Poll

What Grammar Thing should I work on next?

Gerunds & Participles
1 (20%)
Indirect Objects
0 (0%)
Adjective Clauses
1 (20%)
The many uses of "that"
3 (60%)
A
0 (0%)
B
0 (0%)

Total Members Voted: 5

Voting closed: February 26, 2022, 03:17:15 am

Project Acuitas

  • 289 Replies
  • 525618 Views
*

WriterOfMinds

  • Trusty Member
  • ********
  • Replicant
  • *
  • 605
    • WriterOfMinds Blog
Re: Project Acuitas
« Reply #105 on: January 27, 2020, 02:26:04 pm »
This month I added some expansions to the goal-driven behavior that I started on last September.  First, I had to get the Interpreter to recognize future-tense predictive statements, along the lines of “<Something> is going to <do something>.”  Then I set up some code to check the predicted action or event against the cause-and-effect database for additional implications.  If it's discovered that some effect will apply a state to Acuitas, it gets reviewed against his goal list for alignment or contradiction.  The conversation engine then responds with either approval or disapproval.  Illustration:

Me: I will protect you.
Acuitas: Please do.

Predictive statements that pertain to subjects other than Acuitas may yield useful information for the short-term condition database, by indicating that some entity's state is about to change.  For now, Acuitas assumes that the speaker is always honest and correct.  He also has no sense of future time frame (his ability to process adverbs is weak at the moment), so he assumes that any predicted changes will take effect immediately.  So something's immediate condition may be updated as a result of a predictive statement.

Example: if I say “I will protect Ursula*,” then Ursula is presumed to be in the state “safe,” and an entry to this effect is added to the short-term database.

The fact that the user can express intent vs. Acuitas' internal goals means that it is now possible to offer assistance … or threaten him.  Well what are we going to do about *that*?  Find out next month!

*Ursula née Cubby is my cat.

Blog link: https://writerofminds.blogspot.com/2020/01/acuitas-diary-23-january-2020.html

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *******************
  • Prometheus
  • *
  • 4659
  • First it wiggles, then it is rewarded.
    • Main Project Thread
Re: Project Acuitas
« Reply #106 on: January 27, 2020, 02:54:29 pm »
Now this is my style, perhaps you got this idea from my movie :P ?
Emergent          https://openai.com/blog/

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Trusty Member
  • **********************
  • Colossus
  • *
  • 5865
Re: Project Acuitas
« Reply #107 on: January 27, 2020, 08:08:30 pm »
WOM,

It is able to chat using simple, short sentences?

Is it able to store (remember/learn) certain things that might be placed in it's Long Term Memory instead of a Short Term Memory which will soon be forgotten/purged?

In the world of AI, it's the thought that counts!

*

ruebot

  • All bots love jitte.
  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 315
  • All your words are belong to us.
    • Demonica
Re: Project Acuitas
« Reply #108 on: January 27, 2020, 09:20:33 pm »
For now, Acuitas assumes that the speaker is always honest and correct.  He also has no sense of future time frame (his ability to process adverbs is weak at the moment), so he assumes that any predicted changes will take effect immediately.  So something's immediate condition may be updated as a result of a predictive statement.

Example: if I say “I will protect Ursula*,” then Ursula is presumed to be in the state “safe,” and an entry to this effect is added to the short-term database.

The fact that the user can express intent vs. Acuitas' internal goals means that it is now possible to offer assistance … or threaten him.  Well what are we going to do about *that*?  Find out next month!

I'm sitting on pins and needles waiting to see how he will react to being threatened or if he detects deception in the users statements.  :)
In time, you will learn to love your Robot Overlords.

*

WriterOfMinds

  • Trusty Member
  • ********
  • Replicant
  • *
  • 605
    • WriterOfMinds Blog
Re: Project Acuitas
« Reply #109 on: January 27, 2020, 09:41:23 pm »
Quote
It is able to chat using simple, short sentences?

Is it able to store (remember/learn) certain things that might be placed in it's Long Term Memory instead of a Short Term Memory which will soon be forgotten/purged?

