- Dołączono
- 14 Lis 2022
Going out on a limb to make an extreme point. A single counterexample or proof of the possibility is enough to refute me on this.
I despise contemporary discussion around "AI" because, the way I see it, "AI" simply does not exist. At best, it's a misleading marketing term (read: disinformation) for large language models (LLMs) or other ""learning"" algorithms. What does exist is machinery that simulates certain outputs associated with intelligence. My reasoning may be familiar to you if you are very savvy in good epistemology.
A digital computer is a physical switching system. A computer's hardware consists of transistors, voltages, memory cells, circuits, charge states, and causal electrical transitions. The "software" running on any digital computer is not a thinking substance floating above the hardware. When you reduce the concept properly, "software" is a human way of interpreting physical arrangements and transitions in the machine.
A machine does not know that it is running a program. It does not know that some voltage pattern "means" the number 7 or the word "dog" or a chess position or a line of code. These meanings exist for us humans, but they do not exist for the machine. What the computer does is undergo physical state changes which were designed by humans, interpreted by humans, and are used for cognitive purposes.
To make an analogy: count with your fingers.
Are your fingers counting? No. You are counting.
Write some arithmetic on paper. Is the paper doing mathematics? No. You are.
Can you use a calculator to help you add numbers together? Yes. But the calculator is not aware of numbers. It is producing physical outputs according to its physical structure.
And that basic reasoning applies to more complicated machines. Increasing the number of switches does not change what kind of thing a switch is. A billion transistors are no more aware than one transistor. Or do a billion transistors suddenly become aware just by being numerous? A trillion statistical correlations or matrix multiplication weights are not a concept. Even the best state-of-the-art LLM that will be available in a million years is not a mind.
The only reason why things like programmed video game non-player character behavior or chatbots may seem intelligent is because the input and output are already saturated with human meaning.
Think about it, specifically for LLMs. The training data was written by humans (or generated from models that used training data written by humans), the task was defined by humans, the model was built by humans, the model is prompted by humans, the output is read by humans. The very marks on your screen, the ones you are reading right now, you as a human are interpreting these differently colored pieces of light as meaningful language. Everything semantic or genuinely cognitive or "intelligent" is happening in humans outside of the machine. Nothing of the sort is happening inside the machine.
A chatbot can imitate explanations because it has been trained on explanations. It can imitate reasoning because it has been trained on the written products of reasoning. It can imitate personality because it has been trained on human expression. However, imitation of cognitive output is not cognition.
When a tape recorder says "I understand", does the tape recorder understand? No.
Is a wax figure of a man alive? No.
The same way, a chatbot generating an essay about consciousness is not conscious.
Does the physical system have the kind of nature that is required for awareness, perception, concept formation, judgment, valuation, and volition? For humans, the answer is yes. For every computational machine, the answer is no.
Human beings are living organisms that perceive the world, act in it, have needs, face alternatives. Human beings form concepts from perceptual contact with reality (sensor input is by itself not perceptual awareness). Human beings validate ideas by reducing them back to reality. We human beings can direct our focus, question our own conclusions, reject invalid premises, and grasp that something is true or false.
A computer does none of that, because it has no perceptual contact with reality. It has no life to sustain. It has no values, no purposes of its own, no awareness of what "its" symbols refer to. It does not know when it is correct, it does not know when it is "hallucinating", it does not know anything. All it does is "produce" outputs. Which makes it machinery, not intelligence.
Is a telescope an astronomer because it helps humans see the stars? No, and by the same token, calling a LLM intelligent because it produces useful language is false.
Tools, such as telescopes, calculators, chatbots, computers, extend the efficacy and power of the human mind that chooses to use them. Unfortunately, human beings are easily seduced by language. If some thing speaks in full sentences, people will sooner or later start thinking that the thing has understanding. However, language without awareness is not thought. There is no "someone" inside the machine to whom words mean anything. There is no subject grasping an object. There is no mind that's identifying reality. There is no conceptual awareness. There is only a physical system producing patterns that humans then interpret.
However, hand waving is unacceptable. A proof of this point is not accomplished by just saying shit like "emergence" or "complexity" or asserting "the brain is basically a computer".
If you want to prove me wrong, explain how this specific kind of electrical switching (digital computation over interpreted symbols) becomes awareness. Explain how formal symbol manipulation becomes meaning. Explain how a digital machine with no life, no perception, no self-directed focus, and no intrinsic purposes suddenly becomes capable of understanding.
Until that has been done, "AI" remains a marketing term (read: disinformation) at best.
"AI" strongly implies the existence of artificial intelligence, which in turn means cognition, understanding, awareness, or agency. Current systems are not intelligent, therefore the label communicates a false impression, especially when paired with anthropomorphic marketing ("the AI thinks/knows/wants/understands/hallucinated/reasoned"). This, in turn, drives investment, hype, labor panic, policy proposals, and consumer behavior.
If a company knowingly represents a non-intelligent tool as intelligent in order to manipulate investment, labor markets, valuations, or consumer decisions, that's attempted fraud through semantic deception. Categorically impermissible.
But "intelligence"? No. Not now, not in a trillion years.
I despise contemporary discussion around "AI" because, the way I see it, "AI" simply does not exist. At best, it's a misleading marketing term (read: disinformation) for large language models (LLMs) or other ""learning"" algorithms. What does exist is machinery that simulates certain outputs associated with intelligence. My reasoning may be familiar to you if you are very savvy in good epistemology.
