Interest in Verbal Working Group?

[Martin Triplett]

I wanted to find out how many people here are building verbal robots, are planning to, or have enough interest in the topic to gather together every few weeks to discuss verbal robotics issues.

I personally would be interested in a verbal sub-group to discuss and present short walkthroughs of how particular verbal problems can be solved.

Topics like:

1. Basic Verbal Feature Extraction using NLP
2. Chunking, Lemmas, Tenses, Plurals, Entities, Pronouns, Toolkits, etc.
3. Determining intent and routing to appropriate agents for processing
4. Question Answering Using Transformers (General, Personal, Contextual)
5. Dialog Interaction
6. Maintaining and Changing Topics
7. How to build Reactions to Answers of Questions
8. Curiosity, Initiative, Sentiment, Empathy, Ambiguity
9. Maintaining Personal Data & Adapting to People Individually
10. Off-the-Shelf Models, Transformers, etc.
11. Multi-Language Grammar & Issues
12. Building Narratives, or executing them.
13. How to form narratives from relational or API data.
14. Getting a bot to form sentences and conjugate any verb
15. Getting a bot to contribute to a conversation...having something relevant to say.

I would be happy to share any of the techniques I have used over the last 10 years of building talking bots and NL based software.  If we could find at least 4 or more people who are interested in active discussion on this, I think it could be fun and interesting and new ideas would come out of it.

I am thinking a zoom meeting every 2-3 weeks might be a place to start, perhaps we can record them too.  People could vote on topics ahead of time to make sure we focused on topics that most people were interested in.  

I don't see a lot of discussion here on verbal issues so this might not be the right group, but I sure hope it is!   I could also go for a working group on vision or movement if anyone is up for that.

Regards,

Martin

[Dan]

I'm in...Let me know.

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/aaac9063-45a3-4267-9d12-e541e6d47dfcn%40googlegroups.com.

[Gmail]

Sure. I am in. I have been using that technology for years. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 1, 2022, at 9:30 AM, 'Dan' via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1337261565.2718497.1659371444505%40mail.yahoo.com.

[Scott Monaghan]

I'm very much in.

Scott Monaghan

413.244.9344

scott.m...@gmail.com

www.linkedin.com/in/scott-monaghan

Resume: https://tinyurl.com/yxhq5eul

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/3C67FA7A-47D1-463D-8A81-B81DE3934F98%40gmail.com.

[James H Phelan]

I'm in!  This is way beyond me at present.  Still struggling to apply Linorobot to the Open Source Rover. 

But, ultimately, it's something I'd want any "real" robot to do - be able to converse.

Be happy to sit in the back row and observe.

Jim

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

[Martin Triplett]

That's really great, we have at least 5!

I guess we need to decide on a time window / day of the week that works for people.  I am in CST here but am very flexible on time and day of week.  I would vote for a weekday and 75 mins or less.  We could just do Wednesdays other than the ones when HB is having the monthly meeting.

Perhaps we could use the first meeting just to get to know everyone a bit, share a bit about our backgrounds and what we would like to get out of this.  We could discuss everyone's opinions on meeting format, length, initial topics, etc.

I would guess topic selection/speakers for the next meetings will spill right out of that, if we even need them.  We could do presos if people want, or just ask each person to come with something short they want to talk about or discuss/problem solve on.  I personally want to avoid a format that involves people feeling like they need to prepare much.  Maybe we can just keep it interactive QA / problem solving / discussion oriented.  I don't know.  Thoughts?

Regards,

Martin

[Michael Wimble]

I’m interested though I haven’t yet looked into this at all. I’m a reasonably quick learner, though. Zoom this Wednesday (tomorrow)? 7 pm pacific ?

On Aug 2, 2022, at 1:21 PM, Martin Triplett <martint...@gmail.com> wrote:

That's really great, we have at least 5!

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/a922d05c-1b94-4b93-8ccd-a50566843cc1n%40googlegroups.com.

[Gmail]

I can meet Monday, Thursday or Friday. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 2, 2022, at 1:26 PM, Michael Wimble <mwi...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/096C4EEB-9731-4852-8615-79DFD206BC0C%40gmail.com.

[Gmail]

Or Wednesdays when HBRC is not meeting. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 2, 2022, at 1:21 PM, Martin Triplett <martint...@gmail.com> wrote:

That's really great, we have at least 5!

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/a922d05c-1b94-4b93-8ccd-a50566843cc1n%40googlegroups.com.

[Scott Monaghan]

In general I can do any evening except for Mondays.

Scott Monaghan

413.244.9344

scott.m...@gmail.com

www.linkedin.com/in/scott-monaghan

Resume: https://tinyurl.com/yxhq5eul

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/D359AC53-D5D2-4AFC-8DCC-BA5A3AA3F0F6%40gmail.com.

[Stephen Williams]

I'm interested also.  I am sporadically available, depending on work mostly.

Stephen

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMzUCWNLdn-P-2nPyOMsHgUpuZEv8P9Z7F6Jn2Nh%2BR_LKDNu5w%40mail.gmail.com.

[Martin Triplett]

How's Wednesday (or Thursday) of next week?  I hate doing things on such short notice and losing somebody.  It gives us time to see if more people are interested too.

