AI - 2.01 - genAI - build use LLMs

AI - 2.01 - genAI - build use LLMs

Having discussed some of the issues around artificial intelligence, in general, and some of the various historical approaches, we are now, finally, ready to talk about generative artificial intelligence and large language models.  These are the backbones of the current crop of artificial intelligence products that are being promoted quite heavily in our society.

As previously noted, this is built on the mathematics behind Bayesian analysis, Markov chain analysis, neural networks, and so forth.  Using the mathematics here, the companies that have built generative artificial intelligence chatbots have created statistical models based on enormous amounts of text data.  This text data has come from books, it has come from the news media, and, of course, lots and lots and lots of it has come from social media.  Social media is a free source of a huge amount of text based on people conversing with each other.

Building these statistical models is not easy, and the resulting statistical models, themselves, are not easy to understand.  As a matter of fact, if they are honest, the companies that have built these statistical models will, themselves, admit that they do not understand everything that is in the models that they have built.  After all, it is not they that have built the statistical models.  The statistical models have been built by computer programs that have done statistical analysis of these masses of text.

It is hard to explain just how complicated this process is.  In one sense, it is very simple.  It is simply looking at a lot of text, and making a statistical analysis of which words come in what order, what word comes after a certain word, and how often, with some extra statistics thrown in to indicate how often this word comes four words after that word, and so forth.  But the thing is, that the statistical analysis goes on at many levels, and the statistics that are built get modified according to the mathematics of neural networking theory, which is looking for relationships, sometimes relationships between the statistics themselves.  It's all just numbers, and it's all just ones and zeros, but it keeps on going, and the end result is enormously complex.

This is why such enormous amounts of money are being put into this effort.  Yes, there have been artificial intelligence programs that have been built on specialized computer equipment.  When IBM built Deep Blue and Watson, they were built on specialty computers, which were created specifically for the purpose of running those artificial intelligence programs.  The work that went into creating those programs, and the work that went into creating the hardware for those programs, have, certainly, spun off benefits for the fields of both hardware engineering, and program design.  But they were one-off attempts to address specific challenges.

The building of the large language models has required the construction of entire data centers.  Enormous computers, filled with what would normally be specialty processors within other computers, that have been specially designed to perform a certain type of mathematics.  This type of mathematics is one that has been widely used in generating graphics on computers, and so one particular company, formerly known simply for creating the chips that were helpful with making graphic cards for computers, has come to be enormously valuable in the midst of this race to create artificial intelligence.  I should note that the same type of mathematics is the mathematics that goes into trying to break encryption systems, so these type of chips do have more than one purpose.  Prior to the demand for these chips because of the artificial intelligence boom, a lot of people were using them to build cryptocurrency mining devices.

But now there are enormous data centers, which are, in reality, just single computers, created by putting together thousands, and sometimes millions, of these specialty processing chips.  This demand for processing power in order to accommodate research into and the use of, artificial intelligence, and particularly generative artificial intelligence, is so great that other companies are now building power plants, solely for the purpose of powering these particular data centers, solely for the purpose of using creating large language models for generative artificial intelligence.

The creation of chatbots is not new.  Microsoft, rather infamously, tried it some years ago.  They created a chatbot, and put it up on the social media platform Twitter.  In a few hours, the chatbot was taken down.  What had originally been seen as a polite and helpful commentator, had, within hours, turned into a foul mouthed combatant.  The chatbot had been designed in order to use the text that it encountered to build and improve itself.  The thing is, the conversations on social media aren't always polite.  The improvement didn't improve things any.  The chatbot learned to be a troll.

So, it turns out that, one of the things that you really need to be careful of, with regard to generative artificial intelligence chatbots, is that they don't go off the deep end.  You need to build in some kinds of restraints.  You can't just let them learn, and then accept whatever it is that they produce.  No, instead, you need to make concerted efforts to ensure that the chatbot is at least somewhat reasonable in terms of its conversation, and that it doesn't give people useful information about how to kill themselves, or how to make weapons of mass destruction, or various things like that.  Creating these restraints is known, in the field, as guardrails.

Creating guardrails turns out to be a non-trivial problem.  People who are interested in the field have attempted to get around the guardrails, and, in all too many cases, it has turned out to be surprisingly easy.  Sometimes it is the researchers who have found the ways to make chatbots spit out very dangerous information.  Sometimes, unfortunately, it is the users who have found that the chat box are all too willing to encourage them to commit suicide, and counsel them that painful ways of dying aren't really that bad if it ends up fulfilling your objective not to exist.  In addition, there is an ongoing problem, now identified as AI psychosis, which is that, partly encouraged by the publicity and promotion of the generative artificial intelligence companies, people have come to regard chat bots as having personalities.  People have created chatbots with personalities.  People have created chat bots as artificial friends, sometimes artificial lovers, and in a great many cases artificial representations of a grieving individual's dead loved ones.  A number of psychological issues are only just starting to be examined with respect to this particular risk.