A digital computer is a physical switching system. A computer's hardware consists of transistors, voltages, memory cells, circuits, charge states, and causal electrical transitions. The "software" running on any digital computer is not a thinking substance floating above the hardware. When you reduce the concept properly, "software" is a human way of interpreting physical arrangements and transitions in the machine.
A machine does not know that it is running a program. It does not know that some voltage pattern "means" the number 7 or the word "dog" or a chess position or a line of code. These meanings exist for us humans, but they do not exist for the machine. What the computer does is undergo physical state changes which were designed by humans, interpreted by humans, and are used for cognitive purposes.
To make an analogy: count with your fingers.
Are your fingers counting? No. You are counting.
Write some arithmetic on paper. Is the paper doing mathematics? No. You are.
Can you use a calculator to help you add numbers together? Yes. But the calculator is not aware of numbers. It is producing physical outputs according to its physical structure.
And that basic reasoning applies to more complicated machines. Increasing the number of switches does not change what kind of thing a switch is. A billion transistors are no more aware than one transistor. Or do a billion transistors suddenly become aware just by being numerous? A trillion statistical correlations or matrix multiplication weights are not a concept. Even the best state-of-the-art LLM that will be available in a million years is not a mind.
The only reason why things like programmed video game non-player character behavior or chatbots may seem intelligent is because the input and output are already saturated with human meaning.
Think about it, specifically for LLMs. The training data was written by humans (or generated from models that used training data written by humans), the task was defined by humans, the model was built by humans, the model is prompted by humans, the output is read by humans. The very marks on your screen, the ones you are reading right now, you as a human are interpreting these differently colored pieces of light as meaningful language. Everything semantic or genuinely cognitive or "intelligent" is happening in humans outside of the machine. Nothing of the sort is happening inside the machine.
A chatbot can imitate explanations because it has been trained on explanations. It can imitate reasoning because it has been trained on the written products of reasoning. It can imitate personality because it has been trained on human expression. However, imitation of cognitive output is not cognition.
When a tape recorder says "I understand", does the tape recorder understand? No.
Is a wax figure of a man alive? No.
The same way, a chatbot generating an essay about consciousness is not conscious.
But, XL xQgg?QcQCaTYDMjqoDnYpG, human beings are also physical systems!
So? It proves nothing.Does the physical system have the kind of nature that is required for awareness, perception, concept formation, judgment, valuation, and volition? For humans, the answer is yes. For every computational machine, the answer is no.
Human beings are living organisms that perceive the world, act in it, have needs, face alternatives. Human beings form concepts from perceptual contact with reality (sensor input is by itself not perceptual awareness). Human beings validate ideas by reducing them back to reality. We human beings can direct our focus, question our own conclusions, reject invalid premises, and grasp that something is true or false.
A computer does none of that, because it has no perceptual contact with reality. It has no life to sustain. It has no values, no purposes of its own, no awareness of what "its" symbols refer to. It does not know when it is correct, it does not know when it is "hallucinating", it does not know anything. All it does is "produce" outputs. Which makes it machinery, not intelligence.
But, XL xQgg?QcQCaTYDMjqoDnYpG, the machinery of computers is really impressive!
Yes, I 100% agree. Changes nothing about my point though.Is a telescope an astronomer because it helps humans see the stars? No, and by the same token, calling a LLM intelligent because it produces useful language is false.
Tools, such as telescopes, calculators, chatbots, computers, extend the efficacy and power of the human mind that chooses to use them. Unfortunately, human beings are easily seduced by language. If some thing speaks in full sentences, people will sooner or later start thinking that the thing has understanding. However, language without awareness is not thought. There is no "someone" inside the machine to whom words mean anything. There is no subject grasping an object. There is no mind that's identifying reality. There is no conceptual awareness. There is only a physical system producing patterns that humans then interpret.
But, XL xQgg?QcQCaTYDMjqoDnYpG, maybe one day computer software or hardware will instantiate intelligence!
Then the burden is on you to prove it. I promise I'll change my mind if you can pull it off.However, hand waving is unacceptable. A proof of this point is not accomplished by just saying shit like "emergence" or "complexity" or asserting "the brain is basically a computer".
If you want to prove me wrong, explain how this specific kind of electrical switching (digital computation over interpreted symbols) becomes awareness. Explain how formal symbol manipulation becomes meaning. Explain how a digital machine with no life, no perception, no self-directed focus, and no intrinsic purposes suddenly becomes capable of understanding.
Until that has been done, "AI" remains a marketing term (read: disinformation) at best.
But, XL xQgg?QcQCaTYDMjqoDnYpG, why do you care? Why should I care?
Current "AI" is not merely overhyped, but the category label itself is false."AI" strongly implies the existence of artificial intelligence, which in turn means cognition, understanding, awareness, or agency. Current systems are not intelligent, therefore the label communicates a false impression, especially when paired with anthropomorphic marketing ("the AI thinks/knows/wants/understands/hallucinated/reasoned"). This, in turn, drives investment, hype, labor panic, policy proposals, and consumer behavior.
If a company knowingly represents a non-intelligent tool as intelligent in order to manipulate investment, labor markets, valuations, or consumer decisions, that's attempted fraud through semantic deception. Categorically impermissible.
But, XL xQgg?QcQCaTYDMjqoDnYpG, what is the more appropriate term for these things that are driving all the hype and uprooting labor markets?
Maybe something like "automated language engine" or "statistical text engine" or "artificial verbal behavior". If you're mean-spirited, "glorified matrix multiplication".But "intelligence"? No. Not now, not in a trillion years.
Ostatnio edytowane:
