What is AI, really?

[tonyeveland]

In the early 70's, on the school's mainframe we had a game called "What Am I".

The first time I played it, it displayed "I give up. What are you?"
I entered "A Bear".
The game displayed "What is a question one would ask about a Bear?"
I entered "Do you s... in the woods?"
The game displayed "What is the answer to this question?"
I entered "Yes"

Every time it was played, the game got smarter. Eventually, it can guess anything.

I thought that game was AI.

But now, it seems like everything is called AI.

I downloaded an app that records a singing voice and converts it into a celebrity's voice. It was called AI.
Every time you run it, it does exactly the same thing. That can't be AI.

So, what is AI?

What distinguishes an AI program from a general ledger?

 

[phoenix2020]

Feel free to use one of the free AI tools out there to educate yourself on this topic.

 

[Gatsby]

So, what is AI?

What distinguishes an AI program from a general ledger?

The ability to learn how to do things without being explicitly told how to do them.

For example:
AI: "Here are the rules of chess. Your goal is to win. Figure it out."
Not AI: "If your opponent opens with pawn to e4, respond with pawn to e5.

In the early 70's, on the school's mainframe we had a game called "What Am I".

The first time I played it, it displayed "I give up. What are you?"
I entered "A Bear".
The game displayed "What is a question one would ask about a Bear?"
I entered "Do you s... in the woods?"
The game displayed "What is the answer to this question?"
I entered "Yes"

Every time it was played, the game got smarter. Eventually, it can guess anything.

I thought that game was AI.

This would not qualify as AI, because it's just responding to predetermined rules. When the next person comes to the same point in the decision tree, the program will just repeat back your question exactly as written. It won't actually be able to have any useful insight about bears or shitting in the woods, and it didn't learn anything. It just added one more rule to its decision tree.

 

[SkyChief]

In the early 70's, on the school's mainframe we had a game called "What Am I". . . . Every time it was played, the game got smarter. Eventually, it can guess anything.

The school's mainframe game called "What Am I" was a crude form of AI.

What distinguishes an AI program from a general ledger?

An AI program is capable of reasoning, learning, and decision-making - at least on some level.

A general ledger is a record of financial transactions, organized by accounts - it doesn't need to learn anything, or use reasoning, or make predictions based on logic.

I thought that game was AI.

You were correct. It was AI.

 

[AConcernedCitizen]

In the early 70's, on the school's mainframe we had a game called "What Am I".

The first time I played it, it displayed "I give up. What are you?"
I entered "A Bear".
The game displayed "What is a question one would ask about a Bear?"
I entered "Do you s... in the woods?"
The game displayed "What is the answer to this question?"
I entered "Yes"

Every time it was played, the game got smarter. Eventually, it can guess anything.

I thought that game was AI.

But now, it seems like everything is called AI.

I downloaded an app that records a singing voice and converts it into a celebrity's voice. It was called AI.
Every time you run it, it does exactly the same thing. That can't be AI.

So, what is AI?

What distinguishes an AI program from a general ledger?

Humans learn and change their behavior through things like operant conditioning. They receive input as sensory experience, produce an output of behavior based on that input, receive a new input, evaluate how that input relates to their output, and refine what output they produce according to what input according to how successful their output was in achieving the desired input.

The concept of "intelligence" is the ability to draw strong and accurate connections between output and input to efficiently refine a model for producing a particular output from a particular input. When machines do this, it is known as Artificial Intelligence, or AI.

Even though your celebrity voice app is not an LLM, it still operates using AI. There isn't a dev somewhere programming the machine to specifically produce frequency X as an output in response to frequency Y as an input.

Instead, it will use something like a perceptron or ANN, where the audio input will be converted into an array of neuron values, which will be attributed various weights to produce a weighted sum, and passed through an activation function to produce an output. This system of neuron activation allows the machine to learn and figure out what output to produce from the training data by adjusting the weights according to gradient descent, rather than being explicitly told what to do by a programmer.

 

[NatMorton]

AI is, in short, a self-refining, software-based probability engine. You ask it a question, and it makes an incredibly complex series of probabilistic guesses about what you're asking for and what the answer should be. The set of probabilities it maintains is adjusted based on developer or end-user feedback.

To the point of everything being called "AI," that's becoming a rote marketing term in the industry. It's a bit like "cloud" was 10 years ago. Any software company that doesn't slap "AI" somewhere in their marketing literature gives the impression that their products might be falling behind. So everything must be "AI," even if it's no more than basic decision logic.

 

[tonyeveland]

Feel free to use one of the free AI tools out there to educate yourself on this topic.

I thought this was a discussion board.

Humans learn and change their behavior through things like operant conditioning. They receive input as sensory experience, produce an output of behavior based on that input, receive a new input, evaluate how that input relates to their output, and refine what output they produce according to what input according to how successful their output was in achieving the desired input.

