Previous in Forum: Why a 'Language Model' (such as ChatGPT) is not even nearly AI  
Close
Close
Close
14 comments
Guru

Join Date: Feb 2007
Location: Israel
Posts: 2966
Good Answers: 24

How Well Do We Understand Ourselves in Order to Teach a Machine?

05/20/2024 2:21 PM

I happened to be among the first to use ChatGPT (which allows me free entry), and in accordance with my background and curiosity I asked him a lot about the trends and history of expert system development, visual analysis, voice analysis, and learning mechanisms.

.

He blurted out more or less what I expected him to blurt out, but what amazed me was a clear and well-designed ability to articulate complex issues - which, we must admit, is a quality that human beings also seek for themselves, if only to explain to themselves, what they see.

.

<Get to the point laddy, time waits for no one>

.

This ability was invented by a person - the software he used is a unique tool, a mechanism that knows how to manifest an astronomical number of machine states - hence also being a substrate for the wild growth of bugs - a bug is any given machine state encountered in a precedent, a state that the program is not prepared to deal with. This is Godel's great discovery.

.

- as a machine, program or mechanism - a precedent is an immediate stop of any interrupted procedure, even if "stop" is a pre-programmed condition to refer to, in the output of an error message.

.

The computer (including the software that was running on it at that moment) is simply a multi-mode automaton, designed to serve human inventions, but the nature of the service is also planned by a person whose mind is never perfect - (again, according to Kurt Gadel), but he, the person , as an animal, is used to dealing with a precedent - he simply improvises, while immediately updating the picture of the world that exists in his head - a feature that a machine could take advantage of, if the person knew how to articulate to the machine from a precise objective understanding of the structure and operation of his soul.

.

- It would be nice if we knew about ourselves, what we pretend to teach the machine.

.

Here is what GPT3,5 thinks of itself:

*this article was google-translated from Hebrew

*this article was google-translated from Hebrew

Register to Reply
Pathfinder Tags: AI
Interested in this topic? By joining CR4 you can "subscribe" to
this discussion and receive notification when new comments are added.

Comments rated to be Good Answers:

These comments received enough positive ratings to make them "good answers".

Comments rated to be "almost" Good Answers:

Check out these comments that don't yet have enough votes to be "official" good answers and, if you agree with them, rate them!
Guru

Join Date: Apr 2010
Location: About 4000 miles from the center of the earth (+/-100 mi)
Posts: 9629
Good Answers: 1098
#1

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/20/2024 4:03 PM

Natural intelligence learns from experience. If the environment changes, a creature with a brain adapts to cope.

Artificial Intelligence, as far as I know, does not learn from experience. It could be made to do so, but then it could "forget" some of its training, form its own opinions, and no longer be under the control of its developers, much like a human.

But don't take my word for it:

Ask GPT:

Does ChatGPT learn from users?

"ChatGPT does not learn from individual user interactions in real-time or on a continuous basis. The current version of ChatGPT, like other iterations of OpenAI's language models, was trained on a diverse dataset comprising text from various sources up until its knowledge cutoff date in April 2023. This training process involves machine learning techniques but is not influenced by individual conversations post-deployment.

OpenAI may collect and analyze data from interactions with the model to improve system performance, identify and mitigate biases, and enhance the user experience. However, these improvements and updates are incorporated into new versions of the model and are not applied to the model in real-time.

If you have specific questions about how your data is used or privacy concerns, it's best to refer to OpenAI's privacy policy and terms of service."

Register to Reply Score 1 for Good Answer
Guru

Join Date: Feb 2007
Location: Israel
Posts: 2966
Good Answers: 24
#3
In reply to #1

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/21/2024 1:34 AM

True. GPT doesn't even recall past sessions, and if you force-feed it recent discussions, it will relate to those as fresh input albeit acknowledging your claim of it being a quote of a past-discussion. All it can relate to, is you current query, in relation to it's database and training

Register to Reply
2
Guru
Engineering Fields - Nuclear Engineering - New Member

Join Date: Sep 2009
Location: Louisville, OH
Posts: 1784
Good Answers: 34
#5
In reply to #3

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/21/2024 11:10 AM

GPT doesn't even recall past sessions

Now this example makes sense! I asked it to tell me about the interurbans in Louisville, Ohio. It proceeded to tell me about the interurbans in Louisville, KY. I complained to it, and it apologized; then proceeded to again tell me about the interurbans in Louisville in Kentucky!

__________________
Lehman57
Register to Reply Good Answer (Score 2)
Guru

Join Date: Feb 2007
Location: Israel
Posts: 2966
Good Answers: 24
#14
In reply to #5

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

07/02/2024 6:42 PM

That's hilarious : it isn't even capable of relating to its own claims;

Talking about self-awareness...

Register to Reply
Guru

Join Date: Mar 2007
Location: by the beach in Florida
Posts: 32883
Good Answers: 1795
#7
In reply to #1

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/23/2024 2:53 AM

The first chat GPT models were babbling idiots repeating chatroom jibberish...today the new models combined with open AI are quite different, they use a database of facts to work from, and is barely distinguishable from a person, and in some cases better at interacting with a customer...

https://platform.openai.com/docs/models

__________________
All living things seek to control their own destiny....this is the purpose of life
Register to Reply
Guru
Engineering Fields - Nuclear Engineering - New Member

Join Date: Sep 2009
Location: Louisville, OH
Posts: 1784
Good Answers: 34
#2

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/20/2024 11:52 PM

My experience with AI is limited. From what I have observed, it can be very good at sifting thru mountains of data to determine trends. But it is poor at being an author of text material. It doesn't seem to have the flow and linkage that I expect, but it may be good for finding ideas and phrasing that can be used in a text article by a human author.

