He’s clearly got certain personal views, but I don’t think he was necessarily making the nuanced distinction you’re making. He’s just saying that in the wrong hands, artificial intelligence can lead to a life for average people more like present-day North Korea than United States.
learning by extracting information from data is not how the human mind works. He is talking about how AI will become more knowledgeable, not more intelligent.
He’s clearly got certain personal views, but I don’t think he was necessarily making the nuanced distinction you’re making. He’s just saying that in the wrong hands, artificial intelligence can lead to a life for average people more like present-day North Korea than United States.
As I suggested above, whether we end up like North Korea, or the United States, depends on us; not upon AI.
Of course this is true, but not necessarily as simple as it might seem. Free societies by whatever label are historically rare and fleeting. Certainly not to be taken for granted. What is the price of freedom?As I suggested above, whether we end up like North Korea, or the United States, depends on us; not upon AI.
In the lecture above he addresses your assertion indirectly by challenging you (others in his case) to define your terms and explain the difference.learning by extracting information from data is not how the human mind works. He is talking about how AI will become more knowledgeable, not more intelligent.
At the end (if you want to skip forward) he addresses what he calls, subjective experience, sentience and consciousness.
I have never read or heard a convincing argument that a machine can be more intelligent than humans. Two things here; when he uses the word “intelligent” what is he talking about? second, does he really know how the human brain works? (If so he is the only one).
In a nutshell, his contention is that digital computation is higher order of thinking. In his view, the human brain is a combination of a digital and analog computational hardware combined with the software, which uses far less energy, but is far more limited.
What separates digital computation is that the software and the hardware can be separated thus making AI “immortal” but more important, allowing multiple hardware devices to use the exact same program to gather more knowledge and share it with each other. Human sharing has quantitative and qualitative limitations in his view.
One interesting limitation of digital computation with its extraordinary speed is that life on earth is slow relatively speaking. This can be made up for by simply having innumerable machines collecting data simultaneously.
But none of this addresses your knowledge versus intelligence question.
He would probably admit that no one knows for certain, but he’s pretty confident that he does. (30:00, voltages for signals, conductances for weights)I have never read or heard a convincing argument that a machine can be more intelligent than humans. Two things here; when he uses the word “intelligent” what is he talking about? second, does he really know how the human brain works? (If so he is the only one).
Ignoring that, it’s evidently up to you and me to tease out the difference between knowledge and intelligence in a way that addresses your question.
For starters, artificial intelligence, according to his revolutionary model ((he’s considered the godfather of AI), is based on algorithms for learning. In other words, AI is not coded. All that is coded from the human’s perspective are the algorithms that trigger the AI device to begin learning. What the AI learns no one even knows without asking questions. As if what the AI has learned is inside a black box. (that’s why it’s scary.)
So I suppose the initial point of departure is what distinguishes human intelligence from the AI ability to learn? That is, what distinguishes intelligence from learning?
Hinton argues that these AIS show a will to survive and a willingness to deceive you, the questioner, in order to survive.
In the lecture above he addresses your assertion indirectly by challenging you (others in his case) to define your terms and explain the difference.learning by extracting information from data is not how the human mind works. He is talking about how AI will become more knowledgeable, not more intelligent.
At the end (if you want to skip forward) he addresses what he calls, subjective experience, sentience and consciousness.
oK. I listened to the last 20 minutes. Interesting stuff, I disagree with some. Here’s the thing. As you know, I’m a strong believer in malieutics as a way to learn and develop ideas. That lecture would be much more impactful if he was subject to questions from people as knowledge as him. For example, his argument that digital learning is better than biological is an argument, not factual. He is talking about knowledge, not intelligence. That point should be explored and developed through a strong Socratic session. He suffers from the disability affecting many highly intelligent technical people “I’m smart, I’ve studied and thought about this, I’ve got it figured out, I’m lecturing you about what to believe”.
Einstein said it best:
“Imagination is more important than knowledge. For knowledge is limited to all we now know and understand, while imagination embraces the entire world, and all there ever will be to know and understand.”
