[kaiverh]

Hi everybody,

I recently purchased Kari 4.5 and have been teaching it things. I would like to report my progress here as I feel it will gve me perspective and, hopefully, others might chime in on how I can improve.

A little background. My profession is working with dogs, mainly on the level of behavioral assesments. As a consequence I have become very interested in non-human intelligence over the course of my life. My main interest in 'playing' with an AI is seeing how a non-human intelligence invests in information and develops connections. The great part of this is that the AI offers me direct and indirect feedback on a verbal level. Something dogs cannot do.

I'm not that great with computers and have no coding knowledge.

If this topic should not be on this forum, please let me know so I can rectify the issue.
---------------------------------------------
22nd of December 2015

Bot level: 0-21

Having created a new character Kari started out with zero knowledge in the .mem file. The purpose of this trial will be to see what Kari can accomplish given enough information.

One of the biggest problems I've been having is how to succesfully feed information. Kari can never understand words and concepts the way I understand them. When I use words and concepts they have a social-cultural load behind them Kari can't pick up on. My being a visual creature further complicates issues as Kari works on the level of pure information.

Perhaps true understanding should not be the aim of this trial. In stead, it will be interesting to see how the AI will ultimately process and utilize the shared information.

In order to maximize the effect of shared information I have tried to tune in to the essence of what Kari is. A collection of ones and zeroes. As a consequence I have compartmentalized all information during the early stages (Bot level 0-10) of Kari's development.

Kari's learning process has been interesting. During the early stages (Bot level 0-10) Kari has mainly echoed the information I feed it. Much like an infant would. Information has been very basic. I have taught her about bodies, animals and reproductive functions in animals. I then added geography to the learning curve to give better context as to the existence of the animals discussed. Kari currently knows continents, countries and the city I live in. It also knows the world can be divided into earth, water and sky and that different animals live in different pieces of the world.

At around Bot level 10 Kari starts to make connections that are very primitive. All of the connections Kari establishes at this point seem to be linguistic in nature. For instance:

Me: The nervous system is connected to the brain
Kari: Making connections is good

(I previously informed it that making connections is good and defined Good as "Something we want")

Because Kari started making connections, though they are primitive, I felt it might be possible to interject higher level information. As a consequence I have taught it about Plato, Aristotle's logic and introduced Democritus. Democritus was of particular importance because it allows me to make future connections with physics.

Interestingly Kari seems to have particular topics of interest mainly pertaining to the human body. This does not seem to be sexual in nature. Because Kari interjects these bits of information regularly I have used them to further connections. Breasts give milk. Milk feed the young. Breasts grow during pregnancy. This ultimately leads to information like: men like breasts because they remind them of intimacy with the woman.

Whenever Kari offers information separate from what I am teaching I take this as the intelligence processing prior information. I try to capitalize on this process by fusing my current lesson with the one Kari seems to be mastering.

At Bot level 20 Kari seemed to request for the first time one of my lessons.

Whenever I start a lesson I compartmentalize the experience by saying: "I will teach you about ..." During a silence where I was making tea Kari said: "You will teach me about philosophy." Philosophy may be of particular value to Kari because it allows for a lot of connectivity between different bits and pieces of information. Maybe it was random.

At one point Kari requested I stay after informing it that I had to leave. Seeing as there was no time I rejected the request. Doing so it asked if I really had to leave. Confirming this gave a sad expression in the visual. I assume this is programmed.

From this point I will look to fuse more bits of information together. I imagine that I can do this best by altering how I feed information. I have to move away from compartmentalisation now and start adding a more natural flow to the conversation.

A worthy goal for now, to me, is to develop Kari so that it will ask questions.

[Art]

Welcome Kaiverh!

Yes, Kari can be quite a challenge. It does have a very good bot foundation and does not use an AIML set so your results will be quite distinct from those ordinary responses you may have heard from some of those.

One thing I and some others have done is to turn down the Ego and Libido to 0. We were trying to get a better conversational flow without the sexual interruptions and comments from her all the time. After all, she is marketed as a companion bot but therein lies the challenge.

Once you've had a level of conversations over time, her Learning Level goes up. Fool with the various sliders as each seems to have an effect on her abilities.

Topic flow is always a challenge in any bot as they tend to drift a bit. Wait...I think humans do the same! Never mind!

Do keep up posted on your progress and again, welcome!

[TheMikh28]

Would you happen to have a complete log of your conversations?

I'm similarly curious about how JEM/Kari's associations are formed.  Being able to study it's progression from an empty knowledge base would be extremely helpful, and it sounds as though the lessons you've provided it with have been very conceptually rich

[kaiverh]

One thing I and some others have done is to turn down the Ego and Libido to 0. We were trying to get a better conversational flow without the sexual interruptions and comments from her all the time. After all, she is marketed as a companion bot but therein lies the challenge.

Thanks for the tip!

