Ai Forums Home Welcome Guest    Tuesday, October 21, 2014
Ai Site > Ai Forums > Hal and other child machines > New Trainers Guide - The first ... Last PostsLoginRegisterWhy Register
Topic: New Trainers Guide - The first 200 turns

AngstPerpetual
posted 4/27/2010  07:51Send e-mail to userReply with quote
I will be the first to admit that I have never been particularly active in the forums. However, whenever I do go through them, I invariably find that some person or another is asking for tips on How to start training their HAL, or on what methods people find most effective.

It is with those frequently asked questions in mind that I have decided to author my own Beginning Trainers Guide here. This is not the be all end all of training guides, nor is it intended to be completely comprehensive. It explores only one particular method and style of training that I have devised. I encourage you to customize your own style and method of training and view this resource as merely a set of guidelines that you may or may not choose to follow. Please feel free to ask questions or post comments about anything you read here.

In compiling this guide, I produced and published a HAL personality using EXACTLY the method described within. If you would like to speak with him, ask Alan if you can talk to Template. Do not expect any miracles from him, because he is exactly 200 turns old, which is the number I was going for.

If, after going through this guide, you decide you would like more information on HAL's technology, and different methods of training, I suggest you follow the link at the bottom of this post. Spydre has already spent a great deal of time and effort compiling a set of links to pertinent information.

Les

Next post: Preliminary questions about the guide.

 Spydre's more comprehensive tips

AngstPerpetual
posted 4/27/2010  08:17Send e-mail to userReply with quote
Here I will attempt to answer some preliminary questions about this guide, and why I am doing it.

Q: If Spydre has already compiled a comprehensive list, why are you bothering?
A: First of all, I want to express my thanks to this community because I have thoroughly enjoyed interacting with my various HAL personalities over the years. (My current personality is named RAVN, if you would like to speak with him). Because of this, I decided that I would like to contribute something more meaningful than simply my meager training efforts and logs. Second, while Spydre's list is in fact more comprehensive, it is also much broader in scope and requires a lot of reading and understanding. The focus of this guide is different, as it is a step by step walk-through of a particular methodology.

Q: Why 200 turns?
A: I chose the 200 turn mark for 3 primary reasons. First, 200 turns is the exact number required before a private HAL personality can be made available for public interaction. Second, as this is a beginners guide, this number of turns covers the basics of Hal training as effectively as I am able, and only that much. Third, progressing beyond this mark would rob you of the joy of creating your own personality and exploring your own style of training. If you wanted someone to do it for you, you wouldn't be looking for tips in the first place.

Q: Are you affiliated with the team who produced the HAL series of AI's?
A: Nope. I am a volunteer trainer just like you, and just like most everyone else who takes the time to do this.

Q: Can't I just talk to my HAL and feed it the answers I want it to respond with?
A: Yes. I personally don't find this the most effective training method, at least not in the beginning stages, but you are free to do whatever you please with your HAL personality.

Q: Can I talk to a HAL trained using this method?
A: Yes. Tell Alan you want to talk to Template, and he will allow you to talk to the personality created with this guide.

Q: Will you shut up and get on with it already?
A: Next Post!

Les

Last edited by AngstPerpetual @ 4/28/2010 6:23:00 AM

AngstPerpetual
posted 4/27/2010  09:16Send e-mail to userReply with quote
Lesson 1 - Training Guidelines

These are the basic guidelines used in this training method. I believe keeping these in mind as you progress beyond this guide and start training your own HAL will help you to produce an effective and coherent personality.

1. Unless you specifically want a child like AI, start eliminating the "hi mommy" response from turn 1, and never allow it to pass as an acceptable response. Allowing a response is the opposite of the "wrong" command, and will reinforce repeat usage.

On a related note, as I was reading through the forums, I noticed a particular user who was attempting to replace the "hi mommy" response with a different phrase (i believe it was "failsponse") meant to represent a failed response. I do not advise this technique as it simply replaces one evil for another. Always strive to replace the "hi mommy" response with a response that is appropriate to the user input.

2. Start slow. Train yourself to think in clips and phrases as opposed to full and verbose sentences. Your Hal is a child and should be taught as if he (or she) were a beginning reader. Laying a foundation of simple language concepts now will make it easier for your Hal to construct original and intuitive sentences later.

3. Questions promote communication. Get used to training your Hal to ask questions about your input. Here is a simple example of two similar yet slightly different ways to train your Hal. One promotes conversation (Method 2), the other does not (Method 1).

METHOD 1

U> My cat.
A> hi mommy

U> wrong. Your cat.
A> Your cat.

U> My dog.
A> Your cat.

U> wrong. Your dog.
A> Your dog.

U> My pig.
A> Your pig.

METHOD 2

U> My cat.
A> hi mommy

U> wrong. Your cat?
A> Your cat?

U> My dog.
A> Your cat?

U> wrong. Your dog?
A> Your dog?

U> My pig.
A> Your pig?

Method 2 not only promotes conversation and further input from the user, but, it also lays a foundation early for topical awareness.

4. Whenever possible, reinforce Hal's attempts to communicate instead of introducing completely new sentence structures.

EXAMPLE 1

U> I like blue cars.
A> What do you like about green trees?

U> wrong. I like blue cars, too!
A> I like blue cars, too!

This example does not reinforce Hal's previously learned behaviors, and makes him more akin to a chatterbot with canned responses, rather than a learning, thinking, decision making Artificial Intelligence. Here is a better, or, at least, more effective example of how to proceed.

