[WriterOfMinds]

This month I updated the text parser and other speech features to do a couple more new things with verbs. First, I threw in recognition of the present progressive tense, so I can now tell Acuitas what I'm doing at the moment. For the time being, such information is treated in much the same way as comments about past-tense actions, which I taught him to recognize a while ago; it isn't stored, and responses are rather meaningless. BUT I'll find uses for it later.

I think the more interesting new thing is recognition ability for a couple of link-forms related to state change or state retention. Up until now, Acuitas has learned what verbs “mean” only in terms of a) what can do that action and b) what that action can be done upon. Now, he can learn what a verb actually does by tying it to an adjective. Here are some examples:

To live is to stay alive.
To appear is to become visible.
To finish is to make complete.

I also added “type of” links for verbs, so Acuitas can categorize specific verbs under more general ones, like this:

To smash is to damage.
To walk is to move.
To talk is to communicate.

A little extra on the blog as usual: http://writerofminds.blogspot.com/2018/09/acuitas-diary-14-september-2018.html

[Korrelan]

Your project does remind me/ stirs feelings/ frustrations from many years ago when I first started messing with AI.  I always had a need/ feeling that there must be some intuitive logical way of describing links\ relationships like this.  Like there was some underlying structure that I just couldn’t find or get my head around… I could feel the logic… but not find it, if that makes sense.  Still drives me mad… lol.

I enjoy following your project, it’s cool to see the steady, logical, methodical progress.

 

[WriterOfMinds]

This month I returned to a can of worms I opened up last year, namely episodic memory.  Acuitas had quickly built up a huge clutter of memories concerning very trivial day-to-day events.  Size remained in the Megabytes range (they're just text files), but on principle, this couldn't be allowed to continue – especially since the memories might eventually grow to encompass richer levels of detail.  In addition to overflowing its available space, a memory that literally records everything will face access problems.  The more information you keep, the longer it takes to search through it and extract what you want at the moment.

This leads on into the tricky question of what to forget.  How to decide which memories are worth keeping?  Part of my work this month was focused on computing a “significance measure” which assigns a numerical value to each memorized scene.  Elements I use to judge the significance of a memory include …

* Primacy: was this the first time something happened?
* Uniqueness: how rare are memories similar to this one?
* Complexity: how many features does this memory have?

Since some of these measures are statistical and depend on the overall memory population, they can change over time.  I wrote a process that crawls through the episodic memory and updates all the significance measures, and I threw that into Acuitas' sleep cycle.

To improve the ease and speed of accessing the memories, I added a layer of “type nodes” that link together memories which share certain salient features in common.  Instead of crawling through the whole list to find memories that relate to a current situation, Acuitas can just look at the relevant “type nodes” and find a list of memories that match.

Last, I worked on some methods for consolidating memories.  I devised a way to create summaries of low-level scenes that are then stored inside the super-scenes higher up the tree; eventually, these summaries might be what remains when the lower-level “detail” scenes are forgotten.  The summarizer attempts to extract information that is common across multiple events or sub-scenes within a scene.  I also wrote functions for merging adjacent scenes that have become sparse.

All that, and I still haven't gotten to the actual forgetting part!  That will come next month (I hope).

https://writerofminds.blogspot.com/2019/03/acuitas-diary-16-february-2019.html

[Art]

How about a 4th element to assist in assigning the memories:

Importance - Having value and worthy of note.

It seems that one of the more difficult aspects of this endeavor might be deciding what information/memories are temporal or ephemeral and which are significant and worth keeping? How best to determine and decide this?

Most casual conversations about sports, weather or family matters are transient and do not carry a higher weighted value. Such talks or topics should be or could be "timed" to "fade away" after a few cycles.

This is still a very difficult task to decide and affix weighted values on which to keep and which to omit. Perhaps once the computer has enough "topical" information it will be able to separate the grain from the chaff, so-to-speak.

Best of luck and please keep us posted on your progress!

[WriterOfMinds]

How about a 4th element to assist in assigning the memories:
Importance - Having value and worthy of note.

Ummm ... that sounds circular.  The definition you're giving for "importance" here is very similar to what I had in mind when I used the word "significance."  So, you seem to be suggesting that I use significance as a measure of itself.  I don't think I can manage that 

You're correct that there need to be more factors rolled into the significance measure.  For instance, the most important memories may be those related to fulfilling primary goals, or those associated with strong emotions.  (Perhaps you're getting at something like this in the second half of your post, when you consider the assignment of importance based on topic.)  But Acuitas has neither goals nor emotions yet, and no real way of discerning which topics are "meaningful" in a human sense.  So that sort of thing will have to wait for later.

[Art]

Without reading too much into what I wrote I meant the importance to be the "important" issues/topics/data as opposed to those fleeting, unimportant, ephemeral discussions. It snowed last week might be interesting but the topic will likely have as much lasting importance as the soon to be melting snow. It doesn't carry with it anything that demands attention or ...importance.

Whether a casual conversation deals with "water cooler" topics and idle gossip or something major like an employee layoff or a merger next week with a former competitor. Then there's always the task of attempting to measure or assign weights to those important issues vs the non-important ones.