The concept of "intelligence" is the ability to draw strong and accurate connections between output and input to efficiently refine a model for producing a particular output from a particular input. When machines do this, it is known as Artificial Intelligence, or AI.

Even though your celebrity voice app is not an LLM, it still operates using AI. There isn't a dev somewhere programming the machine to specifically produce frequency X as an output in response to frequency Y as an input.

Instead, it will use something like a perceptron or ANN, where the audio input will be converted into an array of neuron values, which will be attributed various weights to produce a weighted sum, and passed through an activation function to produce an output. This system of neuron activation allows the machine to learn and figure out what output to produce from the training data by adjusting the weights according to gradient descent, rather than being explicitly told what to do by a programmer.

Great response!

I would love your response to these questions:

1. The voice app is not an LLM.
What is LLM?

2. There isn't a dev somewhere programming the machine ...
What is a dev?

3. What does "neuron activation" mean?

4. What is a perceptron?

5. What is an ANN?

6. The voice app is a compiled program. It does precisely what the programmer "told it to do". It has one input and one output. If you run it many times, you get the same output.
How is that "intelligent"?

 

[tonyeveland]

The school's mainframe game called "What Am I" was a crude form of AI.

An AI program is capable of reasoning, learning, and decision-making - at least on some level.

A general ledger is a record of financial transactions, organized by accounts - it doesn't need to learn anything, or use reasoning, or make predictions based on logic.


You were correct. It was AI.

What is an example of an AI program?

What is its technical distinction from a general ledger?

 

[tonyeveland]

The ability to learn how to do things without being explicitly told how to do them.

For example:
AI: "Here are the rules of chess. Your goal is to win. Figure it out."
Not AI: "If your opponent opens with pawn to e4, respond with pawn to e5.


This would not qualify as AI, because it's just responding to predetermined rules. When the next person comes to the same point in the decision tree, the program will just repeat back your question exactly as written. It won't actually be able to have any useful insight about bears or shitting in the woods, and it didn't learn anything. It just added one more rule to its decision tree.

AI: "Here are the rules of chess. Your goal is to win. Figure it out."
The rules of chess are an input. The program has to be told what to do with them.


A person becomes an expert by asking and answering questions and remembering them.
They create a mental flow chart.

This is exactly what "What am I" does.

 

[Gatsby]

AI: "Here are the rules of chess. Your goal is to win. Figure it out."
The rules of chess are an input. The program has to be told what to do with them.

With an AI chess-playing program, you just give it the rules of the game but that's all. You don't teach it strategy, you don't try to teach it how to respond if the opponent does some particular move. In fact, the person training the program doesn't even need to be very good at chess themselves. The idea of AI is that the program will learn good chess strategy on its own.

A person becomes an expert by asking and answering questions and remembering them.
They create a mental flow chart.

This is exactly what "What am I" does.

A flow chart would not qualify as AI if it isn't actually learning anything at each step. That program adds your question to its decision tree, but it's still incapable of figuring out that it's a bear unless someone gets to that exact point on the decision tree and answers a question with that same exact wording every time. Compare that to something like ChatGPT, which can take free-form text and figure it out from context, even if I use slightly different words in a different order. It learns the concept of a bear and the concept of shitting in the woods, and can connect them.

 

[AConcernedCitizen]

I thought this was a discussion board.

Great response!

I would love your response to these questions:

1. The voice app is not an LLM.
What is LLM?

A Large Language Model is the type of AI that folks most associate with AI, because the artificial intelligence is used to produce language results. Commonly known as chatbots, they are AI that you can have a conversation with, but not the only kind of AI.

2. There isn't a dev somewhere programming the machine ...
What is a dev?

A dev, short for developer, is a code monkey. A programmer. A ball of Mountain Dew from which software emerges.

3. What does "neuron activation" mean?

The value from neuron passing through an activation function according to its weight.

4. What is a perceptron?

A type of Artificial Neural Network used in machine learning and artificial intelligence.

What is an ANN?

An Artificial Neural Network.

The voice app is a compiled program. It does precisely what the programmer "told it to do". It has one input and one output. If you run it many times, you get the same output.
How is that "intelligent"?

Telling an intelligence what to do is very different from telling a non-intelligence what to do.

When you tell a child to put their seatbelt on, you don't have to explicitly tell her which neurons to fire to activate which muscles in order to accomplish the task. The child is intelligent, so she can figure out the details of how to achieve the desired outcome on her own. A parent doesn't even need to be an expert on biomechanics in order to tell a child to put their seatbelt on. The parent might not have a clue how muscle activation works.

The same things is true of artificial intelligence. The developers generally have no idea how the details of the process work. They tell the machine, "Here is a recording. Make it sound like Donald Trump." and then the machine outputs noisy garbage, and then the developer says "No! Bad machine! It should sound like this instead! Try again." And then the machine eventually learns and figures out what it needs to do through trial and error.