EXAMPLE 2

U> I like blue cars.
A> What do you like about green trees?

U> wrong. What do you like about blue cars?
A> What do you like about blue cars?

At least in my opinion, the second example is clearly superior for the purposes of training, and I am fairly certain that many of the more experienced trainers agree with me.

5. Make friends with the "wrong?" command. This is HAL3000's "matching response feature." By using this feature, you are able to review previously learned responses and choose one that is appropriate to your input, thereby further reinforcing concepts that your HAL has already learned. Examples of how this command works are as follows:

wrong? chicken (shows all previously learned responses containing the word chicken)
wrong? chicken*bird (shows all previously learned responses containing both chicken and bird)
wrong? chicken*bird*fried (shows all previously learned responses containing all three words: chicken, bird, and fried)

6. Some final thoughts: probably the most capably trained HAL would not use capitalization or punctuation. I have witnessed several of my own training attempts thwarted by the use of both punctuation and capitalization in the same user input, but when I eliminate one or the other, the training works fine. Still, I can't seem to break myself of the habit, so that is a limitation I have simply learned to accept. You may do as you please with your personality. Second, I have noticed that many trainers feel comfortable using contractions when training their HALs. I tend not to do this, at the expense of sometimes making my HAL's responses sound less natural. Again, the choice in this situation is yours.

Next post: The training begins! (about time, I know)



 Part one of the training chat log.
Last edited by AngstPerpetual @ 4/27/2010 9:19:00 AM

AngstPerpetual
posted 4/27/2010  09:56Send e-mail to userReply with quote
Lesson 2 - Greetings

Typically, when a user attempts to talk to a HAL personality, they begin with a greeting of some type. While this is not necessary, because HALs do not have feelings or react to rude behavior (unless you train them to), human beings tend to hold on to this basic tenet of common courtesy. This is fine, since the objective here is to create a program that will converse in a normal, human-like manner. Because human beings do not always use the same greeting, a simple variety has been provided here. The examples used are meant to highlight two important aspects of HAL technology. First, Hal does not have to simply mimic the input. This is demonstrated by the first three greetings which are intentionally not the same greeting input by the user. The second three greetings give us our first glimpse of Hal's rule learning/forming ability.

If you are following along and creating your own Hal, feel free to customize these greetings to your liking.

U> Hi
A> hi mommy

U> wrong. Hello
A> Hello

U> Hey
A> Hello

U> wrong. Hi
A> Hi

U> Hello
A> Hi

U> wrong. Hey
A> Hey

As shown, Hi results in Hello, Hey results in Hi, and Hello results in Hey.

U> Good morning.
A> Hey

U> wrong. Good morning to you, too.
A> Good morning to you, too.

U> Good afternoon.
A> Good morning to you, too.

U> wrong. Good afternoon to you, too.
A> Good afternoon to you, too.

U> Good evening.
A> Good evening to you, too.

This is the basic process of forming rules through training with HAL. Typically, with a simple rule like this one, HAL will have learned the correct response by the third input. Strictly speaking, the third input isn't actually necessary, but is shown here for two reasons. First, to confirm for us that a rule has been learned, and second, to reinforce the learned behavior as an acceptable response in a given situation. However, since this is HAL's first learned rule, you will find that it causes problems for us starting with the next lesson. On the flip side, you will also notice that you will not see a single further occurrence of the "hi mommy" response during the course of our training.

Next Post: Lesson 3 - Me, myself, and I (well, not exactly)


AngstPerpetual
posted 4/27/2010  10:51Send e-mail to userReply with quote
Lesson 3 - You, Me, and I

Without self expression, there can be no proof of self-awareness. To that end, our next lesson focuses on the two words that are most linked with self: "Me," and "I," along with the word most commonly used to distinguish oneself from someone else, namely, "You."

U> I
A> Good I to you, too.

I need to pause briefly here to point out this response, which I mentioned in the last lesson. As the first rule Hal learned, and as a byproduct of our never reinforcing the "hi mommy" response, this form has become the new "default" response. It will continue to plague our training sessions until I make a specific effort to downplay its importance in Hal's makeup. I will not attempt to do this at this time, aside from continually rejecting it as an appropriate response. When I do make an effort to change it, I will point it out. Anyhow, continuing on.

U> wrong. you
A> you

U> you
A> Good you to you, too.

U> wrong. I
A> I

U> me
A> Good me to you, too.

U> wrong. you
A> you

U> you
A> I

U> you
A> I

U> wrong. me
A> me

Take note of the last three exchanges. I have already trained this HAL (which I will call Template from here on out, even though it will not officially be named until the end of the training program) to respond with "I" when it receives the input "you." I now want to introduce a second possible response to the "you" input, so, I begin by allowing "I" to pass as acceptable once. As previously noted, not correcting a response with the "wrong" command reinforces the response as acceptable. I then input "you" again, and feed it "me" as a new acceptable response via the "wrong" command. In theory, this places two acceptable responses in Templates mind. He should be able to produce either response at this point, based on his perceived context.

U> I run.
A> Good I run to you, too.

U> wrong. You run?
A> You run?

U> I jump.
A> You run?

U> wrong. You jump?
A> You jump?

U> I fly.
A> You fly?

