One of the areas I am exploring is how to add emotional content to a bot. Have/will your bot have emotions?

Do you have a model you are using for emotions? If so what?
What do you believe are the core emotions of a human?

In a few weeks from now, we’ll introduce our research tab. You’ll love it! It’s full of emo stuff!

I haven’t implemented this yet, but I have an idea in my head that uses “state modifiers”, based on user input, to change stored values in Morti’s database, and are combined and compared, to create an overall “mood”. This idea is inspired by some discussions that have taken place here, and involve a number of community members, and will one day alter Morti’s output, depending on his “mood”. It’s still quite a long way off, but I have a lot of notes and pseudo-code “written down”

I will have simulated emotions.  One thing I want to do is get around to creating an avatar for Grace.  If the conversation turns to something she has no knowledge of, or if she is having problems understanding the input, she will put on a sad or perhaps angry face, and become ‘discouraged’.

But, as you teach her and explain and give definitions to words she initially didn’t know, and she starts understanding, she will light up with a smile and then become happy.

Her current ‘mood’ will also affect the choice of words, and the types of things she says.  So the mood won’t just be a variable assigned to a specific value, it will actually steer the types of things she says.  If she is angry (perhaps at not understanding a lot of what you have said), she may reply with ‘I don’t care’ if you asked her opinion on something, but if in a good mood she will respond more with ‘oh, I’ll be happy to help you out with that….’

Of course, ‘emotions’ will have an off button . . simply uncheck the ‘emotions enchanced chatting’ checkbox.

.....also, Grace will, if she finds contradictory information between what she directly found (or indirectly via logical deduction) and that contradicts with what you just told her, she will put a ‘confused look’ on, and respond with something like ‘What the heck…. you said…......’

if she is experiencing continuous ‘confusion’ like this, she may grow angry

She will even understand jokes, simple ones at first.  If you say ‘Bob is a poor man, he only makes 1 million dollars a year’, it will know that is WAY outside the average salary for a person, and respond with ‘simulated laughter’ and corresponding avatar animation.

Victor Shulist - Mar 15, 2011:

...Of course, ‘emotions’ will have an off button . . simply uncheck the ‘emotions enchanced chatting’ checkbox.

Now THERE’S something I wish I had!

Picard…. “Data… perhaps you should deactivate your emotion chip for now”....

<< Data moves his head to the left >>  ....  done sir.

“Data… there are times where I really envy you”

Merlin - Mar 15, 2011:

Do you have a model you are using for emotions? If so what?

There is a emotions-model developed for AI (I’m using this in my own project), called the PAD-model (Pleasure, Arousal, Dominance).

You can find everything about the model here:

- http://www.kaaj.com/psych/ai.html
- http://www.kaaj.com/psych/scales/emotion.html
- http://www.kaaj.com/psych/scales/temp.html
- http://www.kaaj.com/psych/scales/pl.html
- http://www.kaaj.com/psych/scales/ar.html
- http://www.kaaj.com/psych/scales/dom.html

Enjoy

Of course, ‘emotions’ will have an off button smile . . simply uncheck the ‘emotions enchanced chatting’ checkbox.

I’m also thinking of some way to de-activate the next few stages that I plan to add, so the core stuff remains easy to debug (my testcases use regular expressions to compare the output against).

Now THERE’S something I wish I had!

I haven’t planned any ‘emotion emulation’ for the near future, but I plan to add a proper output generation stage next, which will use an algorithm based on the Meyer-Briggs personality type indicator to modulate the output (at least, that’s the plan).

Jan, using the MBTI is a very useful idea. But although ‘personality’ and ‘emotions’ are very much connected, they are two different things.

I could see however, that the MBTI could be a great ‘template’ that can steer how (for example) the PAD-model would react to certain stimuli.

Hans Peter Willems - Mar 15, 2011:

Jan, using the MBTI is a very useful idea. But although ‘personality’ and ‘emotions’ are very much connected, they are two different things.

I could see however, that the MBTI could be a great ‘template’ that can steer how (for example) the PAD-model would react to certain stimuli.

Yes, they are related, but not really the same, are they?
I am not yet 100% certain how I will integrate this, but I am thinking of a gauge to determine whether the network is more ‘introvert’ or ‘extrovert’ and if it is more ‘judging’ then,....
This would determine how a standard response like ‘yes’, ‘no’,... would be modified.

The way I see it (or will use it) is that I have the PAD-model which describes the current state of emotions, and have the interaction of those states based on stimuli (sensors, conversation, etc.) be calculated in reference of a MBTI profile. So the MBTI profile describes how ‘hard or soft’ certain stimuli will drive the PAD variables up or down. This way ‘personality’ (MBTI) interacts with ‘emotions’ (PAD).

Jan, thanks again for the pointer to MBTI, this is really good stuff

If your system does NLG, here’s a link for some ideas about incorporating personality:
http://mi.eng.cam.ac.uk/~farm2/papers/phd-thesis.pdf

Gary, that’s quite a lot to read but thanks anyway. Added to my research-paper backlog

Another paper I found (this one is just a few pages), that might be of interest to those working on ‘emotions’:

http://condor.depaul.edu/elliott/papers/aaai94n/aaai94n.html

Thanks Gary, it looks very interesting. Skynet-AI does do NLG for some of its responses. In future projects, I am hoping to expand on it and add more persona/emotional driven response systems.