We'll deal with this issue of chatbots in some detail later.  However, there is another side to generative artificial intelligence, and that is in regard to the systems that create graphical images or even video.

These systems use very similar mathematics and technologies to the text-based chat box.  However, the graphical systems are fed masses of image data, usually image data that has some accompanying text.  Therefore, the graphical systems are able to respond to prompts that are involved as queries for certain types of images, by producing images that are going to be similar to images associated with text similar to The prompt that is issued to the system.

And, now that I have used the word prompt, I have to explain it.  Most people who are dealing with artificial intelligence through chatbots are used to thinking that they are asking a question, and the chat bot is giving an answer.  This is, quite simply, not true.  Using a generative artificial intelligence chatbot means that you are issuing a prompt to the system.  The prompt is the "question" that you type in.  This system, however, does not know that this is a question.  It doesn't know what a question is.  It just knows that you have typed in certain text.  And then uses the enormous statistical model to generate a stream of text which is, statistically, probable based on the string of text that *you* typed in.  That is, the statistical model is making a match, based solely on mathematics and statistics, between the words that you have typed in, and strings of words that have followed strings that are similar to those that you typed in, in the masses of data that were fed into the system in order to create the statistical model.  This is not question and answer.  There is no understanding involved here.  What is happening is that the system, with layers and layers of mathematics, is simply generating a stream of text that is statistically probable, based on the analysis that it has previously done of tons and tons and tons of text.

Your question isn't a question.  It's just a prompt.  In cryptographic terminology, we would say that it is a seed.  It'll produce something, but what it produces is based on mathematics, not understanding.

In terms of producing graphics or video, sometimes the situation is even worse.  In terms of encrypting graphics, you have to use methods that are somewhat different from the encryption that you do with regard to text.  If you use methods that work very efficiently in hiding text, in terms of encrypting graphics, very often you will come up with a result where the original image maybe somewhat fuzzy, but you should be able to get the general idea.  That's not good in terms of encryption.  Therefore, the process that we use in encrypting graphics often uses something called diffusion.  This means that we take the actual information in the image, and move it around, so that the information is actually all still there, but it's no longer next to other information that will recreate the image and let you know what the image is and means.

When you ask a generative artificial intelligence system, which creates graphics, to create a picture for you out of something, it usually actually starts with random noise.  And then, using the same mathematics that would go into diffusing an image, so that it no longer appears to be an image, we run that process backwards.  You have heard the old joke that it's easy to create a statue of an elephant.  All you have to do is take a large block of stone, and then cut away everything that doesn't look like an elephant.  Although the process is complex and heavily mathematical, this is, essentially, what image generation generative artificial intelligence systems actually do.  They take noise, and then move it around, throwing away everything that doesn't look like an image that is similar to an image that is associated with something like the text that you typed in.  Again, there is no comprehension or understanding involved here.  This is one of the reasons why, when you first start trying to use the graphical generative artificial intelligence systems, you have to make many tries, and teach yourself, how to word a prompt so that you will get an image that is something like what you want.  (For example, these systems don't understand how many arms or legs human beings have.)  It's a bit of a trial and error and frustrating project.

AI topic and series

Previous: https://fibrecookery.blogspot.com/2026/02/ai-110-history-neural-nets-to-llms.html

Introduction and ToC: https://fibrecookery.blogspot.com/2026/01/ai-000-intro-table-of-contents.html

Next: TBA

AI - 1.10 - history - neural nets to LLMs

AI - 1.10 - history - neural to LLMs

When babies are learning to talk, they reach a stage which pretty much everybody refers to as babbling.  However, if you pay attention, careful attention, to what they're doing, you will realize that they probably think that they are actually speaking.  They have learned the patterns, or at least a number of the patterns, that we use when we are speaking.  The sounds that they make may not sound like English words to us, but you will notice that the pauses that they make when they are babbling, and head tilts, and possibly even movements of hands, copy what we do when we are speaking.

They are learning to speak, and they learn to speak by copying the patterns that they see us using.