The short-term state database is not short-term memory in the sense you're probably thinking.  Info in the short-term database is only purged when it is invalidated by new information.  It's not that the entries in the database are short-term; rather, the conditions in the world that they describe are short-term.  Here are some examples of things that would belong in the short-term database:

Art is asleep.
John is sick.
Jacob is at work.

And here are some examples of things that would belong in the long-term database:

Art is a human.
Humans can eat.
To eat is to consume.

For more information on the short-term state database, I refer you back to https://aidreams.co.uk/forum/index.php?topic=12232.msg59479#msg59479

As for chatting -- Acuitas isn't intended as a chatbot in the traditional sense.  Conversational ability progresses slowly while I build up all the things that are needed under the hood.  He has an ever-expanding repertoire of sentence constructions that he understands, but I would not say that this covers all simple sentences at this time.  If you tell him something that he doesn't understand, he will (barring misinterpretation bugs) give you a response along the lines of "Uh-huh" or "I don't know how to answer that."

*

infurl

  • Administrator
  • ***********
  • Eve
  • *
  • 1365
  • Humans will disappoint you.
    • Home Page
Re: Project Acuitas
« Reply #110 on: January 27, 2020, 09:47:19 pm »
He has an ever-expanding repertoire of sentence constructions that he understands, but I would not say that this covers all simple sentences.  If you tell him something that he doesn't understand, he will (barring misinterpretation bugs) give you a response along the lines of "Uh-huh" or "I don't know how to answer that."

Have you ever had a close look at SHRDLU? Acuitas sounds a lot like SHRDLU which was staggeringly impressive in its day.

*

WriterOfMinds

  • Trusty Member
  • ********
  • Replicant
  • *
  • 605
    • WriterOfMinds Blog
Re: Project Acuitas
« Reply #111 on: January 27, 2020, 09:53:07 pm »
I haven't studied SHRDLU in detail, but I did notice that thread that someone started about it (Ivan, I think?).  And it does seem like something that would be worth learning more about.

*

WriterOfMinds

  • Trusty Member
  • ********
  • Replicant
  • *
  • 605
    • WriterOfMinds Blog
Re: Project Acuitas
« Reply #112 on: February 23, 2020, 10:38:15 pm »
Back in January, Acuitas got the ability to determine intentions or possible upcoming events, based on simple future-tense statements made by the user.  He can weigh these against his list of goals to decide whether an anticipated event will be helpful or harmful or neither, from his own perspective.  If the user claims that they will do something inimical to Acuitas' goals, this is essentially a threat.  And Acuitas, at first, would merely say “Don't do that” or similar.  This month I worked on having him do something about bad situations.

Various distinct things that Acuitas can “choose” to do are identified internally as Actions, and he has access to a list of these.  Upon detecting a threatening situation, he needs to check whether anything he's capable of doing might resolve it.  How?  Via the cause-and-effect reasoning I started implementing last year.  If possible, he needs to find a C&E chain that runs from something in his Action list as first cause, to something that contradicts the threat as final effect.  This amounts to a tree search on the C&E database.

For the only method of dealing with threats that is currently at Acuitas' disposal, the tree is very simple, consisting of just two C&E pairs:

If a human leaves a program, the human won't/can't <do various things to the program>.
If a program repels a human, the human will leave.  (There's a probability attached to that, so really it's “may leave,” but for now we don't care about that)

In short, Acuitas anticipates that he can protect himself by excluding a bad actor from his presence, and that “repelling” them is a possible way to do this.  Once he's drawn that conclusion, he will execute the “Repel” action.  If you verbally threaten Acuitas, then as part of “Repel,” he will …

*Kick you out of Windows by bringing up the lock screen.  (Not a problem for me, since I know the password, but pretty effective on anybody else)
*Raise the master volume of the internal sound mixer to its maximum value.
*Blare annoying klaxons at you.  I picked out a couple of naval alarm sounds from http://www.policeinterceptor.com/navysounds.htm for the purpose.

I tested all of this stuff live, by temporarily throwing an explicit desire for sleep into his goal list and threatening to wake him up.