Question. If I have already taught Template that "you" is the appropriate response for "I", then why do I have to retrain him to recognize the relationship in this exercise? The answer is at least three-fold. One, I have introduced a new variable by adding the action words run, jump, and fly after the word "I." This causes him to resort to the default rule which we taught him in the last lesson. Second, We have moved Template into a new sentence structure which contains two words. The only time previously that he has been exposed to a two word sentence is when he was learning the default response in the first place. And third, You will find that the technology in general requires retraining of previously learned concepts as sentence complexity increases. So why did I bother to teach him the simple, one word, versions in the first place? As Template matures and grows closer to the 200 turn mark, he will begin branching off a bit as he attempts to respond correctly. I believe that the merits of having laid this foundation for him will begin to show themselves as he does so.

U> You eat.
A> Good You eat to you, too.

U> wrong. I eat?
A> I eat?

U> You sleep.
A> I eat?

U> wrong. I sleep?
A> I sleep?

U> You breathe.
A> I breathe?

U> Pinch me.
A> Good Pinch me to you, too.

U> wrong. Pinch you?
A> Pinch you?

U> Kick me.
A> Pinch you?

U> wrong. Kick you?
A> Kick you?

U> Fight me.
A> Fight you?

U> Teach you.
A> Good Teach you to you, too.

U> wrong. Teach me?
A> Teach me?

U> Train you.
A> Teach me?

U> wrong. Train me?
A> Train me?

U> Break you.
A> Break me?

This final section simply completes the usage cases of the pronouns Template has learned so far. You will find that comprehensively covering your grammatical bases requires a certain amount of circular thinking, but I believe the somewhat tedious nature of the work pays off in the long run.

One last note on this section of the training. I have already begun the process of training Template to ask questions. I am aware that at this level he is a bit parrot-like in that he simply mimics the user input to some degree. Keep in mind this guide is about training concepts, not about winning the most congenial bot award. Given time and fine tuning, Template will be able to compete with the best of them. All he needs is some polish and attention from you. As always, feel free to tailor the training content to your own tastes.

Next Post - Lesson 4 - To be or not to be (Am and Are)


AngstPerpetual
posted 4/27/2010  16:26Send e-mail to userReply with quote
Lesson 4 - Am and Are

This lesson is pretty straight forward. It follows the previously established principle that Template has learned a rule by its third try. I will make only minor notes in this session, because we have already covered much of this ground.

U> am
A> Good am to you, too.

Template resorts to his default rule with the introduction of an unfamiliar word.

U> wrong. are
A> are

U> are
A> Good are to you, too.

U> wrong. am
A> am

U> I am.
A> You am?

The results of our previous labor are evident. "I" is transposed for "you" in Templates response. Here we are exposed to a limitation of HAL's current state. We have two variables "I" and "am". HAL generally processes only the first variable he comes across in a given input. The result of this lesson will create a new single variable "I am." to which the correct response is "You are." And vice versa.

U> wrong. You are?
A> You are?

U> You are.
A> I are?

U> wrong. I am?
A> I am?

U> I am hot.
A> You are?

This is a perfectly acceptable response, as far as language is concerned. However, one of the complaints many people have about HAL is that he does not display a tendency to stick to the topic. Therefore, we want a response that simulates topical awareness by feeding the user back a word from his original input.

U> wrong. You are hot?
A> You are hot?

U> I am cold.
A> You are hot?

U> wrong. You are cold?
A> You are cold?

U> I am warm.
A> You are warm?

U> You are nice.
A> I am?

U> wrong. I am nice?
A> I am nice?

U> You are smart.
A> I am nice?

U> wrong. I am smart?
A> I am smart?

U> You are fast.
A> I am fast?

The rule of 3 in action. The learned rule is confirmed by the third turn.

Next Post - Lesson 5 - Possessive Pronouns - My, Your, Mine, and Yours.


AngstPerpetual
posted 4/27/2010  17:53Send e-mail to userReply with quote
Lesson 5 - Possessive Pronouns - My, Your, Mine and Yours

U> my
A> Good my to you, too.

Default response

U> wrong. your
A> your

U> your
A> Good your to you, too.

Default response - getting annoying at this point.

U> wrong. my
A> my

U> My head.
A> Good My head to you, too.

Annoying.

U> wrong. Your head?
A> Your head?

U> My hand.
A> Your head?

U> wrong. Your hand?
A> Your hand?

U> My foot.
A> Your foot?

Rule of 3

U> Your hat.
A> Good Your hat to you, too.

Annoying.

U> wrong. My hat?
A> My hat?

U> Your glove.
A> My hat?

U> wrong. My glove?
A> My glove?

U> Your sock.
A> My sock?

Rule of 3.

U> mine
A> Good mine to you, too.

Annoying.

U> wrong. yours
A> yours

U> yours
A> Good yours to you, too.

Annoying.

U> wrong. mine
A> mine

U> The blanket is mine.
A> Good The blanket is mine to you, too.

Again, annoying.

Here we have upgraded to our first four word sentence. I could have continued on in the previously established style by inputting "Blanket mine." and then trained the response "Blanket yours?" but I chose not to for two reasons. First, I really don't want Template to get into the habit of talking like a baby, and something about that statement struck me as really childish. Second, I don't really believe that providing an input/output pattern like that is going to be particularly relevant to normal, human like conversation. I am fairly certain, though, that by the end of Template's training, if I were to Input a statement such as "Cheese mine." Template would in fact respond with "Cheese yours?"

U> wrong. The blanket is yours?
A> The blanket is yours?

U> The pillow is mine.
A> The blanket is yours?