__________________
Lehman57
Register to Reply Score 1 for Good Answer
Guru

Join Date: Feb 2007
Location: Israel
Posts: 2966
Good Answers: 24
#4
In reply to #2

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/21/2024 1:40 AM

Not only "poor at being an author of text material" but unable to author in the deep sense of the term. It is however, incredibly efficient, clear and concise, at simplifying complex ideas - A trait most of us I guess aspire for...

Register to Reply
Guru
Hobbies - DIY Welding - Wannabeabettawelda

Join Date: May 2007
Location: Annapolis, Maryland
Posts: 7789
Good Answers: 447
#6

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/21/2024 9:37 PM

A clear example of how we don't even understand ourselves is the difficulty people have teaching a car how to drive. They are getting better no doubt, especially with reasonably well-controlled conditions, e.g. road markings, typical landmarks like street poles, signs, etc. but try to take a self-driving vehicle onto some forest trails and enjoy the fun. Also reading body language and facial expressions is extremely difficult for a machine in a dynamic environment.

My son's Tesla Y can do some downright scary stuff on some of these rural Delaware roadways generally devoid of much of anything short of some pavement that looks like about forty years of tar and chip. And that's on a paved roadway. I can only imagine what would happen when you wanted to follow some two-track across a field or up some fire road in the hills.

Register to Reply Score 1 for Good Answer
Guru

Join Date: Apr 2010
Location: About 4000 miles from the center of the earth (+/-100 mi)
Posts: 9629
Good Answers: 1098
#8
In reply to #6

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/23/2024 9:17 PM

One day I realized that, on seeing a car on a side street, I look at the driver's face. If they're not looking at me, I prepare for them to pull out in front. I'm sure most other drivers do this automatically without thinking about it.

I'm pretty sure AI doesn't do this.

Register to Reply
Guru

Join Date: Mar 2007
Location: by the beach in Florida
Posts: 32883
Good Answers: 1795
#9
In reply to #8

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/23/2024 10:41 PM

Ideally the car to car communication would solve this problem, with the cars determining who has the right of way and prohibiting one car from pulling out in the path of another...

__________________
All living things seek to control their own destiny....this is the purpose of life
Register to Reply Score 1 for Good Answer
Guru
Hobbies - DIY Welding - Wannabeabettawelda

Join Date: May 2007
Location: Annapolis, Maryland
Posts: 7789
Good Answers: 447
#10
In reply to #6

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

05/28/2024 5:33 PM

I should also mention that in "normal" developed areas, the Tesla Y does a remarkably good job of navigating around normal surface streets. Limited access highways in decent weather conditions are handled very well. Parking lots, where the driver interprets many visual cues and signals from other drivers and pedestrians is a challenging exercise for the car and stressful for the old-school passenger riding shotgun.

Register to Reply Score 1 for Good Answer
Guru

Join Date: Feb 2007
Location: Israel
Posts: 2966
Good Answers: 24
#11
In reply to #10

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

06/15/2024 5:21 AM

As I have claimed on previous threads - People's fear of AI is misplaced : There is no real AI in existance as of yet.

Instead, the most notable "AI" apps are in fact Expert-Systems at most, and here's the catch: Expert Systems actively manage our civilization for the last 30 years, and they do pose a real danger to human survival, far more than a real AI in a speculated future, in the realms of : Civil Nuclear management, Ground, Air, Sea and Space navigation systems, Comms and broadcast management, Power-Grid management, Water-Grid management, Gas-Grid management, Military Nuclear management - and that list goes on and on - how are those bug-proof ?

Register to Reply
Guru
Hobbies - DIY Welding - Wannabeabettawelda

Join Date: May 2007
Location: Annapolis, Maryland
Posts: 7789
Good Answers: 447
#12
In reply to #11

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

06/17/2024 7:26 PM

I agree. AI is the most abused, over-hyped buzzword in a long time. As far as most people are concerned, a PID controlled system is AI to them.

Few of these systems actually "learn" any more than a PID system finds a stable operating point.

I remember when "Fussy Logic", I mean "Fuzzy Logic" was going to totally revolutionize the entire electronics industry. The late, great Bob Pease took great joy in poking huge holes in the Fuzzy Logic constructs. That whole thing fell flat in short order when the hype did not materialize into useful product.

True AI is still a long ways off but there are a lot of people who have staked huge tracts of land, I mean money, in convincing people that their AI is really AI.

Register to Reply
Guru

Join Date: Feb 2007
Location: Israel
Posts: 2966
Good Answers: 24
#13
In reply to #12

Re: How Well Do We Understand Ourselves in Order to Teach a Machine?

07/02/2024 6:35 PM

Right. Not unlike those who made a personal furtune, merely by promoting the naive hope for Cold-Fusion...

Register to Reply
Register to Reply 14 comments
Interested in this topic? By joining CR4 you can "subscribe" to
this discussion and receive notification when new comments are added.

Comments rated to be Good Answers:

These comments received enough positive ratings to make them "good answers".

Comments rated to be "almost" Good Answers:

Check out these comments that don't yet have enough votes to be "official" good answers and, if you agree with them, rate them!
Copy to Clipboard

Users who posted comments:

Brave Sir Robin (3); Lehman57 (2); Rixter (2); SolarEagle (2); Yuval (5)

Previous in Forum: Why a 'Language Model' (such as ChatGPT) is not even nearly AI  

Advertisement