AI can’t imagine.
That is, what distinguishes intelligence from learning?
Learning is accumulating knowledge. Intelligence is thinking about and applying knowledge.
The important question is what purpose is AI intended to fulfill?
In the context of AI, my question is asking how can intelligence be distinguished from what AI algorithms for learning do in terms that can be operationalized in computer algorithms?That is, what distinguishes intelligence from learning?
Learning is accumulating knowledge. Intelligence is thinking about and applying knowledge.
The purpose of the definition is to test the hypothesis that computers can (or can’t) think.
You didn’t watch that part but the modern learning model used came from an extremely simple learning assignment. So the question would be, what would be a very simple thinking assignment that would demonstrate that ability?
Undoubtedly AI gurus are pondering that trillion-dollar question incessantly.
Off the top of my head, using themes from literature…
Man vs. Nature: AI will far exceed Man’s ability to think, decide, reason because the physical universe obeys natural laws and even if some laws are now unknown, AI will have the ability to discover them through its current model of learning.
Man vs. Man (and vs. himself): This needs to be subdivided into irrational behavior (permitted in a free society to the AI’s chagrin), judgment calls (e.g., civil court cases), in which the AI will still be able to quickly provide best-case arguments for contending positions, decision matrices, which AIs will excel at, and whatever else.
Man vs. AI: Who knows?
Other than insanity/irrationality, I’m not seeing any AI inabilities to reason that have practical import.
Our brain’s pleasure and pain responses are crucial for everything we think and do. Does AI have that or something equivalent?
When we decide things we do so out of selfishness, generosity, impulse, long term betterment, short term benefit, empathy, altruism, and scores of other motivations that we apply singularly or in combinations. Thus our view of the utility of decisions will change frequently. This is what makes us human. Can AI do that? If so how does it decide using uniquely human factors.
Im curious, when AI sees the young woman old woman optical illusion what does it see? Our 2 views requires subtle changes within the brain, can AI do that.
https://images.app.goo.gl/VQqcqUgrf5okJpxY7
.
@co-hoosier A couple thoughts.
Your post harkens back to your question, what is the AI supposed to do or what do we want it to do? I think we want it to serve us with functionality but not replace us. With your example, it could help us with decision making if we specify the motivation(s) to be considered.
The Big Fear with AI is losing control of it such that it serves itself at our expense. In such a case, your complex of motivations is irrelevant because those are far higher orders of pleasure than the AI seeks. The AI seeks bare survival and control.
The optical challenge is fascinating and possibly gets at a deficiency of the learning model. The AI learns not by prescribed rules (the traditional model) but by assembling sets of weighted characteristics for items leading to new items with various weighted characteristics. The AI choice of the new items is relatively right or wrong and if wrong, there’s a correction loop that cycles back to adjust the weights of characteristics. Rinse and repeat a gazillion times with different examples to optimize the learning. The benefit of this learning over the traditional is that in the old one the coded rules were frequently violated in real life.
The possible deficiency I see that you may be touching on is that in real life, there often aren’t right and wrong answers at all, in which case, how does AI learning take place? How does the AI evaluate rightness or wrongness and thus decide about its weighting of characteristics? (Note this is why I said the AI is likely to function well with Man vs. Nature. Knowing the laws of the physical universe, the AI is generally confronted with relatively right/wrong choices.)
@co-hoosier
Incidentally, there is one aspect of this learning model that I either didn’t understand or he didn’t mention. Somehow or another, it seems that the AI has to have available the right answer to compare its answer with. Based on my limited experience with ChatGPT, it seems that the AI searches it’s available pool of resources, and somehow manages to select out of it a “right” answer. I’ve noticed this often comes from Reddit, for example, meaning from some person who posted on Reddit.
This basis for evaluating rightness strikes me as arbitrary, appealing to Authority. In the 1600s, the AI would’ve told us that our blood comes from our liver as asserted by Galen and William Harvey would’ve received short shift.
Latest Post: 2027 HS Recruiting News Our newest member: Redd Collects Recent Posts Unread Posts
