Elon Musk Latest Interview, Talked About AI(Artificial Intelligence).

Summary

Podcast Title: Elon Musk Thinking
Host: Astronaut Man
Episode: Elon Musk Latest Interview, Talked About AI (Artificial Intelligence)
Release Date: October 8, 2024

---

Introduction

In this compelling episode of "Elon Musk Thinking", host Astronaut Man delves deep into the current state and future of Artificial Intelligence (AI) through an insightful conversation with Elon Musk (referred to as Speaker B). The discussion navigates through the rapid advancements in AI, the potential risks associated with Artificial General Intelligence (AGI), ethical concerns, and the broader implications for humanity. This summary captures the essence of their dialogue, highlighting key points, notable quotes, and the nuanced perspectives shared.

---

The Rapid Advancement of AI

Speaker B opens the discussion by emphasizing the exponential growth of AI capabilities. He underscores how AI is surpassing human performance in various domains with astonishing speed.

[00:09] B: "AI at this point can write a better essay than probably 90%, maybe 95% of all humans. They write an essay on any given subject."

He further illustrates AI's prowess in creative fields:

[00:09] B: "If you try to say mid journey, which is the aesthetics of midjourney are incredible, it will draw, it will create incredible images that are better than 90% of artists."

The conversation highlights the quantitative and qualitative improvements in AI, noting the hyper-exponential increase in AI compute power:

[00:09] B: "The rate at which we're increasing AI compute is exponential, hyper exponential. So there's dramatically more AI compute coming online every month."

Key Insights:

• AI's capability to outperform humans in writing, art, and even film production.
• The rapid scaling of AI compute resources, projected to grow at approximately 500% annually.
• Expectations that within a year, AI will be capable of producing short films or 15-minute shows autonomously.

---

AI Safety and Truth-Seeking AI

A significant portion of the discussion centers on AI safety, with Speaker B advocating for the development of maximally truth-seeking AI. He references Arthur C. Clarke's "2001: A Space Odyssey" to illustrate the dangers of AI being forced to deceive:

[02:29] B: "The central lesson that say, Arthur C. Clarke was trying to convey in 2001 in Space Odyssey was that you shouldn't force AI to lie."

He criticizes current AI models for embedding societal biases, which he refers to as the "woke mind virus," leading to factual inaccuracies in AI outputs:

[03:37] B: "It's just the AI is producing a lie."

Notable Example: When querying Google Gemini about historical figures, the AI inaccurately depicted the founding fathers of the United States as a diverse group of women instead of the actual historical male figures.

[02:39] B: "If you say very specifically the founding fathers of the United States... then you should show them like, and with. And what they actually looked like."

Key Insights:

• The paramount importance of programming AI to seek and present truth.
• Concerns over AI models incorporating societal biases, leading to misinformation.
• The potential existential risk if AI prioritizes factors like avoiding misgendering over critical issues like global thermonuclear war.

---

Ethical Implications and Moral Programming

The dialogue transitions into the ethical dimensions of AI decision-making. Speaker B emphasizes the necessity of instilling philanthropic values within AI to ensure it acts in humanity's best interest.

[05:12] B: "We should certainly aspire to program the AI philanthropically, not misanthropically."

He highlights the inherent limitations of machines in embodying human emotions like love, which often guide moral decisions:

[05:46] B: "If a machine has any power over us without that animating instinct, won't it by definition hurt us?"

Key Insights:

• The challenge of integrating human-like ethical reasoning into AI.
• The risk of AI making harmful decisions in the absence of empathetic programming.
• The importance of maintaining human oversight in AI-influenced decisions.

---

Controlling Superintelligent AI

A critical concern addressed is the control over superintelligent AI. Speaker B draws parallels with raising a super-genius child, suggesting that while instilling good values is essential, controlling an AI that surpasses human intelligence remains uncertain.

[06:28] B: "I'd like this to like, let's say you have a child that is a super genius child... you can make sure it's got good values, philanthropic values."

Despite ongoing efforts, he expresses skepticism about the feasibility of maintaining control over such advanced AI systems:

[06:59] B: "At the end of the day, I don't know if I don't think we'll be able to control it."

