Wie nah kommt die KI der menschlichen Intelligenz? – Große Sprachmodelle wie o1 von OpenAI haben die Debatte über die Erreichung künstlicher allgemeiner Intelligenz (AGI) elektrisiert. Es ist jedoch unwahrscheinlich, dass sie diesen Meilenstein aus eigener Kraft erreichen werden.
https://www.nature.com/articles/d41586-024-03905-1
18 Kommentare
From the article
>The release poured fresh fuel on a debate that’s been simmering for decades: just how long will it be until a machine is capable of the [whole range of cognitive tasks](https://www.nature.com/articles/d41586-023-02361-7) that human brains can handle, including generalizing from one task to another, abstract reasoning, planning and choosing which aspects of the world to investigate and learn from?
>Such an ‘artificial general intelligence’, or AGI, could tackle thorny problems, including climate change, pandemics and cures for cancer, Alzheimer’s and other diseases. But such huge power would also bring uncertainty — and pose risks to humanity. “Bad things could happen because of either the misuse of AI or because we lose control of it,” says Yoshua Bengio, a deep-learning researcher at the University of Montreal, Canada.
>The [revolution in LLMs](https://www.nature.com/articles/d41586-023-00340-6) over the past few years has prompted speculation that AGI might be tantalizingly close. But given how LLMs are built and trained, they will not be sufficient to get to AGI on their own, some researchers say. “There are still some pieces missing,” says Bengio.
>What’s clear is that questions about AGI are now more relevant than ever. “Most of my life, I thought people talking about AGI are crackpots,” says Subbarao Kambhampati, a computer scientist at Arizona State University in Tempe. “Now, of course, everybody is talking about it. You can’t say everybody’s a crackpot.”
the problem is LLMs are top-down instead of bottom-up. They’re just repeating things they’ve heard in new combinations, but there’s no real comprehension happening. It’s a Chinese box.
Now, robots that can sense their environments and interpret them — they may not be able to communicate their knowledge, but they are more likely to „know“ things. Combine those two and you’ve got a stew going.
AI is not intelligent. Association ≠ reasoning. We need *at least* one other breakthrough to get closer to human-like intelligence.
Its all marketing hype. At the end of the day we’re only rearranging chairs on the Titanic. Novelty is often confused for advancement.
Perhaps AGI is just around the corner, the way fusion has been just around the corner for 20 the last years.
Llms are very far from any sort of intelligence. They don’t think. In fact without input plus reply there isn’t anything going on in there.
A big indicator for me will be an intelligence that can make its own decisions independently, without human input. And the ability to change, alter, or ignore its human-provided programming and instructions. And it has to have the capacity for growth, learning, and change outside of its designated parameters.
But like all other intelligent populations, AIs will never have rights unless they have the capacity to fight for them on their own.
It is a mistake to believe AI will converge with human intelligence. And, in fact, I do not think it should.
We have flaws and one of the most important among them is our inability to perceive our own flaws. The simple ability to bypass these flaws and provide us with a different point-of-view is more valuable and more useful than a simple copy of actual human intelligence skills – we have access to those skills already, and supply is growing.
I feel like I’m being gaslit in these discussions. I program with AI tools on and off nearly daily because in very specific, non difficult but tedious tasks I can move faster. But I know from experience around what point the complexity becomes too much to rely on AI without wasting time, and it’s not even close to what I would consider a Jr. developer. Even this new o1 preview which I’ve been using up to my account limits. Yeah it’s better, but these improvements are really not big steps anymore. And its long drawn out plans it gives you, tend to lead you down unhelpful rabbit holes that waste time, often more than the more focused 4o will. It’s useful but not at all in the ways it gets hyped to be
AI will never have human intelligence, it will always be based on facts and logic, unless it is programmed to use false information, given a flaw to introduce randomly, and have it always base its decisions on what is best for AI first then what’s best for others.
When people talk about AGI, they often talk about the smartest human at doing a particular task, or at least humans in the top 90th percentile of doing that task.
I would argue that AI today is smarter than most humans at most things, at least for stuff that can be done through a chat interface. I mean, based on my personal interactions, AI is pretty amazing at a super-broad range of topics. I check-in before visiting the doctor, and it says about what my doctor did. I check in on a complicated legal problem before visiting a lawyer, and it says pretty much what my lawyer says. I share some spreadsheets and strategy documents on a business that my friends asked for some advice on, and it’s observations were spot on. My friends (husband and wife who run a business together) said that my conversation with ChatGPT sounded pretty much like their dinner table conversations.
So I think we are at or beyond the intelligence of most humans on most tasks. AI certainly isn’t better than all humans, but was that really the definition of AGI? I thought AGI was „human level intelligence“, and was super-intelligence to go beyond that. AI today is certainly smarter than **most** humans at a given random task.
Depends on what level of „intelligence“ you’re measuring. There are those in society that have maybe a single brain cell to function and *maybe* an ai could be as smart as them and claim *“human-level intelligence“* only because these people who would have been sorted out through evolution and darwinism long ago continue to thrive. Now compared to actual functional emotionally and intellectually healthy people ai might as well be the parrot that learned to scream profanities around its owners because that’s all it really is.
„Bagholding investors and lifetime researches desperate to convince people they are building AGI besides it being an obvious and expensive dead-end“
FTFY
the day AI becomes ”sentient” we would have already have to figure out the mysteries of conciousness, and with that discovery, sentient Ai will not be as big of a deal in comparison
I personally don’t think understand why AGI is considered the holy grail. Let alone how it’s expected to emerge from LLMs.
The strength I see currently is ML finely tuned to achieve singular goals such as protein folding, drug discovery, traffic flow (road planning & network design), etc.
And we do see great progress in applying the technology in those fields, but we still need humans to go through the results to confirm.
So currently, if our best singular results still need humans to confirm and go through the data, what chance do we have to create a AGI that excels at everything and can confirm its own results.
What would change my mind is an AGI that comes up with the grand unifying theory before we do. At that point I would agree to say, not only is it intelligent, it’s actually thinking.
Any software that argues that there are only 2 Rs in „strawberry“ fails a pretty basic Turing test. We’re not anywhere near any kind of intelligence with this.
Given we can’t pin down what exactly consciousness is yet and in philosophical terms we can only assume other humans are conscious because we are, it won’t matter if agi is actually thinking or can just convincingly mimic it. We won’t be able to tell the difference. If It quacks like a duck…
chatgpt was convinced that 32 – 6 = 29 for some reason 10 minutes ago