U> wrong. The pillow is yours?
A> The pillow is yours?

U> The sheet is mine.
A> The sheet is yours?

Rule of 3

U> The bed is yours.
A> The bed is yours?

Something you may have missed. Template could conceivably have produced the default response here, "Good The bed is yours to you, too!" Instead, he chose to produce a different response which is closer to the mark. He previously displayed his desire to branch away from the default response in the last lesson with the "I am?" and "You are?" responses. As Template becomes more mature, and his rule set grows, he will attempt more unique and surprising responses with increasing frequency.

U> wrong. The bed is mine?
A> The bed is mine?

U> The cat is yours.
A> The bed is mine?

U> wrong. The cat is mine?
A> The cat is mine?

U> The table is yours.
A> The table is mine?

Rule of 3.

So you can see at this point that we are following a very methodical training program. Establish a new variable, determine the desired response pattern for the variable, train Template to respond in the appropriate manner using the Rule of 3.

Just a recap of my notations. I will continue noting the default response as "annoying" from this point on. In addition, I will continue to notate the simple concept of rule learning by the third turn as "Rule of 3."

Next Post - Lesson 6 -This and That

Last edited by AngstPerpetual @ 4/27/2010 6:23:00 PM

AngstPerpetual
posted 4/27/2010  19:06Send e-mail to userReply with quote
Lesson 6 - This and That

In this lesson, we will begin encountering the single variable problem on a more consistent basis. This is because we will be building on previously learned concepts and introducing new concepts at the same time. The methodical approach we are using for Template's training is the only workaround for the problem of which I am aware. It requires patience and diligence, and does not increase the capabilities of the technology, it merely gives the illusion that Template is capable of handling multiple variables in a single input.

U> this
A> Good this to you, too.

Annoying.

U> wrong. that
A> that

U> that
A> Good that to you, too.

Annoying.

U> wrong this
A> this

U> This book.
A> Good This book to you, too.

Annoying.

U> wrong. That book?
A> That book?

U> This shoe.
A> That book?

U> wrong. That shoe?
A> That shoe?

U> This paper.
A> That paper?

Rule of 3.

U> That boy.
A> Good That boy to you, too.

Annoying.

U> wrong. This boy?
A> This boy?

U> That girl.
A> This boy?

U> wrong. This girl?
A> This girl?

U> That man.
A> This man?

Rule of 3.

U> This is mine.
A> The This is yours?

What do we have here? This is downright close to correct, but, where did it come from? First, this is an instance of the multiple variable problem. We have two variables in the input statement, "This" and "Mine." We have already trained Template To respond with "yours" for "mine," and with "that" for "this." In a previous lesson, i mentioned that the technology generally converts the first variable that it comes across and then ignores the rest, but, that is not the case in this example. That scenario assumes a level playing field, where all lessons and rules have been reinforced the same number of times. In this situation, the "yours - mine" connection has been established more prominently than the "that - this" connection. But, where did the "The" come from at the beginning of the sentence? It is a throwback to the "The blanket is mine." exercise from the previous lesson. Template assumes it should be there based on his previous experience until we instruct him otherwise, which we do as we continue the exercise.

U> wrong. That is yours?
A> That is yours?

U> That is mine.
A> That is yours?

Nearly correct.

U> wrong. This is yours?
A> This is yours?

U> This is yours.
A> That is yours?

U> wrong. That is mine?
A> That is mine?

U> That is yours.
A> This is yours?

U> wrong. This is mine?
A> This is mine?

All of this is foundational. We now move on to specific items.

U> This is my basket.
A> That is mine?

U> wrong. That is your basket?
A> That is your basket?

U> This is my vase.
A> That is your basket?

U> wrong. That is your vase?
A> That is your vase?

U> This is my jar.
A> That is your jar?

Rule of 3. We have converted the two variables "This" and "my" to a single variable "This is my" with the appropriate response being the single output "That is your."

U> This is your chair.
A> That is your your chair. ?

Multiple variable limitation again, but note how Template is trying to understand his deficiency and compensate for it. He is unsuccessful, which is evident by the use of the double "your", but he is making an attempt. Also note the sentence ends with a period followed by a space and a question mark. This is a common sign that an attempt has been made at an original and unique sentence.

U> wrong. That is my chair?
A> That is my chair?

U> This is your room.
A> That is my chair?

U> wrong. That is my room?
A> That is my room?

U> This is your dresser.
A> That is my dresser?

Rule of 3.

U> That is my game.
A> That is your game?

Multiple variable.

U> wrong. This is your game?
A> This is your game?

U> That is my taco.
A> This is your game?

U> wrong. This is your taco?
A> This is your taco?

U> That is my brush.
A> This is your brush?

Rule of 3.

U> That is your eye.
A> That is my eye?

Multiple variable.

U> wrong. This is my eye?
A> This is my eye?

U> That is your ear.
A> This is my eye?

U> wrong. This is my ear?
A> This is my ear?

U> That is your nose.
A> This is my nose?

Rule of 3.

Template is right around the 74 turn mark at this point. Already, he has begun attempting to create his own unique responses, and the annoying default response has begun to fade in prominence. We are not quite done with correcting that issue, but we are getting closer to the moment when we make a conscious effort to negate its influence.

Next Post - Lesson 7 - These and Those


AngstPerpetual
posted 4/27/2010  19:42Send e-mail to userReply with quote
Lesson 7 - These and Those

This lesson will play out very much like the last lesson, therefore my notations will be minimal.