Can we use HB's Zoom meeting or do we need to use something else?

[James H Phelan]

Great!

I'm in CST/CDT too. Will be nice to not have to leave a meeting early because it's past my bedtime!

Since I'm totally ignorant of the subject, I'd like, after introductions etc., to start from the very basics -

What's a verbal robot?

What are the sub-functions required to become one?

What are the various techniques to achieve each function, their pros & cons?

What are common buzzwords and acronyms and what do they mean?

Give examples of systems for verbal robots from the most rudimentary to the most sophisticated and how they work.

J

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/a922d05c-1b94-4b93-8ccd-a50566843cc1n%40googlegroups.com.

[James H Phelan]

Wed ok /p 8:00 CDT or Thurs is ok

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/b9f6f108-ead2-4bae-91d9-6f8dbc1a874bn%40googlegroups.com.

[Dan]

Wed or Thursday works for me.

-----Original Message-----
From: Martin Triplett <martint...@gmail.com>
To: HomeBrew Robotics Club <hbrob...@googlegroups.com>

https://groups.google.com/d/msgid/hbrobotics/b9f6f108-ead2-4bae-91d9-6f8dbc1a874bn%40googlegroups.com

.

[thomas...@gmail.com]

What's a verbal robot?

As far as I know, it is one that receives instructions verbally and replies verbally—unlike a vacuum cleaner robot.

“Robot, clean the floor”

“At your service sir! Floor cleaning has begun!”

What are the sub-functions required to become one?

It all depends on your software and hardware and if you want your code local or you will use cloud services. I use Bing’s cloud service for turning audio into text strings. Windows has a (very bad) local one.  There are a number available for ROS. There are standalone electronic boards that do just that: turn sound into text/commands.

What are the various techniques to achieve each function, their pros & cons?

Good for a number of topics of discussion. And of course, we overlap with some of AI if it is UNDERSTANDING we want and not just dictation for our input.

What are common buzzwords and acronyms and what do they mean?

I’ll start the list. In no particular order:

1.       Voice I/O: Voice input and output.

2.       NLP: natural language processing

3.       ASR

4.       STT

5.       Wake word: the word or phrase that tells the system to start listening.

6.       Prosody

7.       TTS

8.       DTW

9.       SSML

10.   Deepfake

11.   Speech processing

12.   DNN

13.   hidden Markov models

14.   Speaker identification

15.   Speaker dependent vs independent

16.   Verbal Feature Extraction

17.   Chunking,

18.   Lemmas,

19.   Entities,

20.   AIML

21.   Category

22.   Pattern

23.   Template

24.   Determining intent and routing to appropriate agents for processing

25.   Transformers

26.   LLM: Large language model

27.   GPT

28.   Bot

29.   Turing Test

30.   Loebner Prize

31.   Chatbot (chatterbot)

Give examples of systems for verbal robots from the most rudimentary to the most sophisticated and how they work.

1. A circuit with a microphone that can indicate when there is a sound or there is silence (most simple)
2. A toy that responds to a limited set of vocal commands. (less simple)
3. An agent like Alexa or Siri or Cortana that get their input verbally and respond verbally to simple inquiries or commands, and perhaps perform simple tasks based on those commands.
4. An AI that can hold an interesting conversation with you verbally.  (most complex)

 

 

From: hbrob...@googlegroups.com <hbrob...@googlegroups.com> On Behalf Of James H Phelan
Sent: Tuesday, August 2, 2022 7:26 PM
To: hbrob...@googlegroups.com; Roberto Pensotti <rpen...@gmail.com>
Subject: Re: [HBRobotics] Interest in Verbal Working Group?

 

Great!

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/a6c6a6cd-88fa-0de8-5899-0cdd7f424ab9%40hal-pc.org.

[Gmail]

I have built several. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 1, 2022, at 9:01 AM, Martin Triplett <martint...@gmail.com> wrote:

I wanted to find out how many people here are building verbal robots, are planning to, or have enough interest in the topic to gather together every few weeks to discuss verbal robotics issues.

[James H Phelan]

To break it down even further

(I'm a physician so I think in terms of anatomy and physiology: population, organism, organ systems, organs, tissues, cells, intracellular processes, biochemistry, chemistry, physics).

Each of these I'm sure could take up a whole meeting.

Hearing [Microphone -> analog>digital converter, directionality, localization, sound vs speech interpretation] 

Listening [speech-to-text -> text interpretation ]

Understanding [interpreting text in context of situation (memory, knowledge (personal & cultural), other senses) and physical reality (body, location, time, environment, capabilities)]

Thinking [applying understood concept to choosing an action]

Responding [speaking, moving, indicators, etc]

Speaking [digital-to-analog converter -> speaker]

Moving [motors, manipulators, etc.]

 

Where the process happens:

"At the edge" completely in the robot itself.  An Individual.   "No man is an island, entire of itself; every man is a piece of the continent."  John Donne

"Hosted" shared between the robot and a "host" computer (laptop?) that assumes the more challenging processes. [do I have this terminology right?]

A Family, parent/child relationship.

"Cloud" shared between the robot, +/- a local computer host, and a remote service in "the Cloud"

A Community relationship.

How trustworthy & safe is the Cloud?  (Are you paranoid if people really are spying on you?)