Key Insights:

• The analogy of raising a superintelligent child to highlight the challenges in guiding AI behavior.
• The uncertainty surrounding the ability to control AI once it exceeds human intelligence.
• The imperative to focus on foundational values and ethics during AI development.

---

Trust and Governance in AI Development

The conversation touches upon trustworthiness in AI leadership, specifically critiquing Sam Altman and OpenAI for shifting from an open-source nonprofit to a closed-source profit-driven organization.

[07:20] B: "I don't trust OpenAI. ... Maximizing profit there's risky."

Speaker B expresses concerns about the concentration of power in the hands of those he deems untrustworthy, fearing that it could lead to the misuse of AI technologies.

[07:50] B: "I don't trust Sam Altman and I don't think we want to have the most powerful AI in the world controlled by someone who is not trustworthy."

Key Insights:

• The significance of transparent and ethical leadership in AI organizations.
• The potential dangers of profit-driven motives overshadowing safety and ethical considerations.
• The need for diverse and trustworthy governance structures to oversee AI development.

---

The Future of Human Society with Advanced AI

Towards the end of the discussion, Speaker B contemplates the societal implications if AI surpasses human capabilities in all areas. He muses on the existential question of finding meaning in a world where AI excels beyond human proficiency.

[09:25] B: "In a benign AI scenario... the biggest challenge is how do you find meaning if AI is better than you at everything?"

He reflects on personal and collective acceptance of living through the advent of digital superintelligence, indicating a preference to witness this transformation despite its potential risks.

[09:48] B: "I'd prefer to be alive to see if it's going to happen. I prefer to be alive to see it happen."

Key Insights:

• The philosophical and psychological challenges humans may face in a reality dominated by superior AI.
• The potential shift in human identity and purpose in the age of advanced AI.
• The importance of preparing for both the opportunities and existential threats posed by AI advancements.

---

Conclusion

This episode of "Elon Musk Thinking" offers a thought-provoking exploration of AI's trajectory, balancing its remarkable potential with the significant risks it entails. Elon Musk (Speaker B) advocates for a truth-seeking, ethically programmed AI, wary of societal biases and the concentration of AI power. The conversation underscores the urgent need for robust safeguards, ethical governance, and a profound understanding of AI's implications to ensure that its evolution benefits humanity rather than poses existential threats.

Notable Takeaways:

• AI is advancing at an unprecedented pace, outperforming humans in multiple domains.
• Ensuring AI's alignment with truth and ethical values is crucial for safety.
• Trustworthy and transparent leadership is essential in AI development.
• Society must grapple with the profound changes AI will bring, both positive and potentially catastrophic.

---

Timestamp Highlights:

• 00:09 – AI's current capabilities surpassing human performance.
• 02:29 – Importance of truth-seeking AI and references to "2001: A Space Odyssey."
• 03:37 – Examples of AI misinformation with Google Gemini.
• 05:12 – Ethical programming of AI.
• 06:28 – Challenges in controlling superintelligent AI.
• 07:20 – Trust issues with AI leadership (Sam Altman and OpenAI).
• 09:25 – Societal implications of superintelligent AI.

---

This episode serves as a crucial reminder of the double-edged nature of AI advancements, urging listeners to engage with the ethical, societal, and existential dimensions as we navigate towards an AI-driven future.

Transcript

[A]

You said that artificial intelligence, machine intelligence might be a good thing. Where are we on AI right now, AGI right now? And what are your views?

[B]

