The Lojban Experiments

About 12 years ago, I developed a prototype which used Lojban as a type of meta-language. You could interface with the system in Lojban or English via a microphone, keyboard, text files, or email. If you used English the interface (a separate program) would convert the communication to Lojban prior to it being exposed to the prototype. If you used Lojban, then the interface would simply pass the Lojban through to the prototype.

The original idea was that I wanted to have two prototypes running that could communicate with each other, as well as providing a way for human interaction. What we studied with these two (and eventually six) prototypes was the evolution of language between the two prototypes. Hmmm… maybe I better take a second to explain in more detail.

Generally speaking humans are slow, with respect to the speed at which they learn, conceive of new ideas, and communicate. Alternatively, the prototypes we were building at the time were able to deal with large amounts of concepts and information, in a very short amount of time. So, I had the idea (which isn’t very original today) that if computers would be thinking faster than us, then one human day of thought, might take place in 1 hour of the prototype’s thought. As computers become faster, the less time it would take a computer to think through 1 human day of thought. So, if a computer can live 1 human day of thought in 10 minutes, let’s say, then an A.I. will be able to live the mental life of a human in considerably less time. So look at this for our example of 10 min (computer) for 1 day (human) mental life:

In 1 hour, the computer will live 6 human mental days

01 days … 144 human days

07 days … 2.7 years (1008 days)

30 days … 11.8 years

01 year … 144 years

07 year … 1008 years

75 year … 10,800 years

Okay, so you get the picture that at 10 mins for 1 human day that a lot can go on, especially considering that part of that human day is used for sleep and relaxation. If we looked at it more like humans put in a good 8 hours of mental thought in one day, then that would be more like 3 days (8h X 3 = 24 hours) for each 10 mins. That is 432 human mental days for each day of computer thought. Now consider faster machines where in 1 second of computer thought, 10 or even 100 human mental years pass…

Regardless, the point is more that during this time, not only will the prototypes be able to think faster, solve more problems, and conceive of more new ideas; but it will also be able to evolve intellectually faster. Think about how the system could develop new sciences, more accurate math, and more computer scientific models. Now consider how it will evolve language. Now it can not be designed so that it leaves us in the dust, so the idea was to have two language paths. The first being the meta-language which it uses to communicate it’s thoughts and concepts. The second is to communicate with humans.

Now, the prototype was designed to use Lojban as the meta-language, mainly because it had a long history, simple grammar and rule sets, only 1300 root words which can easily and logically be combined to form millions of words, etc. If you look at the Lojban website, you will see the many other qualities which make this a perfect synthetic language to use for not only computers, but this rapid evolution of language. Also, the other grand notion was that Lojban is easy for humans to learn, not to mention many books have been translated to Lojban.

So, here is the situation. You have two prototypes. One of the prototypes can talk to outside sources via any of the interface options mentioned above. The other prototype can only communicate in Lojban, which means that it can either read Lojban text files, or talk to the other prototype. Over time as we worked with the systems more and more, we had them work more co-operatively on various problems and tasks. (This is the long story short part) What you start to realize is that the two machines make up this community.

The notion of cultural evolution and the idea of family arose. So, we augmented the prototypes to be able to “procreate” in some sense. We want them to evolve after all. Oddly enough they don’t’ have any genetic material, so one would think that that would produce a problem when it comes to breeding. Ha! No, the procreation has more to do with combining two or more A.I. “knowledge-sets” (their minds / memory / knowledge / logic), to end up with a new combined knowledge-set.

Hmmm… I need to clarify again… the point of the procreation is that over time you might have, say for example, two companies that do business together. Each company might have an AI which deals with all of the companies “data.” Over time these two companies might work together exclusively and because of that relationship they might want to dedicate a new system to work with aspects of the combined data; BUT only the data that is relevant for both companies (say 40% of one company’s data and 35% of the others). The rest of the data is confidential. Basically they want a separate AI for this interchange because it will allow them to work faster. The two AIs can “breed” and produce an “offspring” which has only the skills, data, and logic required for the specific combined task, and not the rest of the skills, data, and logic of the two parent systems.

