Robots are cool.

Often, when one hears about AI, the topic of conversation is something grandiose. Strong AI, AI capable of conscious thought, is a fun thing to think about, but hardly something we can achieve currently, if at all. Discussions about this kind of AI are, more than anything, distractions from the equally interesting applications of AI that don’t require us to have a fully conscious machine.

For example, can we design a system that can adequately diagnose a mechanical problem, or a patient? Can we make a system capable of solving theorems or making logical deductions. These aren’t as exciting, but are certainly important to people working on AI, and are far more practical.

One of the first attempts at this was Cyc. Its goal is to be a sort of encyclopedia knowledge base capable of making inferences and answering queries. The database currently has over a million entries, and it’s been building since 1984. You can play around with it on their website, where they have a web frontend to query it. Interesting stuff. Of course, it still doesn’t know nearly enough. One suspects it might never be capable of learning enough.

The Turing Test is an interesting idea. The construction is as follows: you put a person in one room, a computer in another, and then have them communicating with a third person, whom we call the interrogator. The interrogator then asks questions of both and tries, with this information alone, to guess which is the computer and which is the man. The computer’s goal is to communicate with the interrogator so effectively as to convince him that it is the person.

Note that to be fair with the computer, our communication medium needs to be something like a chat room, something that separates physical perception from intellectual perception. Still, the test is weighted heavily in favor of the interrogator, since the computer must be capable of processing and using natural language in answers, and the interrogator has a 50-50 chance of guessing against the computer anyway.

Still, Turing was convinced in 50 years time that a computer would be capable of pulling off this feat… He said this in 1950. It’d seem Turing was off a little bit.

We all have an internal dialogue running in our heads all the time. It seems as though our very thoughts are tied to our linguistic ability. Given that this is the case, is it possible that these thoughts can ever be fully expressed in a language other than the one they are internally represented as? The linguistic relativity hypothesis, or Sapir-Whorf hypothesis, states that thought is so innately tied with language that there are thoughts in one language that are unique to that language and cannot be translated.

The Sapir-Whorf hypothesis has a weaker version that most scientists actually believe to be more accurate – that language impacts the thoughts and perceptions of people such that certain thoughts become more difficult to express in other languages, but not necessarily impossible. There are several studies that have been done about this.

Some of the more interesting ones include a study of people of different languages tasked to separate tokens into groups however they saw fit. Russians chose to separate light blue and dark blue into two separate groups, Setswana speakers would group green and blue. The theory is that these differences come from the differences in language. Setswana has no separate words for green and blue. Russian has two words for blue, one light and one dark. It doesn’t seem surprising to think language might have something to do with it.

I’ve taken several Japanese courses, and I’ve had friends ask me “how do I say X in Japanese?” Often, I just don’t know because I don’t have the vocabulary, but I’ve also seen that certain basic sentences just don’t really fit with any grammatical structures I’ve learned for Japanese. I guess that’s the linguistic relativity hypothesis at work.

Artificial neural networks are kinda nifty. To learn more about them, I thought I’d try playing with one. Here’s a couple of the places I tried: Blackjack with Reinforcement Learning and Java Mouse Neuron Test.

The first site uses an ANN to learn how to play Blackjack. You can start by playing alongside the computer in real time. The computer plays randomly at this point. If you want the computer to start playing better, the computer needs some alone time to get to learn the rules a little better. I set it to a 1000 learning episodes of 100 games each. As you run through the training you can watch the computers win average improving. It steadily goes from about 30%, which is about what you get playing randomly – to 40%. It then sits around 40% slowly improving. It’s impossible for the computer to really get above 50% since the game’s odds are weighted in favor of the dealer (the applet doesn’t account for pushes or splits).

It’s interesting how with just a few training episodes the computer can start really playing a better game. It’s hard to observe the overall difference playing as hands as slowly as a human needs to, but you can see the computer making more informed decisions after the training. ANNs are kind of nifty like that.

I decided to try to improve the training. I experimented with different values for winning and losing weights. Trying to find the right balance of reward/punishment to encourage learning in the quickest manner possible. I found weighting a win heavier than a loss was a quick way to favor learning, since losing is expected over 50% of the time any way. Weighting the wins too heavily did cause a lot more oscillation however, since the network started to think it understood what was going on just because of a chance win more often.