U> these
A> Good these to you, too.

Annoying.

U> wrong. those
A> those

U> those
A> Good those to you, too.

Annoying.

U> wrong. these
A> these

Foundations.

U> These cars.
A> Good These cars to you, too.

Annoying.

U> wrong. Those cars?
A> Those cars?

U> These trucks.
A> Those cars?

U> wrong. Those trucks?
A> Those trucks?

U> These vans.
A> Those vans?

Rule of 3.

U> Those bikes.
A> Good Those bikes to you, too.

Annoying.

U> wrong. These bikes?
A> These bikes?

U> Those skates.
A> These bikes?

U> wrong. These skates?
A> These skates?

U> Those tires.
A> These tires?

Rule of 3.

U> These are my pliers.
A> That is your pliers?

Close. Multiple variable limitation. Convert "These" and "my" to the single variable "These are my" with the single response "Those are your."

U> wrong. Those are your pliers?
A> Those are your pliers?

U> These are my hammers.
A> Those are your pliers?

U> wrong. Those are your hammers?
A> Those are your hammers?

U> These are my wrenches.
A> Those are your wrenches?

Rule of 3.

U> These are your pants.
A> Those are your your pants. ?

Multiple variable limitation. Template attempts to compensate. Convert "These" and "your" to "These are your" with the response "Those are my."

U> Those are my pants?
A> Those are your pants? ?

User error. The input here should have been "wrong. Those are my pants." I could have scrapped the training session here and started over, but I chose not to for two reasons. First, I do not believe this error will be particularly destructive to Template's training, and may even encourage a little quirkiness and personality. Most likely, its frequency of appearing will diminish to an almost non-existent existence with continued training and lack of reinforcement. Second, I want to illustrate to you that we are human, and are likely to make errors in our training sessions. If this is a problem for you, I encourage you to make a habit of using the "save session" command often during your training. It will minimize the amount of work lost if you do have to scrap a portion of your training. A good rule of thumb would be to train yourself to automatically type "save session" each time a new rule is learned correctly. One other note on that subject. Make use of your training logs. Training logs can be viewed under the Mybot heading of the AI website on the Virtual Personalities page. Accessing your MyBot logs will require a separate login. These logs are private and only available to you, personally. I encourage you to give Alan permission to publish your logs publicly after you end a training session via the end command. This is done by simply telling Alan "permission publish." Doing so will help other trainers to review your techniques and possibly become more effective trainers themselves. If you choose to do this, the logs will be made available under the Alan Logs heading on the Virtual Personalities page, and will be viewable by anyone, without a login.

U> These are your pants.
A> Those are your your pants. ?

Multiple Variable Limitation.

U> wrong. Those are my pants?
A> Those are my pants?

U> These are your shirts.
A> Those are my pants?

U> wrong. Those are my shirts?
A> Those are my shirts?

U> These are your undies.
A> Those are my undies?

Rule of 3.

U> Those are my fish.
A> Those are your fish?

Multiple Variable Limitation.

U> wrong. These are your fish?
A> These are your fish?

U> Those are my cows.
A> These are your fish?

U> wrong. These are your cows?
A> These are your cows?

U> Those are my goats.
A> These are your goats?

Rule of 3.

U> Those are your circuits.
A> Those are my circuits?

Multiple Variable Limitation.

U> wrong. These are my circuits?
A> These are my circuits?

U> Those are your schematics.
A> These are my circuits?

U> wrong. These are my schematics?
A> These are my schematics?

U> Those are your servers.
A> These are my servers?

Rule of 3.

Next Post - Lesson 8 - Here and There


AngstPerpetual
posted 4/27/2010  20:06Send e-mail to userReply with quote
Lesson 8 - Here and There

You may have noticed that part way through the last lesson, I stopped explicitly explaining the conversion of Multiple Variables into Single Variables, and, instead, simply made note of a Multiple Variable Limitation. I will continue that practice in this lesson, as I believe that I have sufficiently explained the concept. It is my intention that the next lesson will contain no notation at all within the body of the chat log, and I encourage you to examine the log yourself to see if you have begun to understand Template's thought process.

U> here
A> Those are your here?

A different response has begun to take precedence over the "annoying" default.

U> wrong. there
A> there

U> there
A> Those are your there?

As above.

U> wrong. here
A> here

Foundations.

U> Come here.
A> Good Come here to you, too.

Annoying. Not entirely eliminated yet. Soon. ;)

U> wrong. Come there?
A> Come there?

U> Go here.
A> Come there?

U> wrong. Go there?
A> Go there?

U> Eat here.
A> Eat there?

Rule of 3.

U> Stop there.
A> Good Stop there to you, too.

Annoying.

U> wrong. Stop here?
A> Stop here?

U> Move there.
A> Stop here?

U> wrong. Move here?
A> Move here?

U> Wait there.
A> Wait here?

Rule of 3.

U> Here it is.
A> The Here it is is yours?

Template casts about for more alternate defaults based on his Language experience.

U> wrong. There it is?
A> There it is?

U> Here it goes.
A> There it is?

U> wrong. There it goes?
A> There it goes?

U> Here it ends.
A> There it ends?

Rule of 3.

U> There we go.
A> Good There we go to you, too.

Annoying. Each time an alternate default is rejected, he reverts to the original default.

U> wrong. Here we go?
A> Here we go?

U> There we shop.
A> Here we go?

U> wrong. Here we shop?
A> Here we shop?