There are many patterns in our use of language.  You probably know that the letter "e" is the most commonly used letter in the English language.  The most common consonant is "t."  A number of the patterns are statistical.  When we can copy a sufficient number of these patterns, we can use the statistics, just the statistics, and nothing else, and nothing to do with any kind of meaning, to create a string of text that looks very much like the English language.  In another area that I have studied, forensics, there is a field called forensic linguistics, or stylistic forensics, which we can use to look at even more detailed patterns of statistics in text, and actually determine the specific author of a piece of written text.

Now, some of you may be somewhat suspicious of the proposition that a mere statistical analysis, no matter how complex, can generate lucid English text.  Yes, I am oversimplifying this somewhat, and it's not just the probability of the next word that is being calculated, but the next three words, and the next seven words, and so forth.  The calculation is quite complex, but it still may sound odd that it can produce what seems to be a coherent conversation.

Well, this actually isn't very new.  There is a type of statistical analysis known as Bayesian analysis, or Markov chain analysis.  It has been used for many years in trying to identify spam, for spam filters for email.  And, around twenty years ago, somebody did this type of analysis (which is much simpler and less sophisticated than the large language model neural net analysis) on the published novels of Danielle Steele.  Based on this analysis, he wrote a program that would write a Danielle Steele novel, and it did.  This was presented to the Danielle Steele fan club, and, even when they knew that it was produced by a computer program, they considered that it was quite acceptable as an addition to the Danielle Steele canon.  And, as I say, that was over two decades ago.  And done as a bit of a lark.  The technology has moved on quite a bit since then, particularly when you have millions of dollars to spend on building specialized computers in order to do the analysis and production.

One of the other areas of study that I pursued was in psychology.  Behavior modification was a pretty big deal at the time, and we knew that there were studies that confirmed how subjects form superstitions.  If you gave random reinforcement to a subject, the subjects would associate the reward with whatever behavior that they had happened to be doing just before the reward appeared, and that behavior would be strengthened, and would occur more frequently.  Because it would occur more frequently, when the next random reward happened, that behavior would likely have occurred recently, and so, once again, that behavior would be reinforced and become more frequent.  In animal studies it was amazing how random reinforcement, presented over a few hours or a few days, would result in the most outrageous obsessive behavior on the part of the subjects.

This is, basically, how we form new superstitions.  This is, basically, why sports celebrities have such weird superstitions.  Whether they have a particularly good game, or winning streak, is, by and large, going to be random.  But anything that they happen to notice that they did, just before or during that game, they are more likely to do again.  Therefore they are more likely to do it on a future date when, again, they have a good game or win an important game.  This is why athletes tend to have lucky socks, or lucky shirts, or lucky rituals.  It's developed in the same way.

One of the other fields I worked in and researched was, of course, information technology, and the subset known as artificial intelligence.  One of the many fields of artificial intelligence is that of neural networks.  This is based on a theory of how the brain works, that was proposed about eighty years ago, and, almost immediately, was found to be, at best, incomplete.  The theory of neural networks though, did seem to present some interesting and useful approaches to trying to build artificial intelligence.  As a biological or psychological model of the brain itself, it is now known to be sometimes woefully misleading.  And one of the things that researchers found, when building computerized artificial intelligence models based on neural networks, was that neural networks are subject to the same type of superstitious learning to which we fall prey.  Neural networks work by finding relations between facts or events, and, every time this relation is seen, the relation in the artificial intelligence model is strengthened.  So it works in a way that's very similar to behavior modification, and leads, frequently, to the same superstitious behaviors.

The new generative artificial intelligence systems based on large language model are, basically, built on a variation of the old neural networks theory.  So it is completely unsurprising to see one of the big problems that we find with generative artificial intelligence, is that it tends, when we ask it for research, to present complete fictions to us as established fact.  When such a system presents us with a very questionable piece of research, and we ask it to justify the basis of this research, it will sometimes make up entirely fictional citations in order to support the proposal presented.  This has become known as a "hallucination."

Calling these events "hallucinations" is misleading.  Saying "hallucination" gives the impression that we think that there is an error in either perception or understanding.  In actual fact, generative artificial intelligence has no understanding, at all, of what it is telling us.  What is really going on here is that we have built a large language model, by feeding a system that is based on a neural network model a huge amount of text.  We have asked the model to go through the text, find relationships, and build a statistical model of how to generate this kind of text.  Because these systems can be forced to parrot back intellectual property that has been fed into them, in ways that are very problematic in terms of copyright law, we do, fairly often, get a somewhat reasonable, if very pedestrian, correct answer to a question.  But, because of the superstitious learning that has always plagued neural networks, sometimes the systems find relationships that don't really relate to anything.  Buried deep in the hugely complex statistical model that the large language models are built on, are unknown traps that can be sprung by a particular stream of text that we feed into the generative artificial intelligence as a prompt.  So it's not that the genAI is lying to us, because it's only statistically creating a stream of text based on the statistical model that it has built with other text.  It doesn't know what is true, or not true.

