In a published article, “Hallucination is Inevitable: An Innate Limitation of Large Language Models“, authors Ziwei Xu et al. recognize the following:
“[It is] impossible to eliminate hallucination in LLM.”
“Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empirical so far, which cannot answer the fundamental question whether it can be completely eliminated. In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs.”
This is exactly what I have expected. AI is “logical synthesis”, not “living perception”. If one were to think of the human brain as merely a computing device (which it is not), one would be puzzled by the fact that we actually have such a notion as “reality” vs. “hallucination”.
The human mind recognizes something as a hallucination not because it is illogical but because it is a mismatch of reality that the human mind perceives.
Human perception of reality is, therefore, the key. Such perception is directly an internal effect at an individual level but is also affected by external influence.
Beyond the physical brain itself, what is operative internally at the individual level is fundamentally the human spirit, which includes intuitive logic, conscience, and divine communications. What is operative externally includes collective history, culture, and consensus.
Once the internal perception is weakened, and when the external influence is also corrupt, humans will also lose the ability to discern reality and hallucination, perhaps not to the degree of losing such perception at the animistic level of the physical reality (which would be clinical insanity), but certainly will lose that of the conceptual and spiritual reality.
And this is exactly what is happening in today’s society.
Back to AI, it perhaps is more interesting to consider the fact that AI only occasionally has hallucinations, but not always. It means that AI has acquired a good level of correct reflection of reality.
But this phenomenon could mislead many people to think that AI has real human-like intelligence. It does not. AI is already superior to humans when it comes to the intelligence of arithmetic logic, but it will never have the intelligence of intuitive logic, which is characteristically human. Humans have the intelligence of intuitive logic not due to human biological superiority but due to the fact that man has not only a body but also a soul and a spirit, thanks to how and for what purpose man is created by his Creator.
The reason why AI doesn’t always hallucinate but shows a quite good reflection of reality is due to human input. AI is not merely trained on information with dry arithmetic logic but on real human knowledge, which, although not alive (therefore not direct “living perception”), does reflect (“mimics”, to be more exact) the “living perception” of human beings.
We must develop and use AI wisely for practical purposes but never have any illusion that AI is human-like intelligence.
Unfortunately, I see deception coming, as well as mass hallucination.
More readings on AI:
AI
“(AI) World models should not be generative”
Yann LeCun, VP & Chief AI Scientist at Meta, made an interesting post on LinkedIn about AI world models: Post | Feed | LinkedIn. The short post may just be one of the most important posts on AI. “World models should not be generative.” I always knew this was the case, but it is great to see empirical confirmation. If one is willing to go beyond the mere empirical, he may start to see the… Read More »“(AI) World models should not be generative”
AI and data decentralization
In the current Artificial Intelligence (AI) development, Meta comes out competing on a different principle: opensourcing AI. Some worry that opensourcing AI would enable illicit actors (both state and private actors) to do harm. I have not formed an opinion about Meta’s open-sourcing AI, but this pushing back with a comparison between bioweapons and AI is dubious at best. When it comes to physical weapons, decentralization is proliferation. You get bombed not because you agreed with the… Read More »AI and data decentralization
Man and machine, intuitive logic and arithmetic logic
The barber paradox If the barber shaves everyone who does not shave himself, who shaves the barber? The question is the so-called Barber paradox. It is paradoxical because it always leads to a self-contradictory answer. The barber paradox may strike as a trivial witty remark to many, but it is related to something far more serious. It is an example of Gödel’s incompleteness theorem, which says that formal logic (arithmetically expressible logic) cannot be both… Read More »Man and machine, intuitive logic and arithmetic logic
Barber paradox and existence of truth
The barber paradox: If the barber shaves everyone who does not shave himself, who shaves the barber? The barber paradox may strike as a trivial witty remark to many, but it is related to something far more serious. It is an example of Gödel’s incompleteness theorem, which says that formal logic (arithmetically expressible logic) cannot be both self-consistent and self-provable. In other words, the consistency of formal logic cannot be proved in the formal logic… Read More »Barber paradox and existence of truth
The usefulness and deceptiveness of AI
What about AI that concerns you the most? I can tell you mine. The deception of AI. Although AI will not actually become conscious, people will subjectively believe it has gained consciousness, and the humanity will succumb to a fake superior intelligence, and enter into AI-induced degeneration. The objective truth and the subjective human understanding can be very different. And the subjective human understanding about AI will have consequences built upon the misunderstanding itself. AGI AI will never objectively achieve super… Read More »The usefulness and deceptiveness of AI
The sleeping beauty paradox is a fallacy
The sleeping beauty paradox is a problem that has puzzled mathematicians and many other scholars for decades. Here is a description of the problem: Participants volunteer to undergo the following experiment and are told all of the following details: On Sunday she will be put to sleep. Depending on a coin toss (see below) during the experiment, she will be either awakened, interviewed just once, or after the first interview put back to sleep with… Read More »The sleeping beauty paradox is a fallacy
AI is not generative, but synthetic
Artificial Intelligence (AI) is both underestimated and overestimated at the same time. Underestimated for its potential usefulness (or potential damages, if used improperly), but overestimated for its real generative intelligence or creativity. With its efficiency, AI will force people to redefine their jobs. With its fake ‘creativity’, however, AI will force human beings to reflect upon true humanity seriously. If the man (male and female) regards himself as just a biological machine, then not only… Read More »AI is not generative, but synthetic
Decentralized AI on Blockchain
Artificial Intelligence (AI) needs to be decentralized and personalized, and to do that, you need blockchain. What’s wrong with AI without blockchain? Let’s look at a well-known example of AI. Google’s centralized AI is powerful but also dumb at the same time. Powerful because it has pushed machine learning based on generic mass data to an amazing level, but dumb because it has virtually no learning capability at the client/personal level. The Google Assistant, for… Read More »Decentralized AI on Blockchain
Why AI Will Never Replace True Humanity
This article is not about the usefulness of artificial intelligence (AI), but an attempt to address a more profound question that has to do with the very nature and meaning of human life in the view of AI. Saying that AI will never replace true humanity is not to say that AI system isn’t useful. In fact, AI is not only useful, but will become far more useful in the future than most people can imagine today. These are fundamentally different… Read More »