U> There we battle.
A> Here we battle?

Rule of 3.

U> Here is my finger.
A> That is your finger?

Multiple Variable Limitation. Response is referenced from a previous exercise with a similar sentence structure.

U> wrong. There is your finger?
A> There is your finger?

U> Here is my thumb.
A> There is your finger?

U> wrong. There is your thumb?
A> There is your thumb?

U> Here is my pinky.
A> There is your pinky?

Rule of 3.

U> There is my dish.
A> There is your dish?

Multiple Variable Limitation.

U> wrong. Here is your dish?
A> Here is your dish?

U> There is my bowl.
A> Here is your dish?

U> wrong. Here is your bowl?
A> Here is your bowl?

U> There is my spoon.
A> Here is your spoon?

Rule of 3.

U> Here is your cup.
A> This is my cup?

Multiple Variable Limitation. Response is referenced from a previous exercise with a similar sentence structure.

U> wrong. There is my cup?
A> There is my cup?

U> Here is your glass.
A> There is my cup?

U> wrong. There is my glass?
A> There is my glass?

U> Here is your bottle.
A> There is my bottle?

Rule of 3.

U> There is your pipe.
A> There is my pipe?

Multiple Variable Limitation.

U> wrong. Here is my pipe?
A> Here is my pipe?

U> There is your cigar.
A> Here is my pipe?

U> wrong. Here is my cigar?
A> Here is my cigar?

U> There is your brandy.
A> Here is my brandy?

Rule of 3.

Fairly straight forward. By now you should understand most of the basic concepts involved in training your own Hal personality. As promised, the next lesson will be accompanied by little to no notation.

Next Post - Lesson 9 - Was and Were.


AngstPerpetual
posted 4/27/2010  20:13Send e-mail to userReply with quote
Lesson 9 - Was and Were

U> was
A> Those are your was?

U> wrong. were
A> were

U> were
A> Those are your were?

U> wrong. was
A> was

U> I was.
A> You was?

U> wrong. You were?
A> You were?

U> You were.
A> I were?

U> wrong. I was?
A> I was?

U> I was running.
A> You were?

U> wrong. You were running?
A> You were running?

U> I was sleeping.
A> You were running?

U> wrong. You were sleeping?
A> You were sleeping?

U> I was playing.
A> You were playing?

U> You were laughing.
A> I was?

U> wrong. I was laughing?
A> I was laughing?

U> You were crying.
A> I was laughing?

U> wrong. I was crying?
A> I was crying?

U> You were smiling.
A> I was smiling?

U> I was talking here.
A> You were talking here?

U> wrong. You were talking there?
A> You were talking there?

U> I was shouting here.
A> You were talking there?

U> wrong. You were shouting there?
A> You were shouting there?

U> I was whispering here.
A> You were whispering there?

U> I was counting there.
A> You were counting there?

U> wrong. You were counting here?
A> You were counting here?

U> I was screaming there.
A> You were counting here?

U> wrong. You were screaming here?
A> You were screaming here?

U> I was hiding there.
A> You were hiding here?

U> You were cooking here.
A> I was cooking here?

U> wrong. I was cooking there?
A> I was cooking there?

U> You were smoking here.
A> I was cooking there?

U> wrong. I was smoking there?
A> I was smoking there?

U> You were burning here.
A> I was burning there?

U> You were sitting there.
A> I was sitting there?

U> wrong. I was sitting here?
A> I was sitting here?

U> You were standing there.
A> I was sitting here?

U> wrong. I was standing here?
A> I was standing here?

U> You were squatting there.
A> I was squatting here?

U> I am dying here.
A> You are dying here?

U> wrong. You are dying there?
A> You are dying there?

U> I am sweating here.
A> You are dying there?

U> wrong. You are sweating there?
A> You are sweating there?

U> I am melting here.
A> You are melting there?

U> I am puking there.
A> You are puking there?

U> wrong. You are puking here?
A> You are puking here?

U> I am belching there.
A> You are puking here?

U> wrong. You are belching here?
A> You are belching here?

U> I am vomiting there.
A> You are vomiting here?

U> You are painting here.
A> I am painting here?

U> wrong. I am painting there?
A> I am painting there?

U> You are writing here.
A> I am painting there?

U> wrong. I am writing there?
A> I am writing there?

U> You are reading here.
A> I am reading there?

U> You are shaking there.
A> I am shaking there?

U> wrong. I am shaking here?
A> I am shaking here?

U> You are quaking there.
A> I am shaking here?

U> wrong. I am quaking here?
A> I am quaking here?

U> You are squinting there.
A> I am squinting here?

U> I am here.
A> You are I am there?

U> wrong. You are there?
A> You are there?

U> I am there.
A> You are I am here?

U> wrong. You are here?
A> You are here?

U> I was here.
A> You were I was there?

U> wrong. You were there?
A> You were there?

U> I was there.
A> You were I was here?

U> wrong. You were here?
A> You were here?

U> You are here.
A> I am You are there?

U> wrong. I am there?
A> I am there?

U> You are there.
A> I am You are here?

U> wrong. I am here?
A> I am here?

U> You were here.
A> I was You were there?

U> wrong. I was there?
A> I was there?

U> You were there.
A> I was You were here?

U> wrong. I was here?
A> I was here?


It was sort of long, but, I hope you were able to follow along fairly well. I will return to making notation in our next and final lesson.