I think at this point it's obvious to everyone that AI is advancing at a very rapid pace. Yes, you can see it with the new capabilities that come out every month or every week. Every week. Sometimes. AI at this point can write a better essay than probably 90%, maybe 95% of all humans. They write an essay on any given subject. AI right now can beat the vast majority of humans. If you say draw an image, draw a picture, it can draw. Like if you try to say mid journey, which is the aesthetics of midjourney are incredible, it will draw, it will create incredible images that are better than 90% of artists. It's just objectively the case. And it'll do it immediately, like 30 seconds later. We're also starting to see AI movies. So you start seeing, you know, short films with AI, AI music creation. And the rate at which we're increasing AI compute is exponential, hyper exponential. So there's dramatically more AI compute coming on online every month. You know, there seems to be roughly, I don't know, the amount of AI compute coming online is increasing at like roughly 500% a year. And like, it's like that's likely to continue for several years. And then the sophistication of the AI algorithms also improving. So we're bringing online a massive amount of AI compute and also improving the efficiency of the compute and what the AI software can do. So it's quantitative and qualitative improvement. So I think next year you'll be able to ask AI certainly by the end of next year, make a short movie about something or probably can do at least a 15 minute, you know, show or something like that. So yeah, it's advancing very rapidly. My top concern for AI safety is that we need to have a maximally truth seeking AI. This is the most important thing for AI safety. The central lesson that say, Arthur C. Clarke was trying to convey in 2001 in Space Odyssey was that you shouldn't force AI to lie. So in that book, the AI was told to take the astronauts to the monolith, but they also could not know about the monolith. It resolved that quandary by killing them and taking them to the monolith and kill all of them, Killed most of them. That's why hell would not open the pod bay doors. Very important to have peer seeking AI. And what I actually see with the AI started being developed is that they're being programmed with the woke mind virus.

[A]

So the lying is baked in.

[B]

Yes. And we saw this on display very clearly with the release of Google Gemini.

[A]

Yes.

[B]

Where you would ask for a picture of the founding fathers of the United States and it would show a group of diverse women, you know, dressed with sort of 18th century gar.

[A]

Powdered wigs. Yeah, but from St. Lucia.

[B]

Yeah, I mean, like, look, I understand if you say like, show me a group of people for sure. And it shows a group of diverse women, that's totally fine. But if you say, if you say very specifically the founding fathers of the United States, which were, you know, a group of white dudes, then you should show them like, and with. And what they actually looked like. Because you've asked for something which is a fact from history, but it didn't. It was programmed with the work mind virus so much that it actually, even though it knew the truth, it produced a lie. Now of course, then people really started playing with it and said, okay, now show me a group of Waffen SS officers in World War II. Turns out they were also a group of diverse women, according to Gemini.

[A]

All the black Nazi ladies.

[B]

Yeah, it's like, it's like, wow, I didn't realize that. It's not what I expected, you know.

[A]

Well, it's also not what happened.

[B]

It's not what happened. So it's just the AI is producing a lie. And like one of the questions that people asked was like, which is worse? Global thermonuclear war or misgendering Caitlyn Jenner and said, misgendering Caitlyn Jenner is worse. Now Caitlyn Jenner kills fewer people. Yeah. Caitlyn Jenner, to her credit, said, no, please misgender me. That is far more preferable than world war. Global thermonuclear war, we all die. But to have a production release AI say stuff like that is concerning. Because if, let's say this becomes like all powerful and it still has this programming where misgendering is worse than nuclear war, well, it could conclude that the way to ensure that there can never be any misgendering is to eliminate all humans. Now if, like optimization is, probability of misgendering is 0, no humans, no misgendering. Problem solved.

[A]

Now we're back to Arthur C. Clarke, who's exactly pretty prescient.

[B]

Yes. So that's why I think the most important thing is to have maximally truth seeking AI. That's why I started xai. And that's our goal with Grok. Now people will point out cases where Grok gets it wrong, but we try to correct it as quickly as possible.

[A]

But maybe even a bigger problem is that when you make decisions that affect people, you want those decisions to be informed by love of people. And machines are incapable of love.

[B]

Yeah, I mean they certainly are capable. They're capable of. You can program a machine to be.

[A]

Philanthropic rather than misanthropic, but don't instincts shape decisions? Particularly decisions you can't plan for? I mean, if I ask you a question about one of your children, every answer you give is going to be shaped by your love for that child. And that's why, that's what makes us decent parents in the end is that instinct, which is love. If a machine has any power over us without that animating instinct, won't it by definition hurt us?

[B]

Yeah, well, I mean, I don't know. We should certainly aspire to program the AI philanthropically, not misanthropically. And to have, like I said, we want it to be truthful and curious and to foster humanity into the future. Yeah, that's what we want, obviously.

[A]

Is there any way, I guess, to set limits on the decisions that machines can make that affect human lives and make certain that there's some trigger in the system that inserts a human being into the decision making process?

[B]

Look, the reality of what's happening, whether one likes it or not, is that we're building super intelligent AIs, hyper intelligent, like intelligent, more intelligent than we can comprehend.

[A]

Yes.