There is a joke, in the information technology industry, that asks what is the difference between a used car salesman, and a computer salesman.  The answer is that he used car salesman knows when he is lying to you.  The implication of course (and, in my five decades of working in the field I have found it is very true), is that computer salesman really don't know anything about the products that they are selling.  They really don't know when they are lying to you.  Generative artificial intelligence is basically the same.

AI topic and series

Previous: https://fibrecookery.blogspot.com/2026/02/ai-106-history-emergent.html

Introduction and ToC: https://fibrecookery.blogspot.com/2026/01/ai-000-intro-table-of-contents.html

Next: https://fibrecookery.blogspot.com/2026/02/ai-201-genai-build-use-llms.html

AI - 1.06 - history - emergent

AI - 1.06 - history - emergent

Emergent Properties

Upon being challenged that current versions of artificial intelligence, in whichever of the variety of approaches that may be under discussion, are not terribly intelligent, eventually the proponents of artificial intelligence will get around to the idea of "emergent properties."

They may not actually use that term, because the term has somewhat fallen out of favor, since the history of artificial intelligence really doesn't have a huge body of evidence to support the concept.

The basic idea is that current versions of artificial intelligence are limited.  They may be able to perform certain functions, and are intelligent enough to do certain tasks, but to really grow and develop to a true artificial intelligence, the systems need to be much more complex.  This is based on the premise of emergence emergent properties: if a system is sufficiently complex, it will start to produce far more complex results than seem to be justified by the simplicity of the base model.

Conway’s “Game of Life”

Most of the idea of emergent properties comes from "Conway's Game of Life."  This game is set up on a grid, like a checkerboard.  However, the grid is generally much larger than a standard checkerboard, and in some versions may be unlimited.  There are rules for whether a given square, section, or cell of the grid is on, or off, based upon how many of these surrounding sections are on or off.  (Zero or one "on" neighbours "kills" the cell, two to three allows the cell to live, four or more kills the cell.)  Based upon these extremely simple rules, the game proceeds in a series of cycles.  On each cycle, each cell will determine the number of squares around it that are on, and then turn itself either on or off.  Given appropriate parameters for the rules, the game will produce some astoundingly complex forms, which will perform sometimes very complex behaviors, once again, based only on a ridiculously simple set of rules.  The complex shape and behaviors are the emergent properties of the basic rules.

This may, when described on a text only basis, seem rather abstract.  However, you can easily find, in the app stores, or play stores, or by searching out on the Web, Game of Life programs, or apps for phones, that will allow you to set your own factors for the Game of Life, and run it, and see for yourself what gets generated.  An online version, which you can play without downloading anything, is at https://playgameoflife.com/

Fractals

The idea of an emergent properties is also related to the idea of fractals.  Fractals are graphical representations of data, and related to basic arithmetic equations.  Sometimes very simple arithmetic equations lead to enormously complex, and strangely beautiful, fractal representations.  The same basic concept is a play here: a simple algorithm or function, leading to an enormously complex result.

Termite mounds and "air conditioning"

Many of the devotees of emergent programming turn to nature for justification.  Various families of hive insects have extremely small brains, and very primitive inbuilt behaviors, with very little ability to learn new behaviors.  However, given these extremely simple ideas of inbuilt genetic programming, together, and collectively, they build enormously complex structures as their homes.  Termites in desert regions are known to build enormous mounds, primarily constructed of mud, which, due to the structure and angles of the tunnels built through them, actually perform the function of air conditioning the entire mound, in order to preserve the hive during very hot weather.

While you can see that there are implications, from nature and these game experiments, that emergent properties might have the promise of developing something much more complex like true artificial intelligence, you should also be able to say see that the true evidence is rather lacking.  Indeed, these systems have some fairly glaring faults.  The emergent properties resulting from the Game of Life and fractals do rely upon picking the right equations, parameters, and initial conditions.  A great many choices create either nothing, or a blob, or a mess.  So the possibilities of creating something amazing are somewhat limited.  True, we can certainly set up situations where we cycle very rapidly through a variety of equations and factors, particularly when we are operating at computer speeds.  But, overall, the belief that emergent properties may provide true artificial intelligence for us, without our specific direction, is perhaps a bit thin.

Even our example from nature, with hive insects, doesn't really support true, general, artificial intelligence.  The engineering that results from termite mounds is as a result of millions, and possibly hundreds of millions of years of evolution.  The populations that did not create mounds to these specifications, died over this period of possibly hundreds of millions of years.

