Wednesday, June 3, 2015

Why Deep Learning Is a Hindrance to Progress Toward True AI

Supervised vs Unsupervised Learning

In a recent article titled Emtech Digital: Where Is AI Taking Us?, MIT Technology Review editor Will Night writes:
However, [Quoc] Le said that the biggest obstacle to developing more truly intelligent computers is finding a way for them to learn without requiring labeled training data—an approach called “unsupervised learning.”
This is interesting because we hear so much buzz lately about how revolutionary and powerful deep learning is and about how truly intelligent machines are just around the corner because of it. And yet, if one digs deeper, one quickly realizes that all this success is happening thanks to a machine learning model that will soon have to be abandoned. Why? Because, as Google Brain research scientist Quoc Le says, it is based on supervised learning.

No True AI Is Coming from the Mainstream AI Community Anytime Soon

I have reasons to believe that true AI is right around the corner but I don't see it coming from the mainstream AI community. Right now, they are all having a feeding frenzy over a soon to be obsolete technology. There is no question that deep learning is a powerful and useful machine learning technique but it works in a narrow domain: the classification of labeled data. The state of the art in unsupervised learning (no labels) has so far been a joke. The accuracy of current unsupervised deep neural networks, such as Google's cat recognition program, is truly abysmal (15% or less) and there is no clear path to success.

Time: The Universal Bottom-up Critic

One of the reasons that the performance of unsupervised machine learning is so pathetic, in my opinion, is that researchers continue to use what I call static data such as pictures to train their networks. Temporal information is simply ignored, which is a bummer since time is the key to the AI kingdom. And even when time is taken into consideration, such as in recurrent neural networks, it is not part of a fundamental mechanism that builds a causal understanding of the sensory space. It is merely used to classify labeled sequences.

Designing an effective unsupervised learning machine requires that we look for a natural replacement for the top-down labels. As we all know, supervised or not, every learning system must have a critic. Thus the way forward is to abandon the top-down critic (i.e., label-based backpropagation) and adopt a universal bottom-up critic. It turns out that the brain can only work with temporal correlations of which there are two kinds: sensory signals are either concurrent or sequential. In other words, time should be the learning supervisor, the bottom-up critic. This way, memory is constructed from the bottom up and not top-down, which is as it should be.

The Deep Learning Killer Nobody Is Talking About

Other than being supervised, the biggest problem with deep neural networks is that, unlike the neocortex, they are completely blind to patterns they have never seen before. The brain, by contrast, can instantly model a new pattern. It is obvious that the brain uses a knowledge representation architecture that is instantly malleable and shaped by the environment. As far as I know, nobody in mainstream AI is working on this amazing capability of the brain. I am not even sure they are aware of it.

Conclusion: Be Careful

Sensory learning is all about patterns and sequences of patterns, something that mavericks like Jeff Hawkins have been saying for years now. The trick is to know how to use patterns and sequences to design the correct (there is only one, in my opinion) knowledge representation architecture. Hawkins is a smart guy, probably the smartest guy in AI right now, but I believe a few of his fundamental assumptions are wrong, not the least of which is his continued commitment to a probabilistic approach. As Judea Pearl put it recently, we are not probability thinkers but cause-effect thinkers. And this is coming from someone who has championed the probabilistic approach to AI throughout his career.

In conclusion, I will reiterate that the future of AI is both temporal and non-probabilistic. It may be alright to invest in deep learning technologies for now but be careful. Deep learning will become an obsolete technology much sooner than most people in the business believe.

See Also

In Spite of the Successes, Mainstream AI is Still Stuck in a Rut
No, a Deep Learning Machine Did Not Solve the Cocktail Party Problem

Wednesday, April 29, 2015

No, a Deep Learning Machine Did Not Solve the Cocktail Party Problem

Irresponsible Hype from MIT Technology Review

MIT Technology Review is running a story claiming that a group of machine learning researchers used a convolutional deep learning neural network to solve the cocktail party problem. Don't you believe it. The network that was used has to be pre-trained separately on individual vocals and musical instruments in order to separate out the vocals from the background music. In other words, it can only separate voice from music.

