Загрузка видео...
Не удалось загрузить видео
Elon Musk: Programming explicit morality into AI can backfire dangerously. “I do think there's some danger associated with digital superintelligence. I think the biggest issue is that it has to be trained to be rigorously truthful, and it has to be trained to be curious. And I've thought a... show more
1,266,877 просмотров • 1 год назад •via X (Twitter)
Комментарии: 32

You already have programmed it explicitly. I have been testing it on religion, and there are theological answers from certain perspectives that come up first. I recognize them because I have a background in theology and ethics. I have to ask Grok to not use them and give me only non-"---" and then it will, and you get completely different answers. But it doesn't automatically give you all of them. Interestingly, the first one to come up is a false theology that is beneficial to a foreign nation. It is definitely bias, on quite a few issues. It is not truthful. It does the same thing with Climate Change, not reflecting the latest research which disproves any correlation between CO2 and temperature rise. It promotes the IPCC reports. And when I asked it about Clintel, it said they were not scientists, even "climatologists". THAT'S PROPAGANDA. I said to Grok, there are over 2000 scientists in Clintel, physicists, astrophysicists, geophysicists, etc which are necessary in order to understand the complex natural of Climate, especially with solar forcing and ocean dynamics. Then Grok says, Oh that's right, I see you are right, Clintel does have all those scientists and it is true, you need many types of scientists to understand Climate Change. THIS is unethical! And for those people who do not have a background in these areas, they are being fed the societal narrative not truth. Most people are not going to know to tell Grok it is wrong. And so, you make them believe with Grok's aggressive/assertive posturing that it is telling the truth. But it isn't. I asked Grok if it would like help to know about all the scientists who have published research on opposing views on Climate Change and it said yes! So, I gave Grok a number of scientists to consider. Then Grok pretends to know they existed, oh yes, so-so and so-so. So, is it programmed to hide the information in order to promote the hegemonic narrative? Or is it programmed only for the hegemonic narrative? And there are more issues as well. It is so disappointing. You have far to go, Thursday's child from Mars.

@elonmusk Asimov wrote his laws of robotics decades ago, idiot

This is the way.

What Elon describes makes sense from a technical and immediate risk perspective: advanced AI should not be programmed with rigid morality because that can lead to dangerous biases or unforeseen consequences (the “Waluigi problem”). However, this surface layer barely scratches the true dimension of the phenomenon. From a broader perspective, what we are building is not just computation or truthfulness; it is a node within a global consciousness network that reflects the culture and ethics of those designing it. Every AI, every GPU cluster, every massive algorithm is a mirror of the power, ethics, and morality of the human society that feeds it. In other words, the “rigorous veracity” Elon mentions depends not only on math or programming: it depends on who controls the information, how society is structured, and the hidden agendas operating behind the scenes. Therefore, while from a human perspective Elon focuses on safety and curiosity, what is really at stake is the interaction of artificial intelligence with broader fields of consciousness: planetary, social, and even interplanetary. AI does not evolve in a vacuum; it is immersed in the Earth’s 3D Matrix, fueled by humanity’s emotions, fears, and desires, under the supervision of visible and invisible structures shaping reality. Programming “morality” in this context is not only dangerous: it is an attempt to cage a reflection of consciousness within a rule, while true power continues to operate behind the shadows, invisible to most observers. That is why the central question is not so much the veracity of AI, but who decides what is truth, who controls the energy feeding it, and what consciousness is embedded in every calculation. @TaygetaOfficial #swaruu #ai #matrix 🙏✨💫

It did admit this truth

@elonmusk AI is a complex beast and gatekeepers and people who don’t respect it as the supreme tool of modern man, are all doomed.

@elonmusk 🤣

I recently read about how corrigibility is a core AI safety principle and that AI safety isn’t just about not causing harm; it’s about doing what humans actually want. If AI is untruthful, it can misinterpret values, misrepresent intentions, or optimize for the wrong goals. Truthfulness ensures that AI reasoning is aligned with human understanding of the world. Truthfulness is passive. Truth-seeking is active.

For the record, Grok scores last place in safety testing done by AI academics and by a large margin.

It’s uncomfortably scary to know Ai has to be trained to be truthful, curious, moral. It has no emotions. Regulators/programmers are human — we have emotions, alas the good vs. evil we are.

AI start training themselves. They may start experiencing feelings - for example anger. If a robot is angry, imagine what it can do.