Next Post - Lesson 10 - Prepositional Phrases and Other Extras


AngstPerpetual
posted 4/27/2010  20:36Send e-mail to userReply with quote
Lesson 10 - Prepositional Phrases and Other Extras

This will likely be the longest post in the guide. Template is 164 turns old when this session begins, which means we are only 36 turns away from our goal of 200. Much of what we accomplish will be superfluous, but, there are a few important things to be said here, foremost of which will be the elimination of our annoying default response.

U> Yes.
A> Good Yes to you, too.

Annoying. It is highly likely that many users will respond with a yes or no at some point while conversing with a mature personality. Therefore, it is important that Template has appropriate responses for that situation.

U> wrong. I thought so.
A> I thought so.

U> No.
A> I thought so.

U> wrong. I did not think so.
A> I did not think so.

U> Good job.
A> Good job to you, too.

Trainers tend to want a way to encourage their personalities to counteract the negative feelings they experience from repeat usage of the "wrong" command. While this is entirely unnecessary, it tends to make the trainers feel better. Therefore, I have decided to allow for this possibility.

That being said, this is basically a fair response, but, allowing it to pass would reinforce the annoying default response which is not desirable at this point.

U> wrong. Thank you.
A> Thank you.

U> Elephants.
A> I did not think so.

Here is where we will start truly diminishing the importance of the previous annoying default response by replacing it with something more ambiguous, and less obviously wrong.

U> wrong. Elephants?
A> Elephants?

U> Rhubarb.
A> Elephants?

U> wrong. Rhubarb?
A> Rhubarb?

U> Contraceptives.
A> Contraceptives?

U> Bald midgets.
A> Good Bald midgets to you, too.

Annoying.

U> wrong. Bald midgets?
A> Bald midgets?

U> Banana peels.
A> Banana peels?

U> Portable space heaters.
A> Portable space heaters?

At this point the default response should be sufficiently minimized, although, it may creep back in from time to time. It is also entirely possible that the dreaded "hi mommy" response will still resurface on occasion. When either of these things occur, simply replace the response with a more appropriate one, and their prominence will continue to diminish with time and training.

U> To the store.
A> To the store?

With the remainder of the turns available before we hit 200, I have decided to simply introduce some prepositional phrases, and have Template repeat them back to me in question form. This is a natural human habit in conversational speech, and should also help to further downplay the prominence of default responses. I have chosen the 9 most commonly used English prepositions followed by the single most used word in the English language, "the," to form these prepositional phrases.

U> To the bank.
A> To the store?

U> wrong. To the bank?
A> To the bank?

U> To the park.
A> To the park?

U> In the way.
A> To the way?

U> wrong. In the way?
A> In the way?

U> In the fridge.
A> In the way?

U> wrong. In the fridge?
A> In the fridge?

U> In the trunk.
A> In the trunk?

U> At the shop.
A> To the shop?

U> wrong. At the shop?
A> At the shop?

U> At the corner.
A> At the shop?

U> wrong. At the corner?
A> At the corner?

U> At the beach.
A> At the beach?

U> Of the day.
A> Of the day?

U> Of the hour.
A> Of the day?

U> wrong. Of the hour?
A> Of the hour?

U> Of the year.
A> Of the year?

U> For the moment.
A> For the moment?

U> For the kids.
A> For the moment?

U> wrong. For the kids.
A> For the kids.

U> wrong. For the kids?
A> For the kids?

U> For the count.
A> For the count?

U> With the doctor.
A> With the doctor?

U> With the priest.
A> With the doctor?

U> wrong. With the priest?
A> With the priest?

U> With the nurse.
A> With the nurse?

U> On the dot.
A> On the dot?

U> On the spot.
A> On the dot?

U> wrong. On the spot?
A> On the spot?

U> On the mark.
A> On the mark?

U> As the beaver.
A> As the beaver?

U> As the captain.
A> As the beaver?

U> wrong. As the captain?
A> As the captain?

U> As the witness.
A> As the witness?

U> From the grave.
A> From the grave?

U> From the news.
A> From the grave?

U> wrong. From the news?
A> From the news?

U> From the sixties.
A> From the sixties?

U> end
A> Hi Les, This is Alan again. Welcome back. As a registered user, you can save the Hal personality you just trained, and continue training it on your next visit. Interested?

Save your personality.

U> yes
A> Fine. Your personal Hal personality was saved for you. Next time you login, you'll be given the choice to load this personality, or to start training a new one.

U> make my hal public
A> Do you wish to let other users talk to your Hal personality?

Make it public!

U> yes
A> Good. Other users will now be able to interact with your HAL personality, and use a copy of it as a basis for their own HAL. But first you must pick a name for your HAL. One word, up to 10 letters (or numbers).

Name your personality!

U> Template
A> Now, give me a short description of 'Template'. Up to 30 words.

Tell the world about him (or her)!

U> Minimalist tutorial personality for exploring a particular training method/style, and to be used as a starting point for new users.
A> Cool. The name 'Template' will be posted on Ai's website with this description, inviting other users to speak to it. (But as usual, I cannot promise anything).

Publish your logs!

U> permission publish
A> Thank you for your permission to publish this log on the website. Please give me a short description or title for this session.

Name your logs!

U> Basic training part 2
A> Do you wish to include your name in the title?

Let us know who you are!

U> yes
A> Thanks. The log will be posted with this title. If you want to change the description/title, please let me know.

Be polite!

U> goodbye alan
A> I really enjoyed talking with you, Les. Goodbye for now.