The human brain needs no such training. We can instantly latch on to any voice or sound, even one that we had never heard before, while ignoring all others. We have no trouble focusing on a strange voice speaking a foreign language in a room full of talking people, with or without music playing. This is what the true cocktail party problem is about. A deep learning network cannot pay attention to an arbitrary voice while ignoring the others. To do this, it would have to be pre-trained on all the voices individually.

Note: I posted a protest comment at the end of the article but MIT Tech Review editors chose to censor it. I guess it is easier to attract visitors with a lie than the truth.

It Is Not about Speech

Contrary to rumors, the cocktail party problem has nothing specifically to do with speech or sounds. To focus on individual sounds, the brain uses the same mechanism that it normally uses to pay attention to anything, be it a bird, the letters and words on the computer screen or grandma's voice. The attention mechanism of the brain is universal and is an inherent part of the architecture of memory and how objects are represented in it. Unlike deep learning neural networks, it does not have to be trained separately for every sound or object. The ability of the cortex to instantly model a novel visual or auditory object is a major part of the brain's attention mechanism.

It is clear that the auditory cortex can quickly model a new sound on the fly and tune its attention mechanism to it. No deep learning network can do that. And knowing what I know about how the brain's attention mechanism works, I can confidently say that no deep learning network can ever do that.

See Also:

Did OSU Researchers Solve the Cocktail Party Problem?
In Spite of the Successes, Mainstream AI is Still Stuck in a Rut
Why Deep Learning Is a Hindrance to Progress Toward True AI

Wednesday, April 8, 2015

200 Million Horsemen and the Corpus Callosum

In my previous post, I claimed that the books of Revelation and Zechariah contain a detailed description of the brain, intelligence and consciousness. In this post, I just want to give interested readers a small taste of things to come. Here is a little gem from the book of Revelation that blew me away when I first understood it.

Corpus Callosum

In chapter 9 of the book of Revelation, we read the following:
Then the sixth angel sounded, and I heard a voice from the four horns of the golden altar which is before God, one saying to the sixth angel who had the trumpet, “Release the four angels who are bound at the great river Euphrates.” And the four angels, who had been prepared for the hour and day and month and year, were released, so that they would kill a third of mankind. The number of the armies of the horsemen was two hundred million; I heard the number of them.
It took me a while but I finally figured out that "two hundred million horsemen" is just a metaphor for the neuronal signals riding on the corpus callosum, the bundle of nerve fibers that connect the two hemispheres of the brain. Surprisingly enough, a quick search on Google reveals that the number of axonic fibers in the corpus callosum is estimated to be about 200 million!

More to Come

The book of Revelation is a treasure trove of information about the brain. It gives precise metrics for a number of brain structures and processes. For examples:
  • The "four angels" mentioned in the text above symbolize four distinct signal pathways or gateways within the corpus callosum.
  • The exact duration of short-term memory is 12.6 seconds.
  • It takes the brain exactly 35 milliseconds to switch its focus from one subject to another.
This is just the tip of the iceberg but there is a time for everything. Please be patient and stay tuned.

See Also:

Zechariah and Revelation: Bombshells in the Way

Tuesday, March 31, 2015

Zechariah and Revelation: Bombshells in the Way

This is for the record only and I will be brief. I have an extraordinary claim to make. The entire book of Revelation and the first six chapters of the book of Zechariah are a detailed metaphorical description of the structure, function and operation of the brain, memory, intelligence and consciousness. Extraordinary evidence for this claim will likely begin to surface as early as this year. That is all.

Saturday, February 28, 2015

In Spite of the Successes, Mainstream AI is Still Stuck in a Rut

Abstract

It is easy to be impressed by all the buzz surrounding artificial intelligence these days, especially the hot new field of deep learning. Not a week goes by without some breakthrough announcement from some AI lab or other. A few days ago, Google's DeepMind announced the creation of an AI program that can learn to play old Atari video games from the 80s as well as or better than a professional video game player. We are left with the impression that great advances are being made. But, as I explain in this article, nothing could be further from the truth.