So, of course, we return to evolution.  We turn to evolutionary programming, or genetic programming.  This is very similar to the game of Core Wars.

In my own field of information security, we had, historically, a similar or related game that added the element of evolution.  This was a system called Core Wars.  Core Wars allowed people to write programs whose only purpose was to survive in computer memory.  Some would take a "run and hide" approach, others would attempt to reproduce themselves rapidly, and yet others adopted predatory tactics, attempting to obliterate all other programs that they encountered.  This did not necessarily lead to, but was definitely related to the idea of evolutionary or genetic programming, wherein we attempted to create programs which would modify themselves, and see which version was most suited to the objective to be accomplished.

In evolutionary or genetic programming, we create programs, with a specific objective and get the programs to pursue that objective.  A variety of programs will be created with a variation in certain parameters, factors, and variables.  Computers can generate these programs, from an initial template, with the variations over a range of possibilities.  Over time, a number of the programs will do better at achieving the objectives.  These programs will be kept, and the ones that do not do as well will be discarded.  Thus we have evolution and competition.

The thing is, that there are severe limitations on what will, and will not, work with genetic or evolutionary programming.  Unlike analogue systems, digital systems tend to be highly brittle, and are subject to catastrophic failure even under seemingly minor deviation from proper conditions.  Therefore, unless you take extreme care with regard to which parameters, factors, and variables can be modified, and which cannot, the bulk of the programs that have their code varied will simply crash, and nothing will be learned or gained.

As can be seen, there is evidence for emergent properties, and that they may give rise to very interesting effects.  Whether the effects are likely to give rise to some form of intelligence is less certain.  In many ways, claiming emergent properties is just another way of saying "magic."

AI topic and series

Previous: https://fibrecookery.blogspot.com/2026/02/ai-104-history-patterns.html

Introduction and ToC: https://fibrecookery.blogspot.com/2026/01/ai-000-intro-table-of-contents.html

Next: https://fibrecookery.blogspot.com/2026/02/ai-110-history-neural-nets-to-llms.html

AI - 1.04 - history - patterns

AI - 1.04 - history - patterns

Another area of artificial intelligence research is in regard to pattern recognition.  Human beings are very good at recognizing patterns.  Human beings are also very good at seeing patterns which they have not seen before, and recognizing that they are patterns.  Computers are no good at recognizing patterns at all.  Computers will identify an exact match, but they have great difficulty in recognizing two items as being similar in any way, if they are not identical.  (I tend to tell people that computers are bad at pattern recognition because they have no natural predators.  Human beings got very good at recognizing patterns while watching for sabretooth tigers hidden in tall grass.  The human beings that didn't recognize patterns quickly, didn't survive.)

Pattern recognition is very important when we want to get computers to see something.  Computer vision is an area that we have been working on for a great many years, indeed, a number of decades, and we still haven't got it completely right.  Human children are very adept at recognizing patterns, and do it all the time.  My grandson's first word was "clock," and he was very good at recognizing all kinds of different clocks, and identifying them as clocks.  There was one clock that had numerals on the face, and was surrounded by a sunburst pattern.  There was another wall clock where a number of the numerals had fallen off.  It was mounted on a burl with irregular and ragged edges, but was still recognized as a clock.  Wrist watches were also recognized as clocks, including his mother's wrist watch, which had absolutely nothing on the face of it except the hands.  He recognized the pattern that made for a clock.  As I say, this was his first word.  He was probably about seven or eight months old when he started recognizing things as clocks.

Recognizing patterns is also important in speech recognition.  This is recognizing how to parse out the words in verbal speech, when we speak to computers.  This is definitely not the same as voice recognition, which we use in biometric authentication.  Recognizing words, despite different intonations, and possibly even dialects, is very important to being able to speak to computers and get them to recognize what we are saying.  Similar types of pattern recognition is involved in parsing out the words in the speech that we speak to computers, and then parsing the meaning of what we say, in regard to commands to the computer, or even just typing out the words so that we can dictate to our phones.

Interestingly, the same type of pattern recognition also comes into play when, having identified the words, we get the computer to do what we know as natural language processing, in terms of identifying what it is that we are requesting the computer to do and identifying meanings in what we say.