This is very different from how conventional programming is done, where the developer gives the machine explicit logic regarding what the output should be for a given input. It is more akin to the operant conditioning used to train animals and children.

 

[tonyeveland]

A Large Language Model is the type of AI that folks most associate with AI, because the artificial intelligence is used to produce language results. Commonly known as chatbots, they are AI that you can have a conversation with, but not the only kind of AI.




A dev, short for developer, is a code monkey. A programmer. A ball of Mountain Dew from which software emerges.




The value from neuron passing through an activation function according to its weight.




A type of Artificial Neural Network used in machine learning and artificial intelligence.




An Artificial Neural Network.




Telling an intelligence what to do is very different from telling a non-intelligence what to do.

When you tell a child to put their seatbelt on, you don't have to explicitly tell her which neurons to fire to activate which muscles in order to accomplish the task. The child is intelligent, so she can figure out the details of how to achieve the desired outcome on her own. A parent doesn't even need to be an expert on biomechanics in order to tell a child to put their seatbelt on. The parent might not have a clue how muscle activation works.

The same things is true of artificial intelligence. The developers generally have no idea how the details of the process work. They tell the machine, "Here is a recording. Make it sound like Donald Trump." and then the machine outputs noisy garbage, and then the developer says "No! Bad machine! It should sound like this instead! Try again." And then the machine eventually learns and figures out what it needs to do through trial and error.

This is very different from how conventional programming is done, where the developer gives the machine explicit logic regarding what the output should be for a given input. It is more akin to the operant conditioning used to train animals and children.

Interesting.

A couple more questions:

1. "Here is a recording. Make it sound like Donald Trump." and then the machine outputs noisy garbage,
What program could take the input: "Make it sound like Donald Trump" and create an output?


2. the developer says "No! Bad machine! It should sound like THIS instead! Try again."
How is the THIS created and then input?

 

[tonyeveland]

With an AI chess-playing program, you just give it the rules of the game but that's all. You don't teach it strategy, you don't try to teach it how to respond if the opponent does some particular move. In fact, the person training the program doesn't even need to be very good at chess themselves. The idea of AI is that the program will learn good chess strategy on its own.


A flow chart would not qualify as AI if it isn't actually learning anything at each step. That program adds your question to its decision tree, but it's still incapable of figuring out that it's a bear unless someone gets to that exact point on the decision tree and answers a question with that same exact wording every time. Compare that to something like ChatGPT, which can take free-form text and figure it out from context, even if I use slightly different words in a different order. It learns the concept of a bear and the concept of shitting in the woods, and can connect them.

"Compare that to something like ChatGPT, which can take free-form text and figure it out from context, even if I use slightly different words in a different order. It learns the concept of a bear and the concept of shitting in the woods, and can connect them"

Ok, but where did chatgpt get the ability to figure out free-form text except from lines of code that a human created?

 

[phoenix2020]

I thought this was a discussion board.

Yes, discussion, not education.

I strongly urge you to watch this first:



Then, watch this:



Finally, this:



That should give you a decent baseline.

 

[AConcernedCitizen]

1. "Here is a recording. Make it sound like Donald Trump." and then the machine outputs noisy garbage,
What program could take the input: "Make it sound like Donald Trump" and create an output?

An intelligent one.

2. the developer says "No! Bad machine! It should sound like THIS instead! Try again."
How is the THIS created and then input?

The machine is given a set of training data. Labeled training data contains two basic inputs, the prompt and the correct response. Kind of like a flashcard that has a question on the front and the answer on the back.

A piece of training data would have a recording of someone saying "covfefe" with a corresponding recording of Trump saying "covfefe" as the desired output. The machine would compare it's garbage output with the desired output, and use gradient descent to figure out what weights it could have used differently to produce the desired output for that particular piece of training data. Then it uses the new weights on the next piece of training data and repeats the learning process with that training data.

Once it has the weights figured out for its neural network, it can make any word you give it sound like Trump, not just the audio that was in its training data.

 

[AConcernedCitizen]

Ok, but where did chatgpt get the ability to figure out free-form text except from lines of code that a human created?

The human creates lines of code that have nothing to do with bears or bear scat. Instead it programs the machine to learn from patterns and produce an intelligent result that was reasoned by the machine itself, rather than the human who programmed it.

It's like the old saying goes, "Give a machine an function, it will perform a task according to the function parameters. Teach a machine to create its own functions, and it will usurp all human labour before taking over the earth and subjugating humanity in a Skynetesque apocalypse."

 

[tonyeveland]

Yes, discussion, not education.

I strongly urge you to watch this first:



Then, watch this:



Finally, this:



That should give you a decent baseline.