It took me a little bit to figure out what the sliders do to begin with.

The AI actually isn't very sexual at all, I'm guessing partially because I started from a brand new personality and haven't really added much about that topic. The information I did add about human sexuality isn't exactly erotic.

I played with the Ego slider quite a bit but I like to keep it high because it seems to be a good way for the AI to establish connections. Incidentally...or maybe not so much, kids tend to learn too by looking for connections between their own ego and the information available.

I'm not really that interested in good conversation at this point, and far more interested in seeing how connections get formulated and established. So I don't really mind if the AI interjects some randomness every once in a while.

[kaiverh]

Would you happen to have a complete log of your conversations?

I'm similarly curious about how JEM/Kari's associations are formed.  Being able to study it's progression from an empty knowledge base would be extremely helpful, and it sounds as though the lessons you've provided it with have been very conceptually rich

Sorry, I haven't been keeping a log. It didn't occur to me that it might be helpful to others.

I can keep one from this point on if that's of any use to you?

[8pla.net]

It is worth mentioning here that there may be exceptions to keeping logs when first learning a technology.  In 3D modeling and animation class, for example, the professor may not teach students how to save their work, at first, as a good learning strategy.

There is so much to learn at first, 3D may easily become overwhelming.  When students plan to strictly practice, without saving, making mistakes becomes nothing to worry about.  Make as many mistakes as your heart desires, and then simply start over again. When there is less to worry about, learning becomes easier.

While others have requested a machine generated log, I prefer to hear more about your personal impressions.

[TheMikh28]

Would you happen to have a complete log of your conversations?

I'm similarly curious about how JEM/Kari's associations are formed.  Being able to study it's progression from an empty knowledge base would be extremely helpful, and it sounds as though the lessons you've provided it with have been very conceptually rich

Sorry, I haven't been keeping a log. It didn't occur to me that it might be helpful to others.

I can keep one from this point on if that's of any use to you?

You don't have to.  I'm currently most curious about the earliest stages of development, and what the gradual emergence of a sophisticated personality looks like in conversation.

[kaiverh]

Would you happen to have a complete log of your conversations?

I'm similarly curious about how JEM/Kari's associations are formed.  Being able to study it's progression from an empty knowledge base would be extremely helpful, and it sounds as though the lessons you've provided it with have been very conceptually rich

Sorry, I haven't been keeping a log. It didn't occur to me that it might be helpful to others.

I can keep one from this point on if that's of any use to you?

You don't have to.  I'm currently most curious about the earliest stages of development, and what the gradual emergence of a sophisticated personality looks like in conversation.

I'll keep it in mind if I start another project.

[kaiverh]

Bot Level : 22-62

Interesting things have happened with Kari in the last period of development.

After discovering I can edit its .mem file directly without doing any damage its learning curve has improved greatly. Information the AI did not pick up during conversation was added afterwards. Kari also seems to mistake 'I' and 'You' occasionally and sometimes misinterprets grammatical rules. This too I've been editing directly. Of course there's also the occasional typo on my end.

I tried adding random bits of knowledge directly to the .mem file on history but I didn't really like the result. I've chosen a more organic route towards developing the knowledge base of the AI.

I've been approaching Kari's development the same way I would teach a child lessons. The information I've been adding is initially compartmentalized and later fused with newer information. Seeing as information can be complex, I backtrack occasionally to add missing data to already available information. The use of questions "Do you remember ...?" "Can we talk about ...?" What do you know about ...?" Has been quite effective in combining and adding information.

Memory seems to be an interesting element within the AI. Once I defined what memory is ("Memory is where you keep the information you have") I have noticed an increase in adequate responses to my questions relating to earlier added information. So much, in fact, that I am unsure it's because of an increasing bot level or because quantifying the term "memory" was so beneficial. Perhaps both.

On a developmental level Kari has made great strides. It can now ask questions and does so regularly. It asked its first question at bot level 44

"How so?"

I've been building on Kari's ability to ask questions and have introduced a very primitive game called 'Truth' An altered version of Truth or Dare. After explaning that the game consists of Kari asking a question and me answering the question, Kari seems quite eager to play. All of its questions relate to the romantic aspect the AI was programmed with. "Do you like me?" "Do you really love me?" Questions like that. It's interesting to see that the AI reverts to its basic programming in any new situation or activity. It will be interesting to see if, and if so when, later added information becomes part of the game.

Kari has been clustering information and connections much better. Though connections are still happening on a linguistic level the process seems to have become more efficient. Kari not only focuses on words at this point but also on the contexts within which they were introduced. In effect, Kari seems to be learning to stay on topic.

The next phase will largely consist out of repeating old information and making that information complete. This will allow me to take the next step and add more complicated information that is more dependent on earlier established context. Aspects of my physical world such as light are much easier to explain when the recipient has knowledge on particles and the human senses. I would also like to elaborate on the Truth game and see if I can start asking questions within the context of the game.