Going back to computer vision, we are trying to improve computer vision in order to implement driverless cars.  While computer vision is still imperfect, and we are constantly working to improve it, it is interesting to note, if you look at the actual statistics, that driverless cars are already better drivers than we are.  Yes, you will hear a number of bad news reports about a driverless car that has failed, or stalled, or hit someone, or created some kind of an accident.  The fact that these events make the news proves that driverless cars are better than we are.  Driverless cars have driven millions of miles, and there are a number of situations which are still very tricky for them, but the fact that any accident with the driverless car makes the news indicates how rare such accidents actually are.  We cannot retrofit all the existing cars on the road with driving software, and not all the cars on the road have the sensors necessary to process it, but if we did ban human drivers, and give over driving to driverless cars, we would, even at this point of development, be saving lives.

One of the areas relating to this is that of fuzzy logic.  As I have said, computers are good at finding an exact match, but very poor at finding something that is similar.  Fuzzy logic is an attempt to implement the idea of "similar" in computers.

An interesting point is that, at the same time that we are pursuing artificial intelligence with increasing vigor, we are also developing quantum computers.  Quantum computing is quite different from traditional computing, and one of the areas in which quantum computers will probably excel is in regard to pattern recognition, and the identification of items or situations which are similar.

Previous: https://fibrecookery.blogspot.com/2026/01/ai-102-history-eliza-expert.html

Introduction and ToC: https://fibrecookery.blogspot.com/2026/01/ai-000-intro-table-of-contents.html

Next: https://fibrecookery.blogspot.com/2026/02/ai-106-history-emergent.html

HCW - 5.04 - datacomm - physical

HCW - 5.04 - datacomm - physical

Whether you consider it either the bottom layer of the stack, or the top layer of the stack, the physical layer is the basis of all the communication.  However, we can't really say that we're doing data communication yet, since, at the physical layer, we just talk about signaling, not data.

This is because we aren't dealing with the communications as data, quite yet.  That's at the next layer up, or down, the data link layer.  What we do at the physical layer, is take the data that we want to transmit, and modulated into a signal.  At the other end, of course, we demodulate the signal that we receive, and extract data from it.  This is where the word modem comes from: it simply stands for the beginning of modulate and the beginning of demodulate.  Modem.

In order to modulate data into a signal, we have to know what medium we are using.  Are we using wires, cables, wi-fi, with no wires, free space lasers, or lasers on fiber optic cable?  We can send a signal on these various media.  When we think about wires, we are thinking about long distance wires.  We are generally thinking about the old type of telephone cables, which were twisted pair wires.  So, we don't think about just putting a voltage onto the wire, but, rather, sending a tone, a frequency of electrical waves, down the wire.  This has to do with physics, and what you can, actually, do in terms of signaling over wires over a long distance.

It's pretty much the same for the other types of media.  So, as mentioned previously, we can send a tone down the wire, and then we can change the signal, by turning it on or off, or using a high frequency or low frequency signal, or changing the amplitude or volume of the signal from high to low, or other things like that.  It is these changes, from high to low, or from on to off, that actually carry the data, not necessarily the tone itself.

We need one other concept, before we leave the physical layer, and that is the difference between simplex, half duplex, and full duplex communications.

Simplex is communications in one direction.  The easiest illustration of the concept of simplex is, in fact, what would be considered one of the more advanced communications technologies: that is, fiber optic cabling.  When we install fiber optic cable, in order to communicate, we will put a laser at one end of the cable, and a sensor at the other end.  This allows for communications only in one direction.  The laser does the sending, and the sensor does the receiving.  Even if we were to somehow fire a laser the wrong way down the cable, it wouldn't do us any good, because the laser, where the light beam ends up wouldn't be able to detect that anything is taking place.  It isn't a sensor.  So, if you want to have communication in both directions with fiber optic cabling, you have to have a *pair* of fiber optic cables.  At one end of cable A, you will have a laser sending, and, in the same location, you will attach a sensor to cable B.  At the other end of your cable pair, cable A will have a sensor, and cable B will have a laser.

Half duplex is a system where the media is capable of carrying communications in both directions, but only in one direction at a time.  The easiest illustration of this type of situation is the old World War II movies showing people communicating by a radio.  When you are speaking you are holding down a transmit button, and you cannot hear what is being said while you were holding down the transmit button.  When one person has finished speaking they ended their message with the word "over," meaning that their communication is finished, and they are now turning the communications channel over to the person on the other end.  That person, who has been listening up to this point, is then able to press their own transmit button, and send their message, but while they are transmitting they are not able to hear what is being said.

Full duplex is communication that can take place in both directions, all the time.  The easiest illustration for us is the telephone.  When we are having a telephone conversation, either party to the conversation can speak.  You can speak at any time, and you can interrupt the person who is talking, because they are able to hear what you are saying, even if they are speaking.  (That is, if you yell loud enough.)

