Hi
you say " no understanding "
is it not analogous to "a child" , they learn from everything they are plumed with,
and that can lead to better or worse "responses" over time
But never say a child is not intelligent,

Which comes around again to the fact we have no definition on what counts as "intelligent" and how we measure it

No, it's not analogous to a child learning where preexisting intelligence is a requirement, it's reindexing weights and constraints of the next word response. It will never understand or gain intelligence this way (some completely different type of system in the future likely will). That's not to say it can't be extremely useful as a dumb labourer if we understand and keep within the limits of its processing .

 

I was going to answer dr's post but spook beat me to it ...

 

No, it's not analogous to a child learning where preexisting intelligence is a requirement, it's reindexing weights and constraints of the next word response. It will never understand or gain intelligence this way (some completely different type of system in the future likely will). That's not to say it can't be extremely useful as a dumb labourer if we understand and keep within the limits of its processing .

re " preexisting intelligence is a requirement "
again it comes back to the question of what is and how do we measure intelligence,
we still dont seem to have an answer to that , so this constant rounds of is it / isn't it, will keep going on

 

Look, we don't know at lot of things about the brain starting with how important the signals we monitor and mimic with AI programs are even the true basis for how we think.

https://scitechdaily.com/century-ol...-shape-matters-more-than-neural-connectivity/

For over a hundred years, scientists have held the belief that our thoughts, feelings, and dreams are shaped by the way various brain regions interact via a vast network of trillions of cellular connections.

However, a recent study led by the team at Monash University’s Turner Institute for Brain and Mental Health has examined more than 10,000 distinct maps of human brain activity and discovered that the overall shape of an individual’s brain has a much more substantial impact on our cognitive processes, emotions, and behavior than its intricate neuronal connectivity.
...
This result counters conventional wisdom, in which activity during different tasks is often assumed to occur in focal, isolated areas of elevated activity, and tells us that traditional approaches to brain mapping may only show the tip of the iceberg when it comes to understanding how the brain works,” Dr. Pang said.

 

Look, we don't know at lot of things about the brain starting with how important the signals we monitor and mimic with AI programs are even the true basis for how we think.

https://scitechdaily.com/century-ol...-shape-matters-more-than-neural-connectivity/

For over a hundred years, scientists have held the belief that our thoughts, feelings, and dreams are shaped by the way various brain regions interact via a vast network of trillions of cellular connections.

However, a recent study led by the team at Monash University’s Turner Institute for Brain and Mental Health has examined more than 10,000 distinct maps of human brain activity and discovered that the overall shape of an individual’s brain has a much more substantial impact on our cognitive processes, emotions, and behavior than its intricate neuronal connectivity.
...
This result counters conventional wisdom, in which activity during different tasks is often assumed to occur in focal, isolated areas of elevated activity, and tells us that traditional approaches to brain mapping may only show the tip of the iceberg when it comes to understanding how the brain works,” Dr. Pang said.

do we need to understand how something is working to have a test for is it working ?

 

do we need to understand how something is working to have a test for is it working ?

Yes, if it's our original creation to recreate existing systems and not some wild ass guess instead of a educated one that Y is equal to X.
How do you define that testing beyond some trivial level of on/off?
It's called science and engineering where we use intelligence and understanding to predict how something will act before being built and can communicate that understanding so others can conduct experiments to validate that understanding.

 

Yes, if it's our original creation to recreate existing systems and not some wild ass guess instead of a educated one that Y is equal to X.
How do you define that testing beyond some trivial level of on/off?
It's called science and engineering where we use intelligence and understanding to predict how something will act before being built and can communicate that understanding so others can conduct experiments to validate that understanding.

interesting,
people like Edison as far as I know, did not understand the physics of say how sound worked,
but they "invented" the "phonogram" or what ever it was called.
I'm not certain Edison knew the molecular / atom level of detail as to why carbon made from bamboo worked as a light source when electricity passed.