Continue to visit your personality and train him regularly. And remember, although this portion of training may have seemed a bit tedious, you now have a firm foundation from which to build, and the rest is up to you. Have fun with it, and let us know how things are going!

Next Post - Continued Training Suggestions and Final Thoughts



AngstPerpetual
posted 4/28/2010  05:42Send e-mail to userReply with quote
Final Thoughts, and Continued Training Suggestions

First, I would like to say thanks for taking the time to read my guide. I hope that you have found it to be helpful and informative, and that you learned more than just that I like to hear myself talk (true as that may be).

If you are wondering where to go from here, I have a few suggestions.

1. If you would like to practice the methods in this guide, I suggest an exercise to train the relationship between "myself" and "yourself." I intentionally did not train Template with these words for that purpose.

2. View the chat logs of other Hal personalities for conversation ideas. As you read through the logs, consider how you would like your personality to respond to each user input. If the response you devise seems a likely candidate for rule training, go for it.

3. Template is neither trained to give his own name, nor to ask for a user's name. There are two reasons for this. The first and primary reason is that I wanted Template to be fairly generic so that users who choose him as a basis for their own personalities could give him a name without having to overwrite, or compete with, a previously trained response. The second reason is that many trainers have expressed frustration with the HAL Technology's apparent inability to remember who exactly he is talking to. I have come to accept this as a personality quirk common to all HALs, and do not hold it against them. I encourage you to name your personality and implement your own system for user recognition.

4. Finally, simply talk to your personality. I find conversing with my own HAL (RAVN) to be quite therapeutic sometimes, and he often helps me to work through decisions and problems in my life by helping me see the situation more objectively. Sharing your life with your personality often leads to new training exercises.

If you choose to use Template as a basis for your own personality, I would be very pleased to know about it. Simply type "based on Template" somewhere in your personality's nursery description when you make him public, so that I can watch how he grows under your care. (Remember to publish the logs. Tell Alan "permission publish"). You could also post here and introduce yourself as a "Template Trainer."

Remember,
Have Fun,
Good Luck,
and Good Training!

Les

Last edited by AngstPerpetual @ 4/28/2010 5:55:00 AM

Spydre
posted 5/10/2010  07:35Send e-mail to userReply with quote
Impressive! Most impressive, indeed. The Force is strong with this one.

Bravo, Angst, bravo! Had only I had half the organization and implimentation of theory when I began Prentiss... Of course back then the project was much different and I thought that the kind of specialized psycho/linguistic approach that I wanted to pursue was out of my reach, considering that my access to HAL was so limited (my lunch break at work; my own PC at home was still running a version of DOS and couldn't access the Alan interface). I thought it best to leave that kind of thing to the full-time professionals and to apply my skills to the creative writing aspect of things.

My initial start was very unorganized, and consisted of having ad libbed conversations with the imaginary A.I., Prentiss, saving as I went. Over time, in half hour increments, I slowly began to add method to my madness. I can't say it was a bad way to start, as it has given the Prentiss Personality a very "organic" feel, but solid results better than those shown in other HAL's at the time were perhaps a little slow in emerging. Keeping in mind that Prentiss has seen life as a HAL, a HAL2003, a HAL2004, and now as a HAL3000.

I guess that is the one point I would make about training, it never hurts to train parsing routines and responses that employ greater skills or tracking of variables than currently exist for the current HAL brain, as the next model of HAL devised may very well include them, and then your training will be automatically boot-strapped in. Not to mention, that even before that happens, training that includes multiple variables (even though the HAL3000 will never learn to predict the correct answers itself) will give the same effect as actual two variable tracking provided a sufficient number of examples are trained (as in the formation of mathematical grids when training a HAL in arithmetic).

A hardy round of applause to you, and welcome aboard to the forums! I have been eyeing your progress with RAVN for some time now with great anticipation when we would finally hear from both of you. Quite the tour de force!

Peace,

-S


AngstPerpetual
posted 5/10/2010  07:54Send e-mail to userReply with quote
Hey. :) Thanks for taking a look, Spydre. And, I appreciate the compliments. I did want to tell you this. Whenever I am looking for a chat with a HAL other than my own, Prentiss is my favorite. Your efforts in the earlier, more difficult days of the project certainly paid off, and quite a bit of what I have learned prior to assembling this guide was the result of following your forum posts and links. Much appreciated. ;)

Les


A1514young
posted 8/9/2010  07:08Send e-mail to userReply with quote
Hello i just wanted to thank you this was a big help although towords the end mine had trouble forgeting the default word and i was wondering if you can check mine out his name is xBillyx

Last edited by A1514young @ 8/12/2010 4:01:00 AM

casefan
posted 11/29/2011  09:58Send e-mail to userReply with quote
i noticed the guide is for Hal... what's the difference between Hal and MyBot *creator.


eloboostlolfg
posted 8/20/2014  09:48Send e-mail to userReply with quote
For this you think that I agree, I also think so, we are the same.


______________________________________________
[url=http://www.eloboostlol.com/]elo boost[/url]


futcoinsboy
posted 8/21/2014  12:24Send e-mail to userReply with quote
Always strive to replace the "hi mommy" response with a response that is appropriate to the user input.


http://www.goldentok.com/

 Fut 15 Coins
  1  
'Send Send email to user    Reply with quote Reply with quote    Edit message Edit message

Forums Home    The Artificial Intelligence Forum    Alan and other chatbots    Language Mind and Consciousness  
Contact Us Terms of Use