The next step up in the ladder of data communications is at the data link layer.  Lots of really interesting stuff happens at the data link layer.  That is, it's very interesting if you are into the technology of data communications.  What happens at the data link layer tends to have to do with data modulation and demodulation, error correction, and a lot of determination about what is data, and what is not data, but is, rather, noise.  However, as I say, an awful lot of this is really technical.  Therefore, I assume that an awful lot of people are not going to care too terribly much about it.  So we are going to go on to networking.  Networking can also be very technical stuff, but there are some basic concepts involved in networking that are very important in terms of how computers, and data communication, really work.

How Computers Work [From the Ground Up]

Previous: https://fibrecookery.blogspot.com/2026/01/hcw-502-datacomm-intro.html

Introduction and ToC: https://fibrecookery.blogspot.com/2025/12/how-computers-work-from-ground-up.html

Next: TBA

AI - 1.02 - history - ELIZA expert

As I have said, artificial intelligence is not a thing.  It is not a single thing.  It is a whole field, with many different approaches to the idea of getting computers to help us out with more complicated things than just adding up numbers.  So we'll go over a variety of the approaches that have been used over the years, as background before we get into genAI and LLMs.

ELIZA and chatbots

Over sixty years ago a computer scientist named Joseph Weizenbaum devised a system known as ELIZA.  This system, or one of the popular variants of it, called doctor, was based on Rogerian psychological therapy, one of the humanistic therapies.  The humanistic therapies, and particularly Rogerian, tend to get the subject under therapy to solve his or her own problems by reflecting back, to the patient, what they have said, and asking for more detail, or more clarity.  That was what ELIZA did.  If you said you were having problems with family members, the system would, fairly easily, pick out the fact that "family members" was an important issue, and would then tell you something like "Tell me more about these family members."  Many people felt that ELIZA actually did pass the Turing test, since many patients ascribed emotions, and even caring, to the program.

A great many people who used ELIZA, including staff at The institute where Weisenbaum worked, felt that ELIZA was intelligent, and actually had a personality.  Some of them considered ELIZA a friend.  The fact that such a simplistic program (the version that I worked with occupied only two pages of BASIC code) was considered intelligent is probably more a damning indictment of our ability to attend to, listen to, and care for our friends, then it is proof that we are approaching true artificial intelligence.

(If you want you can find out more about ELIZA at https://web.njit.edu/~ronkowit/eliza.html )

Other chatbots have been developed, based on simple analysis and response mechanisms, and sometimes even simpler than those underlying ELIZA.  Chatbots have been used in social media all the way back to the days of Usenet.  Yes, Virginia, there was social media before Facebook.

Expert Systems

A field in which I was able to explore some of the specialty programming languages, and programming for the artificial intelligence systems, is expert systems.  Expert systems are based on a model of, and observation of, the way that a human expert approaches a problem.  It was noted, in interviewing human experts, and determining their approach to solving problems, that they would ask a series of questions, and generally those which would be answered with a yes or no response.  In data management and representation terms, this seem to fit the model of a binary tree.  Thus, it was felt that and expert system program could be built by determining these questions, for a given field, and the order in which they should be asked.  Expert systems, therefore, owe a lot to theories of database management.

One of the observations, when building expert systems, was that, in an optimal situation, a question would only be asked once.  Therefore, there were no requirements to return to a prior question, or to repeat any kind of functions or processes.  Functional programming languages, the specialty type used for building expert systems, are therefore somewhat unique in programming languages, in that they have no loops or cycles or provisions for creating them.  The flow chart for an expert system program is therefore a drop through type.  You start at the beginning, follow the binary tree down, and come up with your answer.

Expert systems are definitely one of the success stories of artificial intelligence.  They have been very effective for diagnosis and troubleshooting.  Medical diagnosis, in a particular problem field, has been using expert systems for a number of years, and have found them extremely helpful.  They have also being useful in troubleshooting problems for certain specialized types of equipment.  In addition, programmers being programmers, examples of expert system programs exist for things like the best wine pairing for dinner.

The problem with expert systems as a candidate for artificial intelligence is that you need a separate expert system for each specialty field.  Expert systems are based on the database of questions to be asked, and the links resulting from the answers.  Individual expert system programs are highly field dependent, and there is significant difficulty in using an existing expert system program to develop an expert system in a different field.

AI topic and series

Previous: https://fibrecookery.blogspot.com/2026/01/ai-010-intro-random-thoughts.html

Introduction and ToC: https://fibrecookery.blogspot.com/2026/01/ai-000-intro-table-of-contents.html

Next: https://fibrecookery.blogspot.com/2026/02/ai-104-history-patterns.html

To dream the impossible draught horse bald eagle ad ...