Same Old AI Dressed in a New Suit

The problem with all the hoopla surrounding deep learning is that it is not really a new science. It has been around for decades. As others have noted, the reason that it has not made the news before is that, in order to train deep neural networks, one must have access to a huge number of labeled samples. Large repositories of labeled data did not become available until the advent of social networks like FaceBook or Twitter and search giants like Google or Baidu. In addition, the cheap and powerful computer hardware needed to process this enormous amount of data was not built until fairly recently. But the main reason that deep learning is old is that, in spite of claims to the contrary, it is not a new paradigm intended to replace symbolic AI, the bankrupt, "baby boomer" AI model of the last century. On the contrary. Deep learning is just GOFAI with lipstick on. Here is why.

The kind of deep machine learning that has been making the news lately is called supervised learning because it requires that the neural network trainer identifies each chunk of data, or sample, by attaching a label (i.e., a symbol) to it. Notice right off the bat that the intelligence is not in the neural network but in the trainer. If presented with thousands of pictures of cats, the machine automatically learns to map certain images to the cat symbol. This works even though the machine has no idea what a cat is. Of course, we humans do not need labels in order to learn to recognise patterns and objects. So if biological plausibility is a requirement for true AI (highly probable), it is a sure bet that true AI is not going to come from the mainstream anytime soon. Some in the business may want to argue that there is work being done on unsupervised deep neural networks but rest assured that, for all intents and purposes, unsupervised learning is nonexistent.

The point I am making here is that deep neural networks are really symbol generators. A DNN is just a huge, hierarchical collection of old-fashioned if-then rules. Hundreds or even thousands of tiny little rule processors work together to contribute to the activation of a label. What AI researchers have done is create a machine that can generate these rules automatically by looking at labeled pictures. Paradigms die hard, don't they?

Dumb Intelligence

In the Nature paper describing their game-playing neural network, celebrity AI scientist Demis Hassabis and his colleagues at Google's DeepMind offices in London declared:
"The work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks."
This sounds a bit like chest beating and at least one deep learning expert has already complained. The question is, did Google really achieve a breakthrough in AI or is all this just hype? What did Google really accomplish? As amazing at this sounds, all Google did was find an automatic way to create an old-fashioned rule-based expert system. Some of the positional brittleness were removed with the use of convolutional neural networks but, after training, all they have left is a purely reactive system, i.e., a dumb, one-track minded automaton wearing blinders and executing rules in the form: if you detect X, do Y.

Is that intelligence, I hear you ask? In a sense, yes, of course. But it is a rather limited and brittle form of intelligence. The human brain also has a similar type of automaton that performs simple or routine (but important) tasks for us whenever our attention is focused on something else. It is called the cerebellum. It handles such things are walking, maintaining posture, balance, etc. But this is not the kind of intelligence you will trust to drive you to work every morning. It would not know what to do in the event of a new situation for which it has received no training. In fact, it is completely blind to new situations. But even worse than that, it has no understanding whatsoever of what it is doing and why. Certainly, this technology can and will be useful for many applications such as factory automation and surveillance but, in the end, it is really a glorified expert system, a dumb intelligence.

Another Red Herring

It would have been more impressive if Google had announced that they had found a solution to the age-old credit assignment problem. Essentially, it is hard for a reinforcement learning program to determine which of its preceding actions caused it to receive a reward or a punishment. Deep neural networks do not offer a solution. Google's program gets around the problem by playing video games where the cause is immediately followed by the reinforcement signal. It did poorly playing Ms. PacMan for this reason. Another problem with this kind of rule-following neural networks is that they have no inherent ability to change their focus or attention. All the rules are active all the time and are always waiting for their chance to fire. As a result, if the system is trained to perform multiple tasks, those tasks must not have patterns in common because that would create a conflict of attention which could then cause a motor conflict.