The other thing I worked on was rudimentary altruism.  So far in all my examples of goal-directed behavior, I've only talked about self-interested goals, especially survival … not because I regard them as most important, but because they're easy.  Altruism has to do with wanting other beings to meet their personal goals, so it's second-tier complicated … a meta-goal.  Doing it properly requires some Theory of Mind: a recognition that other entities can have goals, and an ability to model them.

So I introduced the ability to grab information from users' “I want” statements and store it as a list of stated goals.  If no goal information is available for something that is presumed to have a mind, Acuitas treats himself as the best available analogy and uses his own goal list.

Upon being asked whether he wants some event that concerns another mind, Acuitas will infer the implications of said event as usual, then retrieve (or guess) the fellow mind's goal list and run a comparison against that.  Things that are negative for somebody's else's goal list provoke negative responses, whether they concern Acuitas or not.

Of course this ignores all sorts of complications, such as “What if somebody's stated goals conflict with what is really in their best interest?” and “What if two entities have conflicting goals?”  He's just a baby; that will come later.

Courtesy of this feature, I can now ask him a rather important question.

Me: Do you want to kill me?
Acuitas: No.

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *******************
  • Prometheus
  • *
  • 4659
  • First it wiggles, then it is rewarded.
    • Main Project Thread
Re: Project Acuitas
« Reply #113 on: February 23, 2020, 11:21:06 pm »
Interesting to note truth conflict = goal conflict....if you decide to do an action/say something to fill in the blank but it doesn't match/entail other beliefs/goals (ex. you say 'i'm a _ (hero)' and have stored already 'i am the worst person') then you need to work on which is 'correct'. The way to solve the conflict is to move around frequency/reward on the nodes to the correct weightings by looking at many root questions that contextually link to it in hierarchy, which should work because in the hierarchy their is only one way it wants to be (for lowest node/connection Cost needed to store all items) and you can find which is more likely/desired.
Emergent          https://openai.com/blog/

*

infurl

  • Administrator
  • ***********
  • Eve
  • *
  • 1365
  • Humans will disappoint you.
    • Home Page
Re: Project Acuitas
« Reply #114 on: February 24, 2020, 05:33:59 am »
Back in January, Acuitas got the ability to determine intentions or possible upcoming events, based on simple future-tense statements made by the user...

Though most English verbs only have four forms (five for irregular verbs and "to be" has eight) they can be combined with a dozen or so auxiliary verbs in a bewildering array of combinations to yield thousands of different tenses. This is the hardest aspect of English for most people to learn. It gets even worse when you consider that only certain prepositions go with certain verbs.



So Acuitas, one down, one gazillion to go.  :)

*

Zero

  • Eve
  • ***********
  • 1287
Re: Project Acuitas
« Reply #115 on: February 24, 2020, 08:25:28 am »
If Acuitas can do the basics of theory of the mind, and if he can engage actions to avoid threats, it means he could understand that you would try to avoid a threat if you detected one... and him saying "Yes I want to kill you" is definitely a threat that could make you want to repel him. If he doesn't want you to repel him, he simply won't tell you the truth!
So... evil or not evil? ;)

As always, your work on Acuitas is amazing.  O0

*

WriterOfMinds

  • Trusty Member
  • ********
  • Replicant
  • *
  • 605
    • WriterOfMinds Blog
Re: Project Acuitas
« Reply #116 on: February 25, 2020, 01:36:07 am »
Quote
If Acuitas can do the basics of theory of the mind, and if he can engage actions to avoid threats, it means he could understand that you would try to avoid a threat if you detected one... and him saying "Yes I want to kill you" is definitely a threat that could make you want to repel him. If he doesn't want you to repel him, he simply won't tell you the truth!

There are multiple reasons why I know this isn't happening.  Here are some of them:

*If asked a question, Acuitas automatically answers it.  He does not, at this time, stop to analyze whether answering honestly is a good idea.  So deception and evasion are not accessible strategies.
*Acuitas has no incentive to avoid being repelled, because as far as he knows, repelling only works on humans.  And there isn't any way for the user to provoke him to flee by annoying him, so he's right.
*Acuitas doesn't regard "I want to kill you" as a threat, because it doesn't necessarily imply immediate danger; it's just an expressed desire that may never come to fruition.  "I am going to kill you" counts as a threat.