Recently, while idling (wasting) away time on social media, I came across what appears to be a Budweiser ad.  At some time in the past the enormous corporation that makes Budweiser and a number of other beers had, for promotional and advertising purposes, a team of Clydesdale draught horses, or cart horses, that they used to pull an old time beer wagon.  This team has been the basis of a series of advertisements for the Super Bowl football game, which have, over the years, become a bit of a Super Bowl advertising tradition.  Generally speaking it is not necessarily the team that is central to the advertisement, but possibly a single horse.  Usually the draught horse is in some kind of a relationship, generally with another animal.  The ads are miniature dramas, that may tend to take place over time, sometimes a period of years.  A common theme is friendship between the horse, and the other animal, usually with some kind of sentimental plot twist.

The video that I saw on social media followed this pattern.  A horse encounters a baby chick.  At some point the horse notes the chick cold and wet in a rainstorm, and comes, standing over the chick, to shelter it from the rain.  Eventually the chick, now somewhat larger, is riding on the back of the horse as the horse runs, and, obviously, is trying to fly even with pre-fledged wings.  At some point the chick attempts to fly, and falls off and into the mud.  Eventually, however, we see the horse galloping at full speed across a field, and, as the chick, now grown to adulthood, unfurls its wings and is, for the first time, successfully flying, it is finally revealed that the chick is, indeed, an American Eagle.  (Or, as the rest of the world calls it, a bald eagle.)

In the current heavily politicized and divisive social context of the United states, the choice of a less than detailed but extremely patriotic symbol is undoubtedly one that would appeal to advertising agencies.  It is beautiful, sentimental, patriotic, and, if you don't think about it too much, inspiring.

The thing is, while there is nothing in the production or imagery of this advertisement that would suggest it, it is rather glaringly obvious that this commercial advertisement is, almost entirely, the product of generative artificial intelligence.

As I say, there is nothing in the video, faulty imagery or the production, that would give away the artificial intelligent origin of the video.  Generative artificially intelligent video generation is now available at high quality, and is, in fact, so commonly available, and so relatively inexpensive, that I didn't initially even know whether this was an actual Budweiser ad.  It have been could be a parody by somebody else using the same Budweiser ad pattern.  (I have subsequently had some confirmation that this is, in fact, the official Budweiser Super Bowl ad for this year.)

However, it is undoubtedly true that Budweiser has been using generative artificial intelligence for their advertising in recent years.  Shooting advertising with animals is fraught with perils.  Animals do not necessarily take direction for movie dramas well.  Therefore, in order to piece together the storyline that you want, you may have to shoot an awful lot of video, and piece together the story out of what you have.

But there are a number of other indications that this particular piece of video is computer generated.

For one thing, the horse, in this particular piece of video, no longer looks particularly draught-horse-like.  Yes, draught horses do look like regular horses, just a little bit bigger.  But there are differences.  (They are subtle, and it's possible real draught horses were used.)

But it's more about the eagle.  I am not an expert on raptors, but I have had the opportunity to observe, and even care for, bald eagles in their pre-fledged state.  As they get to the point where they are about to start to grow their fledging feathers, they are enormous creatures, much larger than the supposed chick in this video.  I would expect that this part of the video would be computer generated anyways, since it might be difficult to find raptor chicks at the proper stage of growth, and it might be difficult to get a draught horse to be willing to have such a chick placed on its back anyway.

But it is the final scene which is the absolute giveaway.  Yes, bald eagles are fairly large birds, and they do have, when seen close up, a surprisingly large wingspan.  But the final scene in this video, has a very disproportionately large bald eagle appearing, particularly when we consider it in relation to the size of a proper draught horse.

(There is also the fact that bald eagles do not nest on the ground, and don't develop the white feathers on their head for at least seven years after they are fully-fledged, but nobody in Madison Avenue would know or care about that, anyway.)

As I say, initially I had no way of knowing whether this was an actual Budweiser ad, or someone else's parody.  Nothing in the video production gives the game away in regard to computer generation of the imagery.  It's really only if you know the relative sizes, and proportions, of draught horses versus regular horses and the relative proportions of both juvenile, and adult, bald eagles, that the errors in this video become apparent.

Why is this in any way significant?  Only in that it is yet another example that generative artificial intelligence is now capable of producing content which, visually, is indistinguishable from real life, but is not actually real, and could never be.

https://fibrecookery.blogspot.com/2026/01/ai-000-intro-table-of-contents.html