During our research however, we only “breed” 4 children. Basically we found some interesting things about how the vocabulary evolved (the root/common words changed over time and generations), but the syntax of Lojban is static mainly because that is a structure of the language. Another strange result is that the systems would short hand whole concepts and ideas. So, this was viewed as interesting from one aspect, but with respect to the breeding idea, it turned out that it was in many ways faster for the systems to just communicate rather than going through the trouble of breeding. The research down that path was ultimately abandoned to more advanced technology that we developed; which in part was due to what we learned about language from this and other projects.

P.S. Sorry if this entry seems like I am rambling... it all jsut brings back a bunch of old memories.  Plus, I can't help but wonder if they were still running, what they would be like...

I've been thinking a lot about the Sapir-Whorf Hypothesis, which is basically the idea that there is a systematic that your native language influences the way your mind perceives the world. One of the more interesting aspects of this hypothesis is that it is impossible to translate ideas/concepts from one language into another language.

One of my on-going goals is working on an ontology which is subdivided by grade levels (K,1,2,3,4,5,6,HS,Univ) allowing the ability to run prototypes with core-knowledge with specific ranges, or specific levels of knowledge. You can specify K-4, or even K,2,6 if you like and study the resulting streams of thoguht, logic, reasoing, association, etc. This allows me to study how various concepts influence other concepts, ideas, language, thought, etc.

Now, the point of mentioning these two things, is that I am curious what would happen if the core-knowledge of an A.I. prototype of mine contained the language / concepts / ideas / culture of multiple languages, in effect making all fo the languages the native language. Because of how the prototype is designed, the system will easily be able to communicate in any language found in the ontology so that is a given. The facinating idea here is that the system will be able to fuse ideas and concepts from various languages without having to translate. Also, the system will be able to translate more accurately than any previous algorithmic method of translation; which is also augmented by the systems ability to derive the context and meaning of any text.

Which brings me to my experience-experiment. Over the next 8 weeks I am going to take 5 introdutory language classes offered by the local school district. In college I took 4 years of American Sign Language (which I love for it's direct, honest, exacting nature). So this will be a very different experience.

Italian -- Monday
Spanish -- Tuesday
Russian -- Wednesday
Mandarin Chinese -- Thursday
German -- Friday

The goal, in part, in taking actual coarses with instructors is to garner more cultural information, as well asto study the experience of being introduced to a new language, as well as observing how each teacher goes about teachign the language (where they start, how they interact with the students), and also monitor the difficulties which other students have and how they address those difficulties. All of the languages were choosen because of my limited exposure to each (although I am familiar with all of them to some extent), and because the cultures of the coresponding countries are all veyr different from one another.

On the other end of the spectrum, the indormation from the course will be added to a duplicate copy of the core-knowledge ontology which is used in my prototypes so that the original can be perserved, and the multi-language/cultural version can be interchanged at any time. Each language will in intergrated into the new ontology with the ability to not only add and remove grades levels and ranges but also add and remove specific languages. The most difficult aspect of this process will be dividing the new language data by grade level, so I may determine an alternate scale by which to delimit the new language data; but the result will be a consitant scale for all of the new languages.

I have also purchased the language voice prints for the speech synthesis system, and derived a method for the the speech recognition system to be able to quantify which language the system is hearing, as well as, being able to jumble a sentence or converation with multiple languages. The language genesis system has also been altered to produce language in one of three ways: (1) the system can be set to always produce english (2) or the system can dynamically utilize it's vocabulary [so that if the word blue is missing from the ontology because grade 2 is disabled, it can use the corresponding word in another language]. The language choosen is not randomly choosen, but rather more probabistic in nature based on associative values of corresponding concepts in the ontology. (3) the system can be set to produce language to match the audience with which it is speaking. This utlimately means that if you speak to the system in one language it will respond in that language. The system will perform in this case just as if it only had the language at hand.

Well my first class starts tomorrow night. I'm excited to get things underway, now that all of the system modifications have been made and tested. Wish me luck.

A question about Einstein theory of relativity being proven was mentioned in response to one of my posts about time travel.

I just wanted to make a few points about that question. First, Einstein’s view of the universe was far from perfect. He made several mistakes; for example, he like most of his peers at the time, was convinced that the universe was static (not expanding or contracting), and in 1929 Edwin Hubble discovered that other galaxies where swiftly moving away from our own. As a result Einstein revised general relativity, claiming how this had been the biggest mistake of his career.