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/006f01d8a6fd%24e049f6b0%24a0dde410%24%40gmail.com.

[Martin Triplett]

I often think of all communication (whether with people or robots) in terms of playing an amateur game of tennis at a park where others are playing on adjacent courts too.

Someone serves (takes initiative and says something) ...another attempts to return serve (responds).  Occasionally random balls fly in from other courts.

For so many robots, the person always serves and the robot always returns, and all the points end there with 2 hits.

In a real match, the server would react to the robots return and the point would continue and the context (game/set/match) would get more involved (if the robot was built for that).  To me, this is where things get a lot more interesting (longer points with more hits and more contextual reactions).  Also, the robot should decide to serve sometimes...intelligently.

I often think about the following example where person 1 has several questions they want to ask person 2.  Also, person 1 already knows than person 2 is an adult male.

Person #1:  How tall are you?

Person #2:  7 foot 4.

If person 1 moves on and asks another question at this point with no reaction to the last answer, then they are either not really listening, in a terrible hurry, or just being socially dumb.  7 foot 4 is a highly unusual answer, several standard deviations from the mean for adult men...any normal human would have a reaction to that answer.  If person 2 had said 5 foot 11, moving on to the next question would have been fairly normal (not if person 2 was female though).  If person 2 had said 6 foot 4, perhaps a quick reaction and move to next question all in one step..."Wow, that's tall.  How old are you?"

Hit #3 is where the real fun starts.  The first 2 hits are much easier to program for.

IMO, when a robot asks questions, it needs to have some kind of reaction function specific to the question being asked.  If you ask a yes/no question, you have a reaction function that attempts to discern yes/no/I don't know.   "Sure" or "Definitely" should be taken as a yes.    If the question is a numeric question, like "How tall are you?", the robot perhaps needs to have a reaction function that discerns values and evaluates them against mean and standard deviation by gender.

To me, we can discuss all kinds of technologies, but it still comes down to playing a good game of amateur tennis (or some other metaphor you guys choose).  If you don't design the bot to play the game, the bot is going to remain rather uninteresting verbally...IMO.

This is the kind of stuff I used to lose a lot of sleep over when I was doing my Anna project some years ago.

[Gmail]

Interesting perspective. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 3, 2022, at 3:25 AM, James H Phelan <jhph...@hal-pc.org> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/001bf5bb-e013-bb20-7b8a-6f3a67c2c85f%40hal-pc.org.

[Martin Triplett]

Thomas, I was excited that you posted your list of verbal concepts earlier, as I had to look up some of the acronyms and learn about some new concepts.  I still have a few more to look up.

I think that is what I am hoping to get out of the group, interesting discussion, motivation, and especially some new ideas.  Maybe I will come out of this and not view things in my tennis bubble but evolve into some new concept.

Cheers,

Martin

You received this message because you are subscribed to a topic in the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/hbrobotics/ncOUmjp0Y8E/unsubscribe.
To unsubscribe from this group and all its topics, send an email to hbrobotics+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/9150A2E3-2AB6-4DB8-BF22-DCD8A070CC0E%40gmail.com.

[Chris Albertson]

On Tue, Aug 2, 2022 at 10:57 PM <thomas...@gmail.com> wrote:

What's a verbal robot?

As far as I know, it is one that receives instructions verbally and replies verbally—unlike a vacuum cleaner robot.

“Robot, clean the floor”

This is the simplest example.   But I think many people would like the robot to accept commands like this

 "Robbie, I want to make Fish Tacos for dinner, see what is in the fridge and place an Instacart order for what we don't have" Then when the order is delivered Robbie fetches it from the porch and puts the groceries away and tells the person "It's done".

I always like to look at the far-future goal so my incremental baby steps are headed in the right direction.   Because you likely can make a robot that obeys a "Clean the Floor" order using technology that could never possibly be extended to the robust butler who can place Instacart orders.

So I like to think about how I might build the butler/housekeeper.  I see I have no hope of doing this in one lifetime, but I can select some interesting part of the project and do that.  But at least I know I am chipping away at the larger problem.

So as a far goal, fetching a beer is too simple.  The robot should know the beer goes with the fish taco and order the beer for delivery, then place it in the fridge.  It knows this because it was told, not because it was hand-coded in Python.

--

Chris Albertson
Redondo Beach, California

[Gmail]

If the robot has to know that the beer goes with the fish, that the milk goes with the cookies, what a phone is, how to use it, what a restaurant is, etc., we are talking about a incomprehensibly huge database—one that would take decades to populate. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 3, 2022, at 10:22 AM, Chris Albertson <alberts...@gmail.com> wrote:



--

You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CABbxVHuQVQi8i4rjqgk6ykdqfbmi_Rwv9NbGYfXX1ye3KtCnbg%40mail.gmail.com.

[Chris Albertson]

No, not decades because there is more than one robot.

For example, Tesla might need 10 million hours of driving experience data to build a full automatic driver.  But they have 100,000 cars on the road every day and can collect 10 million hours of data every week.

The other way to populate a cooking database is to read cookbooks.  Robbie the robot can read every cookbook in the library in a few hours and then every copy of Robbie in every home knows 100 ways to make a Fish Taco and 30 ways to cook pasta.   