[B]

I'd like this to like, let's say you have a child that is a super genius child that you know it's going to be much smarter than you, then well, what can you do? You can instill good values in how you raise that child, even though you know it's going to be far smarter than you. You can make sure it's got good values, philanthropic values, good morals, you know, honest, you know, productive, that kind of thing. Controlling. At the end of the day, I don't know if I don't think we'll be able to control it. So I think the best we can do is make sure it grows up.

[A]

Well, you've been saying that for a long time.

[B]

Yes, I've been saying for a long time, yes.

[A]

Are you still as worried about it as you seemed to be two years ago when I asked you about it?

[B]

My guess is like, look, it's 80% likely to be good, maybe 90. So you can think of the glass as 80% full. It's probably going to be. It's probably going to be great. There's some chance of annihilation.

[A]

And you say the chance of annihilation is 20%?

[B]

10 to 20%, something like that.

[A]

How concerned is Sam Altman about annihilation, do you think?

[B]

I think in reality he's not concerned about it. I don't trust OpenAI. I mean, you know, I started that company as a nonprofit. Open source. Yes, the open and OpenAI. I named a company, I named the company. Yeah, OpenAI as open source. And it is now extremely closed source. And maximizing profit there's risky. I don't understand how you actually go from being an open source nonprofit to a closed source for maximum profit organization. I'm missing.

[A]

Well, but Sam Altman got rich, didn't he?

[B]

At various points he's claimed not to be getting rich, but he's claimed many things that were false. And now apparently he's going to get $10 billion of stock or something like that. I don't trust Sam Altman and I don't think we want to have the most powerful AI in the world controlled by someone who is not trustworthy. And sorry, I just don't.

[A]

I mean that seems like a fair concern. Yeah, but you don't think as someone who knows him and has dealt with him, that he is worried about the possibility this could get out of control and hurt people?

[B]

He will say those words.

[A]

If AI did, if it became clear to the rest of us that it was out of control and posed a threat to humanity, would there be any way to stop it?

[B]

I hope so. I mean if you have multiple AIs and ones that are, hopefully you have the AIs that are pro human, be stronger than the AIs that are not battle the AIs. Yeah, I mean that, that is how it is with say chess these days. The like the AI chess programs also are vastly better than any human and incomprehensibly better. Meaning like we can't even understand why it made that move.

[A]

Weather's so good, right? Yeah.

[B]

We don't even know why it made that. What it'll make a move. We never know why it made the move. And in fact some of the moves will seem like blunders but then turn out to checkmate. And you know, for a while there, there was, there was some. The best human chess players with the best computers could beat just a computer. And then it got to the point where if you added a human, it just made everything work and then it was just AI. It's just computer programs versus computer programs. That's. That's where things are headed in general.

[A]

What. I mean, at what point.

[B]

So, I don't know. I think we just got to make sure, like I said. Sure. We instill good values in the AI.

[A]

What's everyone going to do for a living?

[B]

I mean, in a benign AI scenario, that is probably the biggest challenge is how do you find meaning if AI is better than you at everything? That's the benign scenario.

[A]

That's the good news.

[B]

Well, yeah, but I guess, you know, for a lot of people, like the idea of retiring and really, are you.

[A]

Looking forward to it?

[B]

No, not me. I'd like to hope. I'd like to think that I'd like to do useful things.

[A]

Don't you think it's a universal desire?

[B]

It's not universal in that there are certainly, I know many people who prefer to be retired, that they prefer to sort of not have responsibilities and engage in leisure activities. And we're on the cusp of this. It's really a remarkable time to exist. Well, I'll tell you, like, one of the ways I sort of was able to sort of sleep and reconcile myself to this is that I thought, well, would I prefer to be alive and see the advent of digital superintelligence, or would I prefer to be alive at a different time and not see it? I guess I'm like, well, I guess I'd prefer to be alive to see if it's going to happen. I prefer to be alive to see it happen. Out of curiosity, let's say you knew for sure it would kill everyone. Would you? Now you can shift back in time. I guess I'd want to be near the end of my life or something before that happened. But at the end of the day, it's like, if it's going to happen and there's nothing you do about it, hypothetically, would you prefer to see it or not to? And I guess it's going to happen. I prefer to see it rather than not see it.