Even more interesting, is a physicist named Richardo Carezani, who in the 1940s found mathematical errors in Einstein’s equations. He corrected the equations and derived an alternative theory based on the corrected calculations; this theory is referred to at Autodynamics. The implications purposed by Autodynamics, are that: (1) special relativity no longer exists, (2) light speed is a constant not a limit, (3) mass, time, and length do not change because of relative movement, (4) mass can be created from pure energy, (5) a mechanism for gravity can be described. The basic idea behind Autodynamics, is that the equations for the Lorentz transformation are incorrectly formulated to describe relativistic effects, which as a result causes special relativity and general relativity to be invalid.

The Hafele-Keating experiment is the famous atomic clock experiment; which is often given as the proof to general relativity. It is important to note here that this experiment does not prove the whole of general or special relativity. Much of what is said about the experiment is that the data was altered to give promising results. I believe it was Dr. Domina Spencer who analyzed the raw data from the experiment and came to this conclusion in the mid 1990s. As it turns out, atomic clocks do not track the decay of a single atom, but rather a cloud of atoms. The instruments involved in the process of measuring the decay of atoms is very delicate; which is often given as another reason why the two atomic clocks differ. Regardless, even if the experiment is valid, it does not conclude that everything stated in Einstein’s theories and thought exercises are true.

Next, there is another experiment, the Sagnac Experiment, which is often given as a disproof of the theory of relativity. This experiment involves splitting a laser beam, and having it travel in opposite directions. There is a debate raging about whether new math can correct this problem, or a reinterpretation of the theory of relativity. Some people talk about which frame of reference various components of the experiment are present, some talk about how knowing the relative velocity of the observer disproves the theory of relativity, some look at the difference in time actually working out to be the planet’s rotation; regardless, this is another example of a problem with the theory of relativity which needs to be resolved.

For me, the theory of relativity tells me less about physics and more about how the mind’s perception of it’s environment is changed based on how, where, and what the mind is observing. Also keep in mind, that as science evolves, and as time passes, Einstein’s physics have fallen to the way side, like Galilean physics, and Newtonian physics. Quantum mechanics, M-Theory, and several other branches of physics are now at the forefront. (This does not mean that we don’t use aspects of previous theories and ideas and the math from the older theories.)

I was having dinner with some friends and we were discussing, all of the recent buzz surrounding time travel. Now understand, I have always had a problem with Einstein with respect to a few his predictions based on his theory of special relativity. I am not a big fan of the whole concept of bending space, and the view that time is a dimension, and that the faster you travel the faster through time you move, etc. The main problem I have with the theory of relativity is that it’s based on having an observer and an object of observation. The idea that you travel through time because you observe someone else’s clock moving slower is just silly to me. I have mentioned before in a previous post how the brain has a time loop via which the brain takes snapshots. This loop can be speeded up or slowed down altering your perception of not only the passage of time, but your perception and view of the world around you. (The issues I have with this whole idea are actually more involved but would require me to speak at length about the nature of the mind and my theories). Regardless, of Einstein, I am not a big fan.

So, at dinner I purposed a “humorous” exercise with respect to the idea of Einstein’s physics with respect to time travel. Basically, the “joke” is that since the faster you travel, the faster through time you move, it would be funny if the light we see is actually time traveling. The funny thing would be if sun was about to go super-nova and we couldn’t even tell. Of course, at dinner, the first question is how can that even be possible?

The idea is the same as seeing the light from a star that goes super-nova hundreds of light years away. We receive the light from so many light years away and hence when you look at far away objects we are looking into the "past" because the light we see is from x number of years ago (i.e., a star that is 200 light years away goes super nova 150 years ago, we will not see the super nova for 50 years, and during the next 50 years we just see a star, which is -- in fact -- long gone).

The joke is that the same thing could be true of our sun, that the light that we are viewing and studying from our sun is from the light jumping forward in time, the idea being that rather than the light having to travel a distance, it travels through time. Picture this… the light we are seeing has jumped forward in time 200 years, but the sun went super nova 100 years ago. The sun is "gone" (brown dwarf just giving off ultraviolet radiation, and not visual light), but we still have light and an image of the sun because the light is coming from the past where there is/was a sun. 