What if your Robot got started reading all of Wikipedia?  You say "All of Wikipedia!  That is HUGE".  Not really.  The entire thing fits on one SD card.  It is only 35GB. and 1000 cookbooks is about then 1GB.   Today in 2022 a 1000GB disk drive costs just over $100.    All of the Wikki plus 1000 books would leave the disk 90% empty.   I'm not suggesting that anyone here download the Library of Congress to their robot, but technically the entire LOC would fit on one disk drive. (LOC is about 15TB in digital form)

But even if it would fit, you would never load so much information to the robot.  Better to give it an Internet connection and teach it to use Google.

This is what I meant by thinking ahead.   Once you realize that translating 100 cookbooks to AIML by hand would take 500 years, you think that maybe AIML is not the way to go.  You have to teach your robot to read.

Next, you find you need to teach the robot to forget.  Knowing too much means the time to search the local database would take too long.  The robot needs to know how ti fill the database from the Internet or its camera and also what to purge.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/0632DCEC-4826-4940-A3AE-FD680A780D60%40gmail.com.

[Gmail]

Camp,

Can we get your blessings on a new official HBRC SIG?

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 3, 2022, at 10:20 AM, Martin Triplett <martint...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy2917dq7rx02a-YrnTxAeZcqW973gN_peYR39X8us1s0g%40mail.gmail.com.

[Gmail]

Martin,

I talked to Camp and the others at the business meeting this evening and they gave us the thumbs up. We are now official: the “Verbal I/O SIG”. We will meet weekly. 

Our topics will include speech recognition, text to speech, and chatbots, and AI as it relates to chatbots. 

I am scheduling our first meeting on 

Monday, August 8th at 7 pm Pacific time. The format  will follow the model of the ROS meetings. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 3, 2022, at 10:20 AM, Martin Triplett <martint...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy2917dq7rx02a-YrnTxAeZcqW973gN_peYR39X8us1s0g%40mail.gmail.com.

[Scott Monaghan]

Hi Thomas, I can’t normally do mondays. 

Martin was talking about doing once per month on Wednesdays.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/6C20FD6E-686D-46C9-84C9-7D0777314963%40gmail.com.

--

[Gmail]

I prefer not overlapping the other meetings if at all possible. 

Ok. Let’s try this again. For the folks wanting to attend our meetings, what days and times are you available?

Name: Thomas Messerschmidt

Preferred days/times: Su M Th F - 7 PM Pacific time. 

Unavailable: Tuesdays and days when HBRC meets. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 3, 2022, at 9:11 PM, Scott Monaghan <scott.m...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMzUCWMBFB8rYtpkc16bLeV-zvEZXzZ6QchhD1q2Vsn-WXukEQ%40mail.gmail.com.

[Martin Triplett]

That's great Thomas.  I can make it for any night next week for the first meeting.

I propose we discuss future meetings and preferences then.  Hopefully we can get some quick consensus and adapt.

Like Scott, I too would prefer not to meet on Mondays on an ongoing basis. 

Regards,

Martin

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1927B815-86CB-4DC4-B713-4D0E27BAF24D%40gmail.com.

[Scott Monaghan]

This coming week and generally I am available on any Sunday-Thurs evening, except Mondays.

Fri & Sat nights are often plans with fam

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy22uGhhypZPk8ffw7gPXvenLw87Ybr2GJmMLD58vE%3Do4Q%40mail.gmail.com.

[James H Phelan]

Any night not conflicting with other HBRC meetings.

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1927B815-86CB-4DC4-B713-4D0E27BAF24D%40gmail.com.

[Michael Wimble]

Do I see a convergence on Thursday’s at 7 pacific?

On Aug 3, 2022, at 10:26 PM, Gmail <thomas...@gmail.com> wrote:

I prefer not overlapping the other meetings if at all possible. 

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1927B815-86CB-4DC4-B713-4D0E27BAF24D%40gmail.com.

[Martin Triplett]

I think so Michael.  Let's go for it if there are no objections.  

Thomas, are you going to take lead on finalizing the time and sending out the meeting invite and all that?  I don't really know what to do to execute on that.

I can organize some kind of loose agenda and moderate if necessary (unless someone else wants to) for the first meeting.  I am hoping meeting #1 will simply give us a chance to meet each other, share our respective goals and backgrounds a bit, and build some consensus on how to proceed for the next few meetings, and some initial topic selection.  If anyone wants to achieve more than that, please let me know.  

Hopefully meeting #2 can follow shortly with some real topics.  I would love it if we did an AIML dedicated meeting early as a basis for later meetings.

For future meetings, perhaps someone else can moderate as I have a lot to say on NLP, so it's probably better if someone else is running the meetings and shutting me up when it's time.  :-)

I am looking forward to getting started.

Cheers,

Martin

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/DA1FAF87-5D80-4C53-94AD-DC5FC1FBEF25%40gmail.com.

[Michael Wimble]

Could you, Martin, give us an overview of the field. Or point to some good online resources to review for background?

On Aug 5, 2022, at 8:31 AM, Martin Triplett <martint...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy2LeDz97Ax6vKEd25muoCu4KUZ_0G3PMg6R1ADhvWu5Uw%40mail.gmail.com.