The other ANN I played with didn’t have any sort of training whatsoever. The associated paper discusses designing a neural net that learns more like a person and less like a computer with the hand of god involvement you’ll typically see. In the real world, there’s not someone to tell you every time whether you responded correctly or not, so the network was designed to learn in whichever way it desired, without a person telling it whether its actions were correct or not. I was able with a few tries to get the mouse to start to circle the goal without any sort of training whatsoever. I’d like to play around with this idea further – no training ANNs…

Artificial Life is an interesting concept. The idea is to make computer programs that model life in interesting ways and study them with the hopes of learning more about the process of evolution itself. For example, some A-Life programs start with a single creature programmed to learn about its environment and adapt. These creatures soon develop complex strategies that were not originally programmed. This is the case in the classic example of Animats.

The other possibility for A-Life programs is simulating entire ecosystems of artificial life. These I find more interesting, because they more often include genetic components and simulate natural selection. One such simulation is Noble Ape. I downloaded it and gave it a whirl. (Download Noble Ape Here.)

It includes all sorts of features, including weather simulation, random maps, senses for the simulated monkeys, etc. It even shows a simulation of their brain. The program even allows for scripting the monkeys to give them certain characteristics. All in all, it’s pretty fascinating watching the monkeys roam around trying to learn stuff.

The interface is kind of a pain to learn to use. The monkeys have several variables that define their overall behavior – including energy levels, fear, desire, along with location. These are presented in a sort of nonsensical array of gauges that don’t really make sense at first. It takes a little bit of practice, but it becomes easier to understand what is going on and you begin to see how their programming is really affecting their behavior.

A-Life doesn’t just focus on such complicated models though, it can also deal with simpler models. Lotus Artificial Life has some such examples. From the simple rules of the automata they have created, complex shapes and self replicating automata will form. This kind of study focuses more than the previously discussed examples on the effects of evolution.

One of our guest lecturers for Cog Sci in the past few weeks was a guy from a group at Georgia Tech investigating BCIs – Brain-Computer Interfaces. He had a very interesting topic to talk about – becoming cyborgs. There is a lot of interesting research being done in the area of interfacing computers with nervous systems. There are wheelchairs controlled by thought and robotic arms that respond to the same impulses our brain uses to control our real arms.

To me, one of the more interesting points of research pointed out by the lecture was the work of Kevin Warwick. He has had an array of electrodes implanted into his arm and used it to control, over the internet, an arm. Not only could he control it, but he was also able to receive stimulus from the arm, allowing him to control it blindfolded. His wife got an implant as well, and they set up bidirectional communication between the implants controlled by thought. (Read more about it in his paper: “Thought communication and control: a first step using radiotelegraphy”)

The result – a crude, but completely thought-driven means of communication. Kevin was able to send finger twitches to his wife, who would feel the twitches and could accurately identify them, and she could send back the same. This is a really cool idea to me. The ability to transmit thoughts, even if they are only simple twitches for now, is something that could be an extremely useful tool and lots of fun too.

What’s this mean for the average joe? Well, nothing for right now. But with all of the advances science has been making, bionic eyes, robotic arms, remote controlled rats… we could be seeing a future where people are as much machine as they are person. Being able to connect to the internet with a thought, being able to get a limb replaced with the full ability of the original arm… None of it seems quite as far fetched any more. Science has begun to understand the brain to the point where we can create technology that taps directly into it.

Memento is an interesting movie. I mean, yeah, it’s got the whole doing the entire story backwards thing. But it’s also got a character with anterograde amnesia [wikipedia]. Anterograde amnesia is something I first encountered while watching Memento. Whenever you hear about amnesia in movies, it always seems to be retrograde amnesia – the kind that always happens in soap operas – so I’d never really heard about anterograde amnesia. Apparently it’s more common though.

Anterograde amnesia largely occurs in one of two ways: drug-induced or trauma-induced. Drug induced amnesia can be overcome with time, but trauma-induced damage is typically permanent. Either way, the amnesia is caused by damage to the memory systems of the brain. While we are still unclear on how memory works in the brain, we do know the areas generally responsible for memory. The typical suspect in cases of anterograde amnesia is damage to the medial temporal lobe (MTL). The medial temporal lobe includes the hippocampus.

Memento features a character who is bashed over the head, causing damage to this area of his brain and destroying his ability to form new memories. The extreme extent to which he has this trouble may seem somewhat far-fetched, but if you’ve ever seen anything about Clive Wearing [youtube] then you can consider the main character of Memento as practically normal by comparison. The debilitating effect anterograde amnesia can have on a person is best exemplified in the movie when the main character is trying to remember a person as an enemy and finds himself struggling to find a pen to write this important fact down while he tries to keep the thought in short-term memory using rehearsal.