Okay that was the explanation... The real "example" would be more like the sun is created 10 billion years ago, but the earth is created 5 billion years ago. The earth is created when the sun goes supernova after the sun has been around for 5 billion years; but the light from the first 5 billion years travels through time 5 billion years so that we have light and the image of a sun for the next 5 billion years. Well, we (the earth) are 5 billion years in now (give or take a few hours ;) *laugh*) and the sun is going to go supernova in T-minus 4 minutes. (of course the sun is gone… but the nice thing is that the radiation and light from the supernova will now time travel into the future and kill us.) Hmmm.. Of course the energy from the supernova together with the bending of the time-space continuum from the super heated plasma from the explosion, would according to their logic, keep the fold in time-space active allowing the light to travel into the future, from the point in time space of the creation of the sun, while also keeping it's momentum through the dimension of time… basically causing the loop/fold in time space with a locus at the singularity from the creation of the star to it's destruction. Which would be the only way for the math to balance out just leaving the dimension of time to be traversed… blah… blah… blah… using one abstraction to explain another… think they might have a problem there? I would say yes.

Okay, the “Einstein fact” that no one else at dinner brought to bare, is that Einstein believes two important things about light: (1) light always travels at a constant speed, regardless of how fast you are traveling. (2) What Einstein actually says is that you begin to time travel as you accelerate to the speed of light.

The really funny part to me now, is that someone like Einstein would say that light doesn’t accelerate to the speed of light; but rather, when light is created, it is instantly traveling at the speed of light. This is funny because that doesn’t make since, and there is really no way to explain how light in a star, when it’s created is instantly traveling at the speed of light without accelerating. Oh, the funny part, is that if the light was time traveling from the past, it would arrive in the observers time-space, already moving at the speed of light. Even more funny is that when the light is created in the past, as it almost instantly accelerates to the speed of light, the light (photons) would actually time travel.

Okay, so the “joke” isn’t funny “Ha Ha”… but the more I progressed through my joke, the more I realized that theoretical physicists might actually like this idea. Then I realize, “ugh!” if someone changes the joke where they use a black hole instead of the sun, then the light would travel backwards in time and…. No.. NO! We must stop the madness now! *laugh*

Well, I wasn’t able to sleep very well last night, and as a result I found my self, organizing my thoughts on how to resolve the debate about the two theories of olfaction.

Basically, there are two theories:

1. Shape Theory -- states that the sensation of smell is due to a “lock and key” mechanism by which a scent molecule fits into an olfactory receptor site in the nose.

2. Vibration theory (Luca Turin) -- states that the sensation of smell arises from the nose's sensing of the rate of electron tunneling into the scent molecule. This rate will show resonances with the liquid-state vibrational spectrum of the molecule. Some evidence supports this model, while some evidence supports the 'lock and key' mechanism.

The most fascinating thing about this debate is that it boils down to “is it a particle, or is it a wave,” which is the same debate they are having in physics. When I first heard of this debate about how we smell, I was struck by how everyone went about even thinking about the problem. Even more perplexing to me, was that after the discovery of the receptor genes by Buck and Axel, everyone seemed to resolve that the Shape theory was the accurate solution.

Before I delve into my solution to this debate, first I must point out a few things which I feel are important when viewing the process of olfaction. First, they have taken a very humanistic approach to the study of olfaction. It is very well known that the human sense of smell is very poor on all counts. We have very few receptors and out ability to discern scents is generally very poor compared to most other mammals. Often you will find that other mammals will use there sense of smell as their primary sense for viewing and experiencing the world. (It was almost humors to me, when I realized that a dog wants to stick his head outside of the window, not because he wanted to breath, or because he was car sick, but rather because he wanted to “see” the world, and to do so required him to give his nose access. And even more interesting, a dog can “see” farther with his nose than we can with our eyes – bringing a whole new meaning to taking the scenic route.)

Another interesting idea to consider is how our vision work, or rather the reception of light by the eyes. The eye contains two types of receptors, rods, which detect photons [particle] and, cones, which detect the frequency [waves] of light. The signals are always separate and are processed in different sub-cortexes in the visual cortex (there are over 30 sub-cortexes in your visual cortex.) Some animals do not have cones, or have very limited color vision. To move from just black and white vision to color vision requires time and evolution. The point here is that as the brain evolves it takes on new means to process information while maintaining the previous methods of accomplishing the same or a lesser form of the same task.