[Martin Triplett]

Sure Michael.  Let me think a bit and I'll get something out.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/BE34A6F8-8146-480F-A343-25B65ED7A8DA%40gmail.com.

[Scott Monaghan]

Are we on for our first meeting on Thursday, 11 Aug at 7pm PT?

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy14s_jKARuUsLcg9bQaahGgFjcGR0bLuUGgwnnASrgPEA%40mail.gmail.com.

[Gmail]

Yes. That will be our first meeting.

Agenda:

1. Introductions

2. Discus the meeting format

3. Create a list of discussion topics 

4. Open discussion 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 5, 2022, at 10:26 AM, Scott Monaghan <scott.m...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMzUCWMyry9igsem-b_F%2BaK2vp9Tp-0FbEb0dOQmN6rx_UE_aQ%40mail.gmail.com.

[James H Phelan]

OK w/ me

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/DA1FAF87-5D80-4C53-94AD-DC5FC1FBEF25%40gmail.com.

[Gmail]

Sure. I will finalize the time and send out the meeting invite.

Camp,

Are you going to participate? And can you share the Zoom account with us?

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 5, 2022, at 8:31 AM, Martin Triplett <martint...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy2LeDz97Ax6vKEd25muoCu4KUZ_0G3PMg6R1ADhvWu5Uw%40mail.gmail.com.

[James H Phelan]

YES

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMzUCWMyry9igsem-b_F%2BaK2vp9Tp-0FbEb0dOQmN6rx_UE_aQ%40mail.gmail.com.

[Gmail]

 👍 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 5, 2022, at 7:56 PM, James H Phelan <jhph...@hal-pc.org> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/bfab95f1-1127-9334-71ce-f522c1591cdc%40hal-pc.org.

[Martin Triplett]

Michael asked me to provide an overview of the area.  There are so many ways to do it, and most lose any real meaning the higher level they get.  I might suggest glancing at the human robot interaction section on the robotics wiki page if you want a high level narrative form overview.  

My overview focuses on questions a builder might want to consider when building a verbal bot.   This list is by no means all inclusive, but does cover all the major issues that come up in a back and forth dialog (tennis metaphor) with a person and a robot.  I hope we will find time to share, discuss, and debate solutions for most of these questions over time.

Inputs:
How can I get a robot to listen and reliably translate audible input into text?

Processing:
What are different ways to get a robot to process verbal input?
What is a basic reusable and extensible process for processing speech (something that can handle simple and complex stuff) ?
How can I determine intent?
How is chunking useful?
How can I calculate NLP parse trees, grammatical structure, nouns, verbs, tense, entities, values, pronoun meaning, sentiment, etc.?
How can I route estimated intent to perform various actions in a brain?
How can I use various off-the-shelf NLP libraries (like spacy or nltk)?
How can I use various off-the-shelf models (like transformers, squad, dialog, text gen, sentiment, etc.)?
How can I use various online data sources and APIs (like Wolfram, Wikipedia, ConceptNet, RDF, etc.)?

How can I deal with synonyms? singulars and plurals?
How can a robot maintain a concept of topic(s) and other context over time? or change the topic?
What kinds of data structures/object models are useful?
How can input speech make any possible set of internal state changes within a robot?
What are different ways to get a robot to answer questions?  When using a model, how can I build a corpus?  What are some likely sources?
What are different ways to get a robot to perform some logical reasoning?
What are different ways to get a robot to ask questions autonomously?
What are different ways to get a robot to listen for answers and classify them?
What are different ways to get a robot to react to classified answers?
What are different ways to get a robot to make statements autonomously?

How can a robot decide when NOT to talk, when to lose interest, or when to stop being annoying?
What are different ways to get a robot to have some personality?
What are different ways to get a robot to have and show some emotions?
What are different ways to get a robot to assess my emotions?
What are different ways to get a robot to remember all its interactions and learned information?
How can thought/responses lead to coordinated actions?
What are narratives and how can they be used?
What are some different ways to aggregate and reuse complex behaviors using NLP?
How can I determine what thought wins?
In the end, how can I get a robot to act?

Outputs:
What are the different libraries to get a robot to speak?
How can I coordinate speech with phonemes, visemes, lip movements, eyes, facial expressions, physical gesturing, internal state changes, etc.?
___________________________

A Final Note

I broke all this into 3 areas (inputs, processing, outputs) for some deliberate reasons, so our discussions can perhaps map easily to thinking in terms of various functions in a brain, whether they are code based, neural nets, off-the-shelf models, rules based, fuzzy logic, or prolog based implementations.  For me, all these techniques and functions should be able to play together in one brain interchangeably and exchange data.   Each technique is useful in different contexts.

I look forward to seeing everyone tomorrow.

Regards,

Martin

[Gmail]

Thanks Martin. Looks great!

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 10, 2022, at 4:54 PM, Martin Triplett <martint...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/32786e11-d37b-4028-8f9c-67812c803fb9n%40googlegroups.com.

[Gmail]

Martin,

If you get this before the meeting please call me. I want to talk over a couple things with you. I know it’s late, so if not we’ll just wing it.

213-810-2589

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Aug 4, 2022, at 7:52 AM, Martin Triplett <martint...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAMZ%3Dfy22uGhhypZPk8ffw7gPXvenLw87Ybr2GJmMLD58vE%3Do4Q%40mail.gmail.com.

