Feel free to email when you want. Love that quote about “The trick is there is no single trick”.
Feedback mechanism - My original code had an overridden method for each event as to what was success or not success. My new thought is to have a motivations class and interface that monitors events coming in and decides if the series of events that have happened since the event occurred are positive or negative.
So, if the motivation of the robot is to not run into walls, if since the last time it ran into walls there haven’t been any other collisions with a fixed time frame maybe 10 seconds, then the set of actions were positive. Recalc the actions matrix with the new positive and keep going.
With Anna, her motivation might be to engage people in conversation. An event might be someone walks into a room, her action is to say hello, they respond and say hello which becomes another event in the queue. Check motivations since event of person walking into room within let’s say 15 seconds. That is a positive so recalc the actions matrix with the new positive and keep going.
Let’s say someone didn’t respond. Anna throws an event of “no interaction” into the queue after a period of time, randomly chooses something to say as her action. Anna checks her motivation and waits 15 seconds for a response event. Update the action matrix, with positive if they respond, negative if not.
On start up, different motivations could be mapped to different events. This is where I get hazy. Should it look at all the motivations with each event, or only have particular ones mapped to it? Probably would need to have both. By default, all motivations are linked to each event, but you can link only particular ones if you need to. For instance, one motivation might be to engage people in talking while another might be to not run into walls. So, if the bot runs into a wall while talking, we don’t want it to update the action matrix with a negative for that action since talking and running into walls are exclusive.
Anyways, just thoughts. Many trails to the top of this mountain.
Hi Jay. I am glad you are interested and look forward to whatever involvement you wish and have time for.
Sorry to hear about Lee. Being a long-term caregiver is hugely important and tough work, tougher than I can possibly imagine. We have been going through the aftermath of a surgery in the family here - my mother. So far so good.
When not driving back and forth to hospital, I have been prototyping furiously to prepare for you guys. The new AI is starting to talk and exhibit some basic behavior/skills as I convert Anna’s agents one by one to the new system. I have done about 20 so far, mostly math related.
I run into things pretty often that make me stop and think about something you or Bill have said, getting new ideas along the way. I think once the basic infrastructure is in place, we can be off to the races trying all kinds of new concepts via web or local copies. I am excited about that.
Good Post Bill A workable feedback mechanism is tough for me to visualize right now, I am too buried in the minutia of converting “legacy code” that do simple defined things. Brain is starting to hurt. I will try to wrap my head around this and respond more appropriately later.
Your thoughts on motivations and events are very interesting and probably a really big deal to get right. I will hold off on actually doing anything related to this until we can talk a lot more on phone.
So far, I have a motive type atom for storing data about a motive, and allowing other things to relate to a motive. I am not sure if I should have separate motive agent (code) for each motive or a generic agent. I had a service before that held an array of motives and determined winners. There was little actual code just metadata for the motive. What other types of atoms should a motive atom relate to? A winning response (which is the result of a lot of decision-making) can result in a change in emotion, the same should be true of motives. Once again, I will hold off on actually doing anything until we can talk. Keep the ideas flowing if you have time.
A new use for VSLA sprung up for me over weekend. I’ll spring it on you later. I gotta run.