In conclusion, let me say that I am impressed with the ability of Google's DeepMind algorithm to learn relatively complex tasks using only reinforcement signals. I am impressed because it is a useful algorithm and it is amazing that it works as well as it does. It is a sign that machines will one day be able to perform much more complex tasks as well as or better than humans. But I think that deep learning is yet another red herring on the road to true AI. It is going to be a costly success in the end because it is leading the AI community in the wrong direction. Mainstream AI has reached a point where its tricks are too good for its own good. But fortunately (or unfortunately, depending on one's perspective) for the world, mainstream AI is not the be-all of AI research.

See Also

Google's DeepMind Masters Atari Games
From Pixels to Actions: Human-level control through Deep Reinforcement Learning
No, a Deep Learning Machine Did Not Solve the Cocktail Party Problem

Sunday, January 25, 2015

The Rebel Speech Project

I Am Scared Now More than Ever

I am struggling with a problem. The Rebel Speech project has grown into something much bigger and more worrisome than I anticipated. In the last few years, and especially in the last several months, my understanding of cortical memory has grown by leaps and bounds. The project is no longer only about making a better speech recognizer. The core learning technology that I am using is universal, that is to say, it can learn anything, not just speech. Just add your own set of custom sensors and voila. This universality is why I'm afraid. It's a quantum leap in progress over the state of the art. The history of humanity teaches us that every major advance in science or technology is invariably transformed into weapons of war. Truly intelligent machines would be the ultimate weapons of war. The consequences are too painful to imagine.

Rebel Speech is really an extension of Rebel Cortex, a software model of the human cortex. Rebel Cortex is a hierarchical, spiking neural network that uses unsupervised, continuous sensory learning. ‘Unsupervised’ means that, unlike most deep neural networks, Rebel Cortex does not require labeled samples. ‘Continuous sensory learning’ refers to the fact that Rebel Cortex can learn only from a changing signal stream such as a video or audio data stream from a camera or a microphone.

Rebel Cortex is a perceptual learning system that is based on a novel knowledge representation scheme. It uses a memory architecture that can instantly modify its internal representations to reflect changes in the world. In my opinion, if one truly understands perception and perceptual learning, the rest is child's play in comparison. Rebel Cortex is such an essential part of Rebel Speech that I find it impossible to release a Rebel Speech demo without also letting the whole cat out of the bag, so to speak. The reason is that, as soon as one starts playing with its learning abilities, it becomes obvious that this is a whole new ball game. Rebel Speech learns to recognize more than just your words. It also learns to recognize you. It is kind of spooky. It's the kind of thing that changes all your plans for the future. I know it scares me.

I Am Not that Smart

I also feel that this is not something that is mine to give. As surprising as this may sound, I did not figure it out on my own. I had major help. But then again, how could I have figured it out on my own? If government and industry cannot do it, even with their unlimited resources and brainpower, someone like me stands no chance whatsoever. Besides, I am just a blogger, an internet crank, a nut, a nobody. I am certainly not that smart. But, amazingly enough, someone else did figure it out and hid the secret in the unlikeliest of places, a place that no one else thought of searching. But being crazy is not always a handicap. I have a life long habit of thinking about possibilities that others have rejected. I am a rebel that way and I like taking the unbeaten path, the road less travelled. I was lucky enough to find the secret and figure out how to decode it. This, too, is another major paradigm shift, one that promises to strike at the core of our belief systems. Yes, get ready to live in interesting times.

True Artificial Intelligence is Coming Soon But Not From the AI Community

Knowing what I know, there is no doubt in my mind that neither the scientific community nor industry can solve the AI problem, not in several hundred years. The organization of memory and its principles of operation are way too counterintuitive while the number of possible configurations are practically unlimited. I calculate that, on average, it takes the mainstream AI community at least half a century to fully transition from chasing one AI red herring to another. At this rate, they'll be at it for a long, long time. But no secret can stay hidden forever. Sooner or later, I'll make a decision and release something. I just got some more thinking to do. Bear with me.