They had "models" , they could predict, they could make tests and learn,
hence my comment on the lines of, we often have tests for something, based upon the results we expect, not on understanding,

 

interesting,
people like Edison as far as I know, did not understand the physics of say how sound worked,
but they "invented" the "phonogram" or what ever it was called.
I'm not certain Edison knew the molecular / atom level of detail as to why carbon made from bamboo worked as a light source when electricity passed.

They had "models" , they could predict, they could make tests and learn,
hence my comment on the lines of, we often have tests for something, based upon the results we expect, not on understanding,

Edison is a poor example IMO.
Yes, he understood at some level the physics of sound at a qualitative level.

You really believe the modern Edison myth of invention, improvement and perfection?
He hired people when he need more than he understood.
https://en.wikipedia.org/wiki/Francis_Robbins_Upton
Francis Robbins Upton (1852 in Peabody, Massachusetts – March 10, 1921 in Orange, New Jersey) was an American physicist and mathematician. Upton worked alongside Thomas Edison in the development of incandescent light bulbs, electric generators, and electric power distribution. He was the first president of the Edison Pioneers.

He did understand the current details to improve and perfect the things Edison is known for.

 

hi,

 

Edison is a poor example IMO.
Yes, he understood at some level the physics of sound at a qualitative level.

You really believe the modern Edison myth of invention, improvement and perfection?
He hired people when he need more than he understood.
https://en.wikipedia.org/wiki/Francis_Robbins_Upton
Francis Robbins Upton (1852 in Peabody, Massachusetts – March 10, 1921 in Orange, New Jersey) was an American physicist and mathematician. Upton worked alongside Thomas Edison in the development of incandescent light bulbs, electric generators, and electric power distribution. He was the first president of the Edison Pioneers.

He did understand the current details to improve and perfect the things Edison is known for.

Ok, selective on your "intelligent" people then
sounds like a test to me

 

hi,
View attachment 298899

dont dis agree with your comments about who invented the bulb
were talking about ChatGPT , and AI
I was using the engineer edison as being intelegent,
and the technique he used as an example of intelligence,

 

Ok, selective on your "intelligent" people then
sounds like a test to me

What? Your poor examples makes me think this is a waste of time.

 

What? Your poor examples makes me think this is a waste of time.

Can you expand please @nsaspook on what you mean by "makes me think this is a waste of time"
is the complete AI / chat GPT thing your alluding to being a waste of time ?

We have talked about AI here,
and its very informative that we can all have our own views and debate them here
like being back at university debating , very enlightening,

For myself, I hear many feelings on AI / chatGPT / LL models et all

What I don't hear is a definition of what intelligence is that can be a test and is measurable
And I think, unless we as humanity can do that, then we are never going to agree on AI.

Im feeling a definition / test of AI is like trying to make a test for "beauty" ,
which seems subjective, and variable over time / location.

 

do we need to understand how something is working to have a test for is it working ?

No, but we do need to understand what qualifies as "working".

 

No, but we do need to understand what qualifies as "working".

can you expand on that please @WBahn , what example would you give

 

can you expand on that please @WBahn , what example would you give

It would seem, to me, to be rather self evident. If I have a machine that does something and need to devise a test for whether or not it is working, I need to know what qualifies as "working" for that machine if I am to have any hope of devising a test that measures whether or not the machine is successfully doing it. The test for a machine that is supposed to make bricks is going to be different than the test for a machine that is supposed to make loaves of bread. Even machines that might be identical are going to have different tests for "working" if they are being used for different things. A machine that pumps water from a pit into a tank needs a different test of working if it is intended to keep the water in the pit below a certain level as opposed to keeping the water in the tank above a certain level. Not only do I NOT need to understand how the system pumps water in order to devise a test of "working", but even if I understand exactly how the machine pumps water, that does not enable me to devise a test of "working" if I don't understand what that machine is supposed to accomplish.

