the retina. At the time, nobody in artificial intelligence gave a second thought to spiking neural networks and the importance of precise timing to intelligence and the brain. I thought I was alone.

Second, Hawkins is an avowed materialist and atheist. As an unabashed, card-carrying, Yin-Yang dualist, I knew that Hawkins' explanation of vision was correct but that it made no sense from a materialist point of view. Hawkins intuitively understood that there was no picture in the brain even though we clearly see a picture. How can that be possible?

I was astounded that Hawkins could still call himself a materialist while holding the refutation of materialism in his hand. Inexplicably, he never reached the logical conclusion that something other than the brain converted those neuronal spikes in the visual cortex into a picture, the fabulous 3D vista that we swear we see in front of our eyes but that exists nowhere. Something non-physical, something supernatural is obviously responsible. Had Hawkins made the leap to dualism, he would not have come out with his pathetic (I can't think of no other word to describe it) location hypothesis.

Materialism is a powerful mind blocker, a set of blinders that mainstream AGI researchers are more than willing to wear. It blinds them to the elephants that are standing right in front of them. Hawkins' steadfast refusal to give up his religious belief in materialism is the main reason that he has made no real progress since 2004. In spite of his amazing early insights and all of his millions, he really has had nothing to show. His HTM program is essentially another me-too program, nothing to get excited about. Even the clueless deep learning crowd have no respect for him. A shame, really. My advice to Hawkins is simple: throw away your blinders and return to your first love.

Coming Soon

In my next article, I will explain in more details how the brain solves the invariant object recognition problem without representation, using just timing and lots of sensors. Hang in there.

See Also:

The Curse of Materialism or Why People Like Jeff Hawkins and Geoffrey Hinton Will Never Figure Out AGI
A Critique of Numenta's Location Hypothesis
Why We Have a Supernatural Soul
The World Is Its Own Model or Why Hubert Dreyfus Is Still Right About AI
Dynamic Routing Between Capsules (Hinton et al)
A Theory of How Columns in the Neocortex Enables Learning the Structure of the World (Hawkins et al)
Does the Brain do Inverse Graphics? (Hinton et al)


Robert said...

... a single perceptual mechanism based on a single principle to process every type of sensory input… At its core, the brain is a massive timing mechanism… return to your first love.

Genesis 1:14; Exodus 18:22; Leviticus 23:4; Ecclesiastes 3:1-11; Daniel 4:17; Hosea 1:11; Revelation 1:8; 1:11; 22:13

Brian Bovaird said...

Hi Louis,
I greatly appreciate your work and look forward to all your articles. One question I have is if the world is the representation, how does your system deal with the hidden object and how we still know there is an object there even though all of our senses can't sense it is there? How is that object represented without the physical world to show us that it is there?

Louis Savain said...

Hi Brian,

Thanks for the comment. First, let me say that the brain cannot represent objects in the world because it cannot see or sense them until it builds complex internal sensors to detect them. The only way it can represent an object is by using another object in the world to serve as the representation. For instance, the symbols on the screen can represent sounds, words or other objects.

The brain learns to see the world by building complex, hierarchically organized sensors to directly sense objects. The entire process of building these internal sensors is based on the timing of sensory spikes and their temporal relationships. The sensors have no idea whether the object they are sensing is a visual, auditory or some other object. The brain only works with neuronal spikes. The colorful 3D vista that you see in front of you is supernatural. The flavors that you taste, the odors that you smell, the sounds that you hear are also supernatural. Almost everything that you experience is a manifestation of yourself, your own spirit. Some call these manifestations "qualia". They are not part of the physical world. The neuronal activity in your brain just awakens them.

An internal sensor is a sequence. It can be low level or high level. When it is activated, it retains a memory trace of that activation. The trace is a recording of both the time of activation and the manner (speed) of activation. From this trace, the brain can extrapolate the next trace. This is how a hunting dog, after momentarily losing sight of a running prey who suddenly disappears behind a hedge, can predict that the prey will reappear on the other side and act accordingly. Jeff Hawkins' original brain/intelligence hypothesis rests on this ability to predict and he is right about that.

As you can see, I view the world from the perspective of a true dualist.