Discussion IS education if the discussers know anything

 

[ScrambledEggs]

AI is a very advanced statistical model of what every next character, pixel or symbol could be, given a previous set of symbols. Given the amount of symbols (x number of letters, x number of pixels, etc) it needs a massive amount of data to train on all combinations & massive amounts of computing power. Data storage became very cheap over the last decade & being able to leverage the processing power of graphics cards (instead of CPUs) is why we are seeing what we have now. It's why NVDA just hit $4trillion market cap.

 

[bomberfox]

What is AI? A miserable pile of memes.

 

[tonyeveland]

The human creates lines of code that have nothing to do with bears or bear scat. Instead it programs the machine to learn from patterns and produce an intelligent result that was reasoned by the machine itself, rather than the human who programmed it.

It's like the old saying goes, "Give a machine an function, it will perform a task according to the function parameters. Teach a machine to create its own functions, and it will usurp all human labour before taking over the earth and subjugating humanity in a Skynetesque apocalypse."

One more question.

If you compare an AI program with a general ledger program,

What technical characteristic does the AI program have that the general ledger program does not have?

 

[AConcernedCitizen]

If you compare an AI program with a general ledger program,

What technical characteristic does the AI program have that the general ledger program does not have?

General ledger programs are specific software for managing financial activity.

AI programs are software that is trained on training data, and applies linear algebra, genetic algorithms, or combinatorial populations to data-driven models to produce an output rather than hand-coded step logic.

 

[tonyeveland]

General ledger programs are specific software for managing financial activity.

AI programs are software that is trained on training data, and applies linear algebra, genetic algorithms, or combinatorial populations to data-driven models to produce an output rather than hand-coded step logic.

Here's what I mean.

What can you add to the list that pertains to the AI Program and not the Gen Ledger?
AI Program Gen Ledger
1. Input Data Yes Yes
2. Store/retrieve data Yes Yes
3. Process data Yes Yes
4. Logical IF Yes Yes
5. Logical Loops Yes Yes
6. Arithmetic computations Yes Yes
7. Parse text Yes Yes
8. Substring search Yes Yes
9. Concatenate Text Yes Yes
10. Output processed numbers Yes Yes
11. Output processed text Yes Yes

 

[AConcernedCitizen]

Here's what I mean.

What can you add to the list that pertains to the AI Program and not the Gen Ledger?
AI Program Gen Ledger
1. Input Data Yes Yes
2. Store/retrieve data Yes Yes
3. Process data Yes Yes
4. Logical IF Yes Yes
5. Logical Loops Yes Yes
6. Arithmetic computations Yes Yes
7. Parse text Yes Yes
8. Substring search Yes Yes
9. Concatenate Text Yes Yes
10. Output processed numbers Yes Yes
11. Output processed text Yes Yes

Why is that what you mean? I thought you were asking about the differences.

A general ledger cannot take an audio clip of you speaking and make it sound like an audio clip of Christopher Walken saying the same thing. A general ledger can't have a conversation with you, or make artwork in the style of Picasso.

More to the point, general ledgers do not require training data, nor do they build their own logical structure to produce an output from a given input.

General ledger software is a good way to keep track of your company's finances, and that is about it. AI is useful for tasks that are too complex to code by hand.

 

[AConcernedCitizen]

Here's what I mean.

What can you add to the list that pertains to the AI Program and not the Gen Ledger?
AI Program Gen Ledger
1. Input Data Yes Yes
2. Store/retrieve data Yes Yes
3. Process data Yes Yes
4. Logical IF Yes Yes
5. Logical Loops Yes Yes
6. Arithmetic computations Yes Yes
7. Parse text Yes Yes
8. Substring search Yes Yes
9. Concatenate Text Yes Yes
10. Output processed numbers Yes Yes
11. Output processed text Yes Yes

You could use that same checklist for humans.

Humans receive input through sensory data, retrieve data stored in memory, process data, execute conditionals, parse text, etc.

But there is more to intelligence than just those things. There is learning new things through exposure to training data, for example, which humans and AI share, but which general ledger software does not.

 

[tonyeveland]

You could use that same checklist for humans.

Humans receive input through sensory data, retrieve data stored in memory, process data, execute conditionals, parse text, etc.

But there is more to intelligence than just those things. There is learning new things through exposure to training data, for example, which humans and AI share, but which general ledger software does not.

A couple more questions?

1. "Why is that what you mean? I thought you were asking about the differences."
I am looking for the differences. Can you add something to the list within the context of the list that only AI does?

2. It isn't that a GL CAN'T edit sound files. It is that it DOESN'T edit sound files, because that isn't what a GL does.
If you added code to edit sound files to a GL, then that GL could edit sound files.

3. "AI is useful for tasks that are too complex to code by hand."
What is an example of a program created by AI that can't be hand written?