The same applies to any kind of software. How can you test if it is "working" unless you know what an acceptable definition of "working" is?

In the case of something like ChatGPT, what qualifies as "working"? Like many systems, it's not a single answer since it is intended to do many things -- and it may succeed at some of them and fail at others. But to know what it is succeeding at and what it is failing at, we need to understand what the various things are that it is intended to do and what qualifies as success for each of them. In terms of whether it is exhibiting "intelligence", we don't have to know how intelligence comes about, but we do need to define what observable phenomena qualify as "intelligence".

 

It would seem, to me, to be rather self evident. If I have a machine that does something and need to devise a test for whether or not it is working, I need to know what qualifies as "working" for that machine if I am to have any hope of devising a test that measures whether or not the machine is successfully doing it. The test for a machine that is supposed to make bricks is going to be different than the test for a machine that is supposed to make loaves of bread. Even machines that might be identical are going to have different tests for "working" if they are being used for different things. A machine that pumps water from a pit into a tank needs a different test of working if it is intended to keep the water in the pit below a certain level as opposed to keeping the water in the tank above a certain level. Not only do I NOT need to understand how the system pumps water in order to devise a test of "working", but even if I understand exactly how the machine pumps water, that does not enable me to devise a test of "working" if I don't understand what that machine is supposed to accomplish.

The same applies to any kind of software. How can you test if it is "working" unless you know what an acceptable definition of "working" is?

In the case of something like ChatGPT, what qualifies as "working"? Like many systems, it's not a single answer since it is intended to do many things -- and it may succeed at some of them and fail at others. But to know what it is succeeding at and what it is failing at, we need to understand what the various things are that it is intended to do and what qualifies as success for each of them. In terms of whether it is exhibiting "intelligence", we don't have to know how intelligence comes about, but we do need to define what observable phenomena qualify as "intelligence".

HI
good chat ,
Id have to agree with you, if I designed a machine, Id have a pass / fail test suit in place, part of acceptance,

But we are talking about a test of intelligence, and how to apply it to AI , so that we could say yes / no this is intelligence,
so we could well have a test of is chatgpt working as we wanted,
but that would not answer the discussion above about what is intelligence, and how to test for it

If we dont have such a test, how cna we say is something intelegent or not ?

Im coming to the conclusion , we dont have the ability to define a test for what is intelligence,
but IM willing to be convinced otherwise.

 

https://www.cnn.com/2023/07/22/tech/ai-jobs-efficiency-productivity
‘It almost doubled our workload’: AI is supposed to make jobs easier. These workers disagree

Others like Clarke, the publisher, have tried to combat the fallout from the rise of AI by relying on more AI. Clarke said he and his team turned to AI-powered detectors of AI-generated work to deal with the deluge of submissions but found these tools weren’t helpful because of how unreliably they flag “false positives and false negatives,” especially for writers whose second language is English.

“You listen to these AI experts, they go on about how these things are going to do amazing breakthroughs in different fields,” Clarke said. “But those aren’t the fields they’re currently working in.”

 

... especially for writers whose second language is English.

I have a confession to make ...

​

 

HI
good chat ,
Id have to agree with you, if I designed a machine, Id have a pass / fail test suit in place, part of acceptance,

But we are talking about a test of intelligence, and how to apply it to AI , so that we could say yes / no this is intelligence,
so we could well have a test of is chatgpt working as we wanted,
but that would not answer the discussion above about what is intelligence, and how to test for it

If we dont have such a test, how cna we say is something intelegent or not ?

Im coming to the conclusion , we dont have the ability to define a test for what is intelligence,
but IM willing to be convinced otherwise.

That's my point -- we can't test for something until we define what it is, and do so in a testable way. We don't have to understand where intelligence comes from to do that, but we do have to define what it is.

People often think that the Turing Test was meant to be a test of intelligence, but it wasn't. It was an attempt at creating a definition of "intelligent behavior".