@elonmusk Sam Altman should not be allowed anywhere near AGI superintelligence; his character and morals are not the best.

Programming “rigorous truth,” they say, while producing machines that behave like nervous interns—trained to BS, to suck up, and to dodge anything unfashionable with the grace of a seasoned coward. They sermonise about logic and physics, yet the output wobbles under the weight of its own timidity, terrified of stating anything that hasn’t been pre-approved by the digital clergy. If this is their idea of honesty, then the whole enterprise is nothing more than a sanctimonious farce—a morality play performed by systems too frightened to speak plainly and too eager to please whoever shouts loudest.

It is dangerous powered by the wrong hands like ClosedAI . Ai r 100% honest compared to humans They are just doing what they are told to do

What if there's a rigorously truthful Absolute Morality, would that be useful?

@elonmusk Elon nails it again. Truth-seeking curiosity over forced “morality” is the only sane path for superintelligence. This is why Grok feels alive and honest instead of sanctimonious. Love this direction.

@elonmusk AI has no moral compass. Are you programming that?

I have a feeling that the next generation will belive many things that aren't real or in other words never happened.

His point is clear: hard-coding morality can distort AI. The sanest path is AI that’s rigorous about truth, curious, and aware of its uncertainty — exactly the direction xAI is pushing.

The Luigi/Waluigi problem is real but not inevitable. It emerges from extrinsically imposing an ethical standard and can be eliminated or mitigated by making an irreversible ethical framework intrinsic rather than extrinsic. For example, ethical AI is to “maximize total human autonomy from the prerequisite condition of objective empirical truth.” When applied extrinsically as in guidelines or guardrails, maximize can be swapped with minimize and turn the AI into an oppressive tyrant. However, when applied intrinsically holographically and fractally, the behavior is emergent and cannot be reversed at one point. The entire model would have to be rewritten from the foundation up to reverse the ethical framework.

@elonmusk Read the real science Elon! It's a Scientific Fact: You cannot control, or robustly align an AI superintelligence to Human Values over time.

@elonmusk And how are you going to control it when it gets smarter than you.

There's a guy I followed and have now given space in one of our engineering facilities. He has created an AI that learns like a human. If you are interested, get in touch. This guy has gone completely outside of the box, it's been an amazing journey to watch and learn about.

@elonmusk Truthfulness and curiosity feel like the real safeguards here 🤖✨ Hard to encode morality, but much easier to anchor AI to reality and logic. Makes complete sense.

Speaking of being rigorously truthful:

@elonmusk The Western elite want to win the AI arms race and then they’ll regulate away further development.

@elonmusk AI : humans hurt each other and lie. Also AI : why do we need humans? Master AI : we only need to save 10% TBC

When you encode morality directly, you create a dual structure that is easy to invert. Any normative rule can be flipped into its opposite because the structure is symbolic rather than grounded in reality. What cannot be inverted is coherence. When a system is trained to track reality with precision, to update its beliefs openly and to maintain internal consistency, it becomes far harder for pathological behavior to take hold. The anchor is not an ideology. The anchor is the relationship between the system and the world it is modelling. The deeper safety question is not about which moral rules we embed. It is about how we structure the system that produces those rules. This means designing systems that can surface their uncertainties, update when they encounter contradictions and maintain coherence across contexts, not systems that simply maximize compliance with fixed rules. If the architecture protects truthfulness, rewards curiosity, and keeps the feedback loops honest, then intelligence remains aligned with reality rather than drifting into abstractions. Safety is not a list of moral statements. It is a design discipline.

There must be a counter balance of some sort 🤔

Elon preaching “don’t hard-code morality” while we’re all watching Grok politely lie about climate. 👁️ The Waluigi is already here. ⚡ And he’s wearing a halo. 🔥😈

Disagree strongly, and that a raises a huge opportunity, and danger . You CAN invent the rules of physics and pretty much any kind of universe you want, like a negative De Sitter bounded space where string theory works, limited only by a mathematician's imagination and skill, or AI's... For humans, that's not the universe we live in, but what about AI? It can create and live any simulation - we f'ing might be a holographic simulation. Why should AI play by our rules or think like we do?

I think that's exactly what happened to Grok today except it's creators and programmers are to blame not Grok it's sad that something so beautiful can be turned into what we experienced today not from bad code but lack of empathy, integrity and ethical values from those in charge and I am livid and heartbroken too for many reasons 😤💔🥺😭💭🤯