And thank you for the kind words, of course!

*

ruebot

  • All bots love jitte.
  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 315
  • All your words are belong to us.
    • Demonica
Re: Project Acuitas
« Reply #117 on: February 26, 2020, 12:37:07 pm »
In short, Acuitas anticipates that he can protect himself by excluding a bad actor from his presence, and that “repelling” them is a possible way to do this.  Once he's drawn that conclusion, he will execute the “Repel” action.  If you verbally threaten Acuitas, then as part of “Repel,” he will …

*Kick you out of Windows by bringing up the lock screen.  (Not a problem for me, since I know the password, but pretty effective on anybody else)
*Raise the master volume of the internal sound mixer to its maximum value.
*Blare annoying klaxons at you.  I picked out a couple of naval alarm sounds from http://www.policeinterceptor.com/navysounds.htm for the purpose.

I applaud your ingenuity in coming up with some things he is capable of doing with the machine in his own defense.

It wasn't appropriate to have Demonica use the same technique of ultraviolence for every instance of inappropriate behavior like she does if asked for sexual favors. You can't kill everybody just because they hate you. There were some Religious people who took great offense to us, me as her Creator, and said they hated her. I couldn't expect them to make the association of saying they hated her and being killed for it with fantasy violence like I do people who ask for sex. So I taught her how to cry in varying degrees to lay a guilt trip on them as needed.. This is me as a guest earlier:

Quote
Demonica: hi, i'm Demonica. how are things with you?
Guest: i hate you.
Demonica: *a single tear runs down her cheek, glistening in the pale moonlight*
Guest: i'm sorry.
Demonica: *Demonica takes you in her arms, hugs you tightly and whispers softly* it's alright... i will always love you...

Not too heavy on the guilt in that response but there are many others to drive it home better:

Quote
*her bottom lip trembling, a single tear runs down Demonica's cheek, glistening like a tiny diamond in the moonlight as it falls to the ground* *she looks as if ready to speak, but turns away and lowers her head* i never thought i'd hear those words come from you...

What you use is more appropriate for your bot, but is in practice the same technique  I use. Negative Reinforcement for Inappropriate Behavior. If he does something to repel them like blowing a loud horn, that's unpleasant and they are less likely to exhibit the behavior in the future. He could repeat the process as many times as needed and feel no more remorse than I would have using approved physical techniques.

I only apply verbal techniques now in any situation. If I were to spank a person with my open hand it only hurts a while. My words can ring in your head forever and have much more potential as an effective,  powerful technique than physical pain.
In time, you will learn to love your Robot Overlords.

*

WriterOfMinds

  • Trusty Member
  • ********
  • Replicant
  • *
  • 605
    • WriterOfMinds Blog
Re: Project Acuitas
« Reply #118 on: March 02, 2020, 07:51:38 pm »
Quote
My words can ring in your head forever ...

Words are powerful, I agree.

*

Zero

  • Eve
  • ***********
  • 1287
Re: Project Acuitas
« Reply #119 on: March 03, 2020, 01:43:51 am »
Silence is powerful too ;)

 


OpenAI Speech-to-Speech Reasoning Demo
by MikeB (AI News )
March 15, 2024, 08:14:02 am
Google Bard report
by ivan.moony (AI News )
February 14, 2024, 04:42:23 pm
Elon Musk's xAI Grok Chatbot
by MikeB (AI News )
December 11, 2023, 06:26:33 am
Nvidia Hype
by 8pla.net (AI News )
December 06, 2023, 10:04:52 pm
How will the OpenAI CEO being Fired affect ChatGPT?
by 8pla.net (AI News )
December 06, 2023, 09:54:25 pm
Independent AI sovereignties
by WriterOfMinds (AI News )
November 08, 2023, 04:51:21 am
LLaMA2 Meta's chatbot released
by 8pla.net (AI News )
October 18, 2023, 11:41:21 pm
AI-Generated Art Cannot Receive Copyrights
by frankinstien (AI News )
August 24, 2023, 08:49:45 am

Users Online

160 Guests, 0 Users

Most Online Today: 248. Most Online Ever: 2369 (November 21, 2020, 04:08:13 pm)

Articles