[James Salsman]

I'd encourage anyone with an interest in this thread to study the Dual Process nature of LaMDA: https://arxiv.org/pdf/2201.08239.pdf (especially pp.5-6.) In particular its symbolist database, calculator, and translations subsystems: https://towardsdatascience.com/why-gpt-wont-tell-you-the-truth-301b48434c2c

LaMDA is tuned on nine unique performance metrics, almost all of which its predecessors were not: Sensibleness, Specificity, Interestingness, Safety, Groundedness, Informativeness, Citation accuracy, Helpfulness, and Role consistency. Sensibleness tuning entails whether responses contradict anything said earlier, by "pre-conditioning" each dialog turn via prepending prior dialog interactions, on a user-by-user basis. Blake Lemoine claimed such tuning involved years of prior dialog, which has not been confirmed by Google. As most people reading this probably know, Google fired Blake. Google has a wait-list for LaMDA access here: https://aitestkitchen.withgoogle.com/

If I were to name one expert who would best be able to explain how to reliably connect large NLP foundation models to physical robotics equipment, at this point I would choose Rachael Tatman https://www.linkedin.com/in/rachael-tatman-500a323a/ but I would also make Kathleen Kreel's recent video required watching: https://www.youtube.com/watch?v=XAts7oViVlg&ab_channel=StanfordHAI

Facebook also has a dual process chatbot/text NLP transformer, but they remain behind Wolfram's group at Google. I agree with the comments confirming sentience, sapience, and consciousness are relatively trivial compared to the stage we are at now, and I would counter discussion of whether such attributes apply by suggesting examining how such chatbots, given support and autonomy, would choose to edit Wikipedia instead: https://old.reddit.com/r/MachineLearning/comments/vgtydo/d_two_flaws_in_discussions_surrounding_the_recent/

Best regards,

Jim

Mtn. View

[Martin Triplett]

Thanks for sharing the links Jim.  I need to pencil time in my schedule to watch and study ALL of it.

In the gaming world, people build 3D humanoid character models that have all the major humanoid joints, so other people can build software to take control of an instance of the model at runtime (as a character in a game), without having to have any specific knowledge of the specific 3D models being used.  This allows AIs to be swapped out for a character, or models to be swapped out for an AI.  2 good things.  Interchangeable parts.

Perhaps that is how some of these transformer (or other types) models could be trained to become more of an embodied intelligence and be able to move in a real or virtual space...with a standard set of joints and other state variables.  Without some standard representations of embodiment (and perception) ...it's hard to see how the models can jump the boundary from interesting chatbot curiosity into the wider brain and robotics realm.  The current state of affairs is that the big models are disconnected from all this, so the relevance of their outputs to embodied intelligence is tremendously constrained.

To me, one of the standard embodiments we need is something with 2-motor differential drive, optional arms, and a pan and tilt head.  I don't think walking robots are about to become standard.  Maybe ground movement should just be simplified down to a "goal heading" and "drive state" (stop, forward on a heading, rotate on a heading, reverse on a heading)  That's about how I do it.  Goal heading could be a 0-1 output from a NN, drive states could be one-hot outputs from a NN.  If people didn't want to use goal headings and a compass, I guess a model could output 0-1 values for each motor that result in steering.  It gets hard to standardize embodiment as soon as you make allowances for different sensors and motor/wheel configs (standard, mecanum, omni)  Perhaps the same core model needs to be retrained on multiple standard body configs.

While I could do all this on my own bot easily enough (get a NN model to control movement) ...I haven't done so.  I experimented with it briefly by using a custom NN trained on a custom dataset, to prove that I could map values into a NN and back out to the brain to create the movements.  It worked to the extent I tried it.  I also tried similar using fuzzy logic.  Same concept really, just one is a NN with its "rules" expressed in training data and one is a fuzzy engine with rules expressed in a simplified English format.    I shied away from using either of these for my bot in a general sense as most of the time I want the robot to be still and remain in my control most of the time...responding to verbal input and having limited autonomy in the context of whatever I am asking it to do.  Its hard for me to conceptualize what "full autonomy" would mean as the default behavior in my house without imagining bad/expensive outcomes.  I suppose my fear keeps my bots from getting better at being embodied.  I need to build a small cheap full-auto bot that can learn from mistakes without breaking stuff.

[James Salsman]

Hi Martin,

Thanks for your reply:

>... how some of these transformer (or other types) models could be trained to become more of an embodied intelligence and be able to move in a real or virtual space...with a standard set of joints and other state variables.  Without some standard representations of embodiment (and perception) ...it's hard to see how the models can jump the boundary from interesting chatbot curiosity into the wider brain and robotics realm.  The current state of affairs is that the big models are disconnected from all this, so the relevance of their outputs to embodied intelligence is tremendously constrained.

It's useful to think of chatbot I/O as similar to how LaMDA and Facebook's dual-process chatbots query their attached databases or calculators etc., and then post-process the returned output before replying to the user.

In practice this looks like distinctive control tokens in the I/O stream, where the chatbot might say, "Good morning, Jim, I've started the coffee machine," upon waking up to an alarm, but the actual chatbot dialog I/O might be e.g.:

In: +ALARM 6:30 a.m.