Anterograde amnesia often only involves what is know as declarative memory – memories of facts, such as what happened when. Often time the amnesiac retains the ability to form new habits or even simple skills and this is known as non-declarative memory. Also, in many cases, the person completely retains memories from before the cause of their illnesses. The contrast between these types of memory and how they are affected by anterograde amnesia definitely shows that the MTL is not the only place memory is dealt with in the brain. We see the main character in the movie using these differences to his advantage by using procedural habits to deal with his lack of memory.

Altogether, Memento is as intriguing a movie as anterograde amnesia is a condition. It’s definitely worth watching, even (or especially) if it was edited together to play out the story backwards. (Each colored scene in the movie happens chronologically before the scenes shown after it, and the black and white shots spliced in happen before all of the color events.) In fact, this style of story telling makes perfect sense from the point of view of the main character, whose life is not linear, but just a bunch of independent scenes.

Last Tuesday we had a class discussing Autism and Autism Spectrum Disorders. But… I wasn’t there. In light of this, I decided to read up on Autism and watch the video shown in class.

(The video is: The Mind of a Visual Thinker [youtube])

Temple Grandin is a fascinating person. Her talk (see video above) on visual thinking was a great look into the thought process of someone one the Autistic Spectrum. Apparently she doesn’t think in words like many people, but instead thinks completely in visual pictures. When she hears the word steeple, she doesn’t just call to mind attributes of steeples, or even a generic steeple, but actual pictures of steeples she has seen at some point.

To Grandin, the major feature tying together the various parts of the Autistic Spectrum, is that all have a different way of thinking about things from those that are not in the Autistic Spectrum. She mentions three specific ways of thinking that many in the Autistic Spectrum find themselves in. One is thinking in pictures, which is the way Grandin herself thinks. Others may think in patterns or even in sounds, but the common theme is that these people think in one way more exclusively, rather than being able to easily switch and relate the different ways of thinking.

The physiological causes of Autism have not been found, but the current theory of how the behavioral differences develop follows this model:

The idea being that the genetic and possible environmental factors lead to a difference in the way the brain develops, which affects how the person is able to think, and finally results in an observable difference in behavior. Work is currently being done to develop a model that explains exactly how this process occurs in a way that fits current evidence. (A Cognitive Model of Autism [cc.gatech.edu])

In understanding Autism Spectrum Disorders, we do important work not only in helping those with Autism survive in a world of what Grandin refers to as neurotypicals. but also in understanding how the mind in general works.

http://en.wikipedia.org/wiki/Gestalt_psychology

(All of my pictures are stolen from Wikipedia)

Gestalt psychology focuses on the idea that the whole is more than the sum of its parts. Gestalt psychology brings with it four main principles: emergence, invariance, multistability and reification. On top of that, there is also the application of gestalt to perception: prägnanz – our tendency to order our experience in a manner that is regular, orderly, simple and symmetric.

Emergence, the first principle given by gestalt theory, is demonstrated very well by above picture. The dog in the picture is not found by noticing a nose, then an ear, until we say, “a ha! That must be a dog because of all the features like a dog.” We notice the dog all at once. The dots organize themselves into that pattern almost magically.

Invariance is the second principle which states that our mind has a tendency to recognize objects regardless of scale, rotation and translation. Further, we can recognize objects despite warping and skewing, or differences in representation.

Third we look at multistability, our brain will waver between various interpretations of ambiguous experiences. Look at the figure below:

Finally, there’s reification – or generative perception. Our brain has a tendency to fill in the blanks to make a picture more understandable. In C below, our brain has a tendency to see a ball with spikes, not because a ball is drawn, but because the organization of spikes makes more sense in a 3d perspective of a globe.

The rules of prägnanz or the “Gestalt laws” of perception are all pretty simple. The law of common fate, the law of closure, the law of continuity, the law of symmetry, the law of similarity and the law of proximity, are all fairly self-explanatory. (See wikipedia if you need help). They have a real application for HCI purposes as well. If we know how people tend to group objects, we can apply these grouping to apply logical groupings for users.

Overall, the only real problem with Gestalt psychology is that is more a list of observations, and does almost nothing to explain how the brain works in this way. This has drawn a lot of criticism from other psychologists, seeing as it does nothing to extend our understanding. Still, knowing how people tend to group and classify objects is useful, especially when you’re going to be creating some sort of interface.