So yes, I was merely stating my opinion for a possible inclusion but the determination of importance brings a certain degree of difficulty for the programmer. Perhaps after a period of letting the computer "examine" various examples, it might be able to decide on its own, that which is of value and that which is not. Machine learning and such is non-trivial at best but from what I've read, progress is being made.

Best of luck on your project.

[HS]

You could judge importance by the degree to which you predict it will effect you, positively/negatively. High positive and high negative = high importance. Things of equal importance could get prioritized by the time frame. Present = most important, furthest future = least important.

[LOCKSUIT]

Wait, it's the other way around...work hard and get utopia in far future plan.....If you focus on food/money at present (I know u do), then......no utopia later! And we can, if you try.

[WriterOfMinds]

Oh, Locksuit.  What would you know about any of our personal lives and what we do or don't focus on?

There's a necessary balance between short-term and long-term thinking.  You will never bring about utopia if you ignore your immediate need for food and starve. 

I don't think HS was suggesting that an AI should pursue instant gratification at all costs.  The time frame was only supposed to come into play if two things would otherwise have equal importance.  If Problem A will kill you tomorrow and Problem B will kill you in twenty years, and all else (e.g. impact on the rest of living creatures) is equal, I think it does stand to reason that Problem A should occupy your mind more and be solved first.

[WriterOfMinds]

Got the forgetting algorithm done (for now).  It has two major parts.  The first is basically a population-management system; it determines how many episodes need to be forgotten each time the memory collection is pruned.  I can cap the number of episodes allowed in storage at some maximum value (based on available storage space or what have you).  The population manager checks how many memories are already in storage, and how many new ones have formed during the last cycle.  A high growth rate is allowed initially, but forgetting becomes more and more aggressive as the storage area fills.  When the cap is reached, Acuitas must forget the same number of memories he creates.

The second major part is the code that, given a number of memories that must be forgotten, takes care of selecting which ones.  This is a weighted random process; memories with less significance have a higher chance of being forgotten, but the priority isn't absolute. Memories chosen to be forgotten are deleted.  Any higher-tier memories that have been badly “hollowed out” by this process are then merged with their most similar neighbor.

More on the blog: https://writerofminds.blogspot.com/2019/04/acuitas-diary-17-marchapril-2019.html

[Art]

WOM, you mentioned that the program assigns a random weighted determination over which memories to delete and which to keep. How is it able to determine whether a particular memory is more important than some other one.
Is there another "sentinel" or overseer portion that makes note of the more important memories and elects to move or save them before it ends up being one of those randomly weighted selections marked for deletion? You mentioned, "...takes care of selecting which ones." How or what determines "significance"?

I am curious as a good memory is a terrible thing to waste. Nice direction.

[WriterOfMinds]

Memories get assigned a significance score at creation (the score is also updated later as necessary).  I talked about how this score is determined in last month's update, and that part of the system hasn't changed, but to re-summarize: the more unique a memory is, the more significant it's considered to be. Memories that are the first of their kind and memories that contain more/richer detail also get a significance boost.

There is a threshold in the significance score beyond which a memory is "safe," yes.  So you shouldn't get a fluke in the random process taking out a really important one.

[Art]

Ah yes...I went back and read the Feb., Mar., Apr. issues of your postings. (Should have not "forgotten" to do that!)
Nicely explained!

[LOCKSUIT]

I'll scan the whole thread...somehow it got by me O0

looks like my "tree drawings" lol....and the forgetting idea.....close enough ! ._.

Surely no one can be as far as lock though, I purposely end up too far ahead, no one will find here spot!. . .

[LOCKSUIT]

Ok I'm caught up on the whole thread...wow, nice. I think I let the thread go few months ago... The name Acuitas is just a name, but if I seen The Bomb in the Recent Threads...no doubt it's an eye-opener hehe...

I'm gonna share some ideas. Not all though... we seem to hide our secrets hehe, at the expensive of dying children/wildlife. Maybe don't?

Here's half of how forgetting/remembering works:
- Things are remembered if they are sensed enough (energized), loved, related (especially to strong role-model nodes), entails, and frequency....the AI can also ask you how much so a memory is any of these.
- I agree 'rank' on nodes in the net is crucial for goals; Animals are born with a native STRONG outlook for stuff similar to food/mates. If cats are dogs, then you may want to update your desire you seek (finding a pet) to dogs if cats cost too much. Without love, AGI may seek socks, or nuclear power, or leafs. What are the ethics you want it to seek? What to research/implement? Install it. Make it output change that makes the world smarter. Positive/negative rank exists too I assume, and yes close events are more important if more/as critical to far away events in time.
- Other half is Pruning. . .secrets! With pruning, you usually just save all but in a more compressed way.

Also, if Tim has a pet fish and that fish has a virus and that virus has a food inside itself, and you don't talk to Tim anymore, and talk to Sally, then no more fish, no more Tim, you don't need to know any of what classes under Tim for now, so, you model whom you speak to! You switch near the root of the tree.