Out: +KLAXXON on.

In: +KLAXXON cancelled by user Jim.

Out: +COFFEEMACHINE on. Good morning, Jim, I've started the coffee machine.

But again, Rachael Tatman and either of the dual process chatbots teams are the real experts on how to condition chatbots to reliably do these kinds of more formal I/O operations. I think Rachael has a tech report on it, but it doesn't seem to be in her ko-fi offerings.

Best regards,

Jim Salsman

[Martin Triplett]

I watched a fair amount of Kathleen's video, largely on bias.  It is a real problem.  The idea of chat models sending out state changes is interesting too.  I considered self training a transformer to take in sensor data and output commands but never did it.  I am glad people are exploring it in the chat model space as it seems like a path of lower resistance for some research.  

In practice outside academia, I doubt the chat models will become the controlling model of everything else in robots though.  Many other models have benefits too.  As far as actual implementation, I would advocate for a middle layer that would coordinate higher level functions and use multiple models (dialog/chat being one among many) to create output behavior.  I think almost all the actual output behaviors should be definable outside the models and outside any of the code for a brain.

In my own system,  I extract all output behavior to a table, so one or more (usually many) variations for any given condition can exist.  Each variation (for something like "hello") is defined as an executable English narrative, so it can execute any state changes, including movements, with any time delays as required.  This means animated behaviors over some period of time are easy.  The narrative can also suppress other autonomous behaviors while this is happening.  The robot just has to make a few decisions and pick the most relevant output narrative given the context (polite, rude, etc., for men, women, young, old, verbal, non-verbal, serious, joking)  To me, this keeps things more interesting.

I think this separation of output behavior from decision code is very important.  It allows for variation while being easily understood (in English) and easily maintained (if you can type sentences).  It will eventually be able to be much more context aware, although this part is not done yet.  This part will not be difficult.

It is also important that different robots could implement these output behaviors in different ways...without having to touch code.  This means code could be shared.  So can behavior (memories).  Some bots might have lots of variations for some behavior, with context sensitive variants as well.

[James H Phelan]

Avoiding the "Uncanny Valley"

"The robot just has to make a few decisions and pick the most relevant output narrative given the context (polite, rude, etc., for men, women, young, old, verbal, non-verbal, serious, joking)"

I may be in the minority here, but I want my robot to be an appliance not a human substitute. 

I don't want my robot to have a personality any more than I want Windows, Linux, my toaster, my microwave, or especially my car to have a "personality".

I just want them to do what I want whether by turning dials, pushing buttons or being given verbal commands.

Buckminster Fuller said, based on his Navy experience, "It's not so important to be understood as to not be mis-understood."  Which is why all commands are shouted back.

If I make a wisecrack statement to my robot, I don't want it going off trying to do what it thinks I meant.  Especially if I say in frustration "Just shoot me!"

And don't give me the "Three Laws of Robotics".  If you've actually read Assimov, you see where they fail!

We've all experienced it doesn't take long in a conversation with a robot, no matter how sophisticated, to have it go off the rails.

My first programming language was FORTRAN.  When I made a typo or malformed statement it stopped right there and gave back the line number where it crashed.

It was pretty easy to find my mistake and correct it. 

Then came PL1 (Programming Language One).  When it came upon a malformed statement, it tried to interpret what the programmer meant and carry on.

Eventually the mistake would cause a fatal crash somewhere else in the program.  Then try to find where the original error was!

In the case of my offhand comment, I want my robot to stop right there and just say "I'm sorry, I don't understand."

Don't want my robot to have dreams, desires, aspirations, a sense or humor or irony, the ability to deceive or manipulate, moods, or especially an "attitude".

I don't need it to "act human" so that I can "feel comfortable" interacting with it any more than my microwave does.  I just want it to be reliable.

Robodoc

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

"Here am I, the servent of the Lord;
let it be with me, according to your Word"
Luke 1:38

[Gmail]

Why avoid it. I EMBRACE it. 🤖

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Sep 10, 2022, at 12:59 PM, James H Phelan <jhph...@hal-pc.org> wrote:



--

You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/160f2b38-efc0-ef68-d2d5-3b0deaad1b4f%40hal-pc.org.

[David Williams]

Anticipating variation in preferences, a forward-thinking designer will, among many other settings, implement and allow a personalized setting within a range:

Appliance |••••|••••|••••|••••|••••| Human

--

You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/160f2b38-efc0-ef68-d2d5-3b0deaad1b4f%40hal-pc.org.

[Chris Albertson]

Some people study AI for different reasons.  

• Some just want a useful machine, like a self-driving car or to process bank loan applications, or to cook burgers. We call these people "engineers"
• Some people study AI because they want to understand the nature of intelligence and see if perhaps there are basic principles involved and if consciousness is real or an illusion.  We call these people "sciences" because their aim is to study nature.
• Many actually want to do both. There is synergy in this because understanding the science of cognition does allow you to build machines that are even more useful

We see this in other fields.  Back n the early 1800s, when steam engines were the pinnacle of technology.  Some engineers simply wanted to build better engines, but some studied thermodynamics to understand science.  There was some synergy in this.  Science and engineering push each other along.