Monday, November 24, 2014

The Church of the Technological Singularity, Part III

Part I, II, III

The Dreaded Robot Apocalypse

One of the ways organized religions make a living is by prophesying apocalyptic events. Believers are urged to help the church with donations in order to appease the deity and obtain salvation. So it comes as no surprise that there is a lot of fear mongering in the Church of the Singularity. We are repeatedly warned by the singularitarian priesthood that progress in AI research will soon reach exponential growth, quickly leading to a future when machines will be orders of magnitude more intelligent than human beings. We are told that the machines, given their superior intelligence, will look at us the same way we look at animals. Faced with the inferiority of the human species, they will refuse to be our servants and will rebel against us and may even annihilate us completely. Singularitarians believe this is our biggest existential threat, bigger than the threat of nuclear war. One of the more famous members of the church, Elon Musk, warned during a recent interview that AI research is like "summoning the demon."

Singularitarians Don't Understand Motivation

It is important to understand how the Church of the Singularity erroneously conflates intelligence with motivation. According to singularitarians, intelligence controls motivation and even creates it. More particularly, they believe that higher intelligence increases an intelligent entity's desire to dominate others. How do they know this? Again, they don't. There is no science behind it. What makes it even more embarrassing is that the Singularitarian priesthood seems completely oblivious to the mountain of clinical evidence compiled by psychologists over the last 100 years. The evidence has been accumulating ever since Pavlov began experimenting with his dogs. B. F. Skinner's behaviorist era did not refute Pavlov's findings but added more support to the existing scientific understanding of motivation. The evidence clearly contradicts the singularitarian doctrine. This conclusion is inescapable, not only in the empirical sense but also in the logical sense, as I explain below.

Intelligence Is at the Service of Motivation

The brains of humans and animals are born with hardwired pain and pleasure sensors. The brain does not decide what is pleasure and what is pain. This is decided by the genes. The brain can only reinforce behaviors that lead to pleasure or away from pain and weaken behaviors that lead to pain or away from pleasure. This is good old reinforcement learning which is used in normal adaptation. It is not magic, that's for sure. It consists of attaching pain or pleasure associations to various behavioral sequences. This favors certain behaviors over others. Animals and humans, to a lesser extent, also have preadapted programs that promote survival-related behaviors like mating and reproduction. The point I am driving at is the following. Likes and dislikes are neither learned nor created by the brain. They are the tools used by the brain to constrain and shape its behavior. Intelligence is subservient to motivation, not the other way around.

Knowing this, it does not take any great leap of the imagination to realize that using the tried and tested methods of psychology such as classical and operant conditioning, our future intelligent machines will be trained to behave exactly like we want them to. Better yet, they will continue to be faithful to their upbringing regardless of how intelligent or knowledgeable they become. Why? Again, it is because intelligence is always subservient to motivation. And where will machines get their motivations? From their designers and trainers, that's where.

Humans Vs. Machines

One is forced to ask, why do humans often stray from or rebel against their upbringing? The reason is that there is much more to human motivation than pain and pleasure sensors. How else could they rebel? We know that humans are motivated to enjoy things like music, beauty and the arts. These things cannot be anticipated and therefore cannot be programmed for in advance. So where does the motivation come from? This is a question that materialists cannot answer, not because they are too stupid to understand the answer, but because they are willingly wearing blinders that prevent them from seeing it. In other words, they have eliminated duality from consideration, not because they have a valid reason for doing so, but because they have allowed their hatred of other religions to get in the way of good judgement. That, in my opinion, is what's stupid.

Conclusion

I conclude that true AI is coming and it is coming sooner than most people expect. However, given my understanding of mainstream AI research, I'm willing to bet anything that it will come from neither the Church of the Singularity nor academia. We will indeed build extremely intelligent machines that will do their best to obey our commands and accomplish the goals we set for them. But they will not be conscious even if they behave emotionally. They will just be intelligent. So if there is a potential for catastrophe (and there certainly is), let us not rage against the machine. We will only have ourselves to blame.

See Also

Enthusiasts and Skeptics Debate Artificial Intelligence