In most mammals, the olfactory stimulus is processed in the piriform cortex. As mammals became more evolved the dorsal part of this cortex increased in volume. In humans, it evolved to an amazing extent, forming the entire neocortex. This is very interesting to me, and in part the reason why I have an interest in the olfactory system (Knowing how one functions will give you some insight into how the entire neocortex works in some way).

Now with those two ideas in place, let me interject why the solution isn’t as simple as purposing that the olfactory system functions like the visual system. The “catch” is that the olfactory system only has one type of receptor, not two types like the rods and cones of the eye. Thereafter, it is also important to note, that the odorant receptor in your nose, does not fire when the odorant molecule is placed in the receptor site of the neuron. What actually happens is that odorant specific proteins (G_olf) are released in the cell that in turn causing a “chain reaction” (GTP -> ATP -> cAMP => influx of Na+ and Ca+) which results in the cell opening ion gates which as a result depolarizes the cell. The point here is simply that the act of receiving the molecule itself does not cause the receptor neuron to fire (there are two types of ion gates and an exchanger at play here).

SO! What’s my resolution you might ask… well first let me say that in the case of a system like vision where you receive and process both “particle” and “wave” stimulus, it makes me think of something like music. Music? Well by taking the analogy of music, think of the visual system, where a percussion instrument would be like a rod detecting a particle, in that a percussion instrument plays one note sharply. In the case of the cones, you would have a string instrument, which would play the note more melodically. The result in both cases is that a note, or signal, is sent and detected as middle-C.

Now, you can make music with either type of instrument, but the percussion instrument has less precession, and with the respect to sensory reception the particle method is less evolved. On the other had, the String instruments add more precession, but has a trade off of wether something is absolute (more green than yellow, more blue than green, etc.).

So, this still does not resolve the most pressing issue; that of the single receptor type for olfaction. Now, given there are 347 different types of receptors, the difference has to do with which molecules are detected by which receptors (each molecule fitting in it’s receptor like a key in a lock). As mentioned before, there isn’t a distinction like with rods and cones, where you have a symphony of instruments playing together to create master pieces of music. So we still need something to bridge the gap between string instruments and percussion instruments.

Have I mentioned that I took piano lessons when I was younger? It is a beautiful instrument creating an entire spectrum of sound and music. The interesting thing about a piano is that in many ways it is both a string and a percussion instrument. (There are also clavichords, and harpsichords. A harpsichord, plucks the string, where a clavichord strikes the string leaving the tangent in contact with the string to set its pitch, and a piano strikes the string and is immediately removed.)

So the simple explanation of my idea is that the receptor may have evolved to detect the vibration of the molecule in several ways. Which can range from the quantity of the various types of ion gates opened and to the exchange rate, to a variation in the duration of time the molecule resides in the odorant receptor (so, if a similar molecule enters and triggers the process which leads to depolarization, but the cell ejects it prior to the depolarization, then another molecule with the missing part of the frequency is received and the same process occurs with the ejection of that molecule; but both molecules resulting in the cell crossing it’s threshold and depolarizing. I would go on, but I have a problem with all of this now.

My problem deals with the simple concept of the sense of smell. The current idea is that an odorant enters and odorant receptor and that is not only the smell that we experience, but is also the experience of smelling. Unfortunately, nothing works like that in the brain, not to mention that I rarely hear someone say that they smell something specific. Generally I hear people say, “this smells LIKE almonds.”  Regardless an even bigger concern is that they are lost in the details of reception, and have lost site of the fact that there are several molecules which make up smells and out perception of smells and odors. Think about wine tasting, and how you describe the bouquets of smells present.

This leads me to the more complex version of my music analogy, in which you consider things like rhythm, harmony, and melody, where you can consider things like cords and. At an even more advance level you could relate the qualities of musical textures (monophony, hererophony, polyphony, homophony, monody, micropolyphony); and even extending the concept further to simultaneity with respect to musical textures. All of this ins very interesting to me, especially considering that they are learning more and more about the temporal aspects of olfaction in insects.