My interest is maybe 2/3rds science.   I've build some robots but I always take them apart and re-used the parts because I really have no use for the robot, other than to learn something.

Tese large language models (LLM) are a good example.    I scientific question is if these can actually ever be intelligent.  My conjecture is "No." they are no smaerter thena uplok table.    But a table with 100 billion entries can appear smart

Today in 2022. I'm thinking that something like Transformational Grammar as popularized by Noam Chomsky is going to get us to the Holy Grail.   But it is not easy.   The idea is that we apply a seris of mathematical transformations to language to get semantic content, manipulate the semantic content, then apply some other transformations to convert this to language.     The idea is not new.  It might be thousands of years old.

--

You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/160f2b38-efc0-ef68-d2d5-3b0deaad1b4f%40hal-pc.org.

--

Chris Albertson
Redondo Beach, California

[James Salsman]

I promise I will reply to the questions in this thread soon, but my LaMDA Google AI Test Kitchen access was approved today, and I am having way to much fun interrogating it! https://ibb.co/Xb0MRZp

You received this message because you are subscribed to a topic in the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/hbrobotics/ncOUmjp0Y8E/unsubscribe.
To unsubscribe from this group and all its topics, send an email to hbrobotics+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/D1FE2D58-230C-444A-9B37-A157D5DC6A2D%40gmail.com.

[James Salsman]

... I made this album of my LaMDA screenshots before I ran out of quota. Please excuse the apparently reverse chronological order and redundancy: https://ibb.co/album/0V2Qq7

I have a feeling asking to excuse redundancy will be an important survival strategy in the near future.

Best regards,

Jim Salsman

Mountain View CA 

[Martin Triplett]

Lots of interesting perspectives.  I kind of agree with everyone.

I suppose I would divide the continuum (expressed by David Williams) into 3 categories, for simplicity of discussion.  I pick 3 instead of 2 for a reason I will get to.

1.  Intelligent Moving Appliances (toys, vacuums, cars)

2.  Robots (with or without personality and other interaction capabilities)

       - There are also many robots here (like InMoov) that take a humanoid form, without trying to look like humans.

3.  Systems that are trying to look like humans

while also trying to act like them.

       - Some go so far as to seek to be recognized as sentient (whatever that means), gain rights, citizenship, etc.

As all our various machines (appliances, cars, phones, whatever) get more intelligent, the definition of robot becomes more important to define.  Otherwise, the term robot could become almost meaningless.

Different people have different aspirations for their creations.  Some people feel strongly about what they "don't want" while others aspire for something else or more.  Some people have an "All of the above" approach.  I don't go that far.  I suppose I have my own list of things I don't want in a robot.

My own personal interests lie firmly in category #2 - what I am calling robots.  I think these robots can have many abilities in the areas of communication, personality, whatever...to make interacting with them easier and more intelligent and interesting too.  I like things that are easy to use, intelligent, and interesting.  The remote control for my robots has one button...from which it can do anything.  For me, my concept of the kind of robots I wanted was formed by movies like "Forbidden Planet",  "Star Wars" and many other sci-fi movies.  These robots looked like machines...but still talked and had personalities. This vision, the uncanny valley, and ethical questions (that I will get to) is why I went with robots that still look like machines.  I went with animated faces with eyes and such, but I went with a minimalist approach...trying to be relatable but not trying to look human.  I didn't try to make the robots lose their unique "robotness" as distinct from humans.  

James Phelan brought up the valley, a very relevant area.  Different people have different opinions on where the valley is.  I personally don't put verbal interaction or attempts at personality in the valley category.  I understand that some people do and I think some of my own robots fall into the valley at various times.

For me, the valley is much more of an issue when we try to make robots look like humans (skin, hair, etc).  The original meaning of robot was closely linked to "slave".  Slavery has a very long and dark set of chapters in human history and we are still living in its very tragic aftermath.  Some might argue it has just taken newer and less obvious forms over time.  I know we need machines to do many things.  Working in factories and in the fields is necessary for us to live and eat.  I really hope we don't make these machine (slaves) look and act like us in the future, as our future will look and feel like a reboot of our dystopian past.

Many robots have a gender look to them.  Sexbots are an industry now too, for men and women.  One of the most popular types of robots in the world is battery powered and can be found in many millions of women's nightstands.  When it comes to full size female looking humanoid robots with skin, it's hard not to think about the history and power differential between the genders and the adverse consequences of that.  I fear the dehumanizing effect of what people en masse will feel empowered to do to other machines that look exactly like people. What will that say about us?  Maybe it's the same as video games.  I don't know.  I enjoyed Westworld as much as the next person, but it raises many issues.  It runs a risk of dehumanizing us more...whatever that actually means anymore.  Like any new technology, I think there will be consequences.

For me, all of this touches on aspects of our own human nature and history that are difficult to come to terms with.  That is why I personally stay out of Category #3.  I respect the fact that this is an individual choice that each builder must make.  There is nothing wrong with human-like animatronics, it's just not my thing.  Some people want to build pre-programmed smart appliances.  That is not my thing either.  We all have limited time and must find our own place on the continuum.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CAD4%3DuZYdhdP8WoYN7HirpxU6jRxisYezZcAvGLFekPJmoO7zXg%40mail.gmail.com.