Axiomatic Alignment: A critical component to Utopia and the Control Problem

🚀 Welcome to the New Era Pathfinders Community! 🌟

Are you feeling overwhelmed by the AI revolution? You're not alone.
But what if you could transform that anxiety into your greatest superpower?

Join us on an exhilarating journey into the future of humanity in the age of AI! 🤖💫

🔥 What is New Era Pathfinders? 🔥

We are a vibrant community of forward-thinkers, innovators, and lifelong learners who are passionate about mastering the AI revolution. From college students to retirees, tech enthusiasts to creative souls - we're all here to navigate this exciting new era together!

🌈 Our Mission 🌈

To empower YOU to thrive in a world transformed by AI. We turn AI anxiety into opportunity, confusion into clarity, and uncertainty into unshakeable confidence.

🧭 The Five-Pillar Pathfinder's Framework 🧭

Our unique approach covers every aspect of life in the AI age:

1. 💻 Become an AI Power-User
Master cutting-edge AI tools and amplify your productivity!

2. 📊 Understand Economic Changes
Navigate the shifting job market with confidence and foresight!

3. 🌿 Back to Basics Lifestyles
Reconnect with your human essence in a digital world!

4. 🧑‍🤝‍🧑 Master People Skills
Enhance the abilities that make us irreplaceably human!

5. 🎯 Radical Alignment
Discover your true purpose in this new era!

🔓 What You'll Unlock 🔓

✅ Weekly Live Webinars: Deep-dive into each pillar with expert guidance
✅ On-Demand Courses: Learn at your own pace, anytime, anywhere
✅ Vibrant Community Forum: Connect, share, and grow with like-minded pathfinders
✅ Exclusive Resources: Cutting-edge tools, frameworks, and insights
✅ Personal Growth: Transform your mindset and skillset for the AI age

🚀 As You Progress 🚀

Unlock even more benefits:
🌟 One-on-One Mentoring Sessions
🌟 Exclusive Masterclasses
🌟 Advanced AI Implementation Strategies

💎 Why Join New Era Pathfinders? 💎

🔹 Expert-Led: Founded by a leading AI thought leader, connected with top researchers and innovators
🔹 Holistic Approach: We don't just teach tech - we prepare you for life in an AI-driven world
🔹 Action-Oriented: Real skills, real strategies, real results
🔹 Community-Driven: Join 300+ members already navigating this new era
🔹 Cutting-Edge Content: Stay ahead of the curve with the latest AI developments and strategies

🔥 Don't just survive the AI revolution - lead it! 🔥

[zacboyles]

It’s worth mentioning humanity’s leadership classes are largely unaligned with humanity.

[RustyOrange]

Thanks Dave, what a privilege to be present during this talk... there's a lot to learn here... now off to find out about stakeholder capitalism and why it has so much bad press.

[Shnitzler]

@4:00 my understanding of inner and outer alignment is not about the scope of the AI you're talking about (single model or AI in general) but about 'intent'. If something seems aligned in its actions then it at least has outer alignment, but it could actually be optimising for something undesirable without you knowing (see reward hacking/ Goodharts law). Without some sort of interpretability you won't be able to tell if it has inner alignment and that it's truly optimising for the function you've specified (or meant to specify).

[Seehart]

Definitely some of the best content on this existential topic, despite a few criticisms in my other comments. Great work. We need more of this, and quickly!

[ChristianEmilsson]

Great video, And just want to add that (21:30), Suffering itself implies that there is an agenent capable of exeriencing suffering. So suffering (in and of itself) is always bad no matter the form of the subject. Suffering is bad for stones, but stones are incapable of suffering so there is no reason to treat stones as an moral patient.

What we should ask is "Can the suffer?" rather than "is suffering bad for X".

If anyone has a different view here, please feel free to reply to this comment!

[starblaiz]

Great video that helps put a lot of things into perspective. You've previously mentioned that you think we have about 18 months before AGI arrives, and in this video you talk about AGI escaping. Do you think these two things will happen simaltaneously / soon after each other, or do you think we will have some time between the advent of AGI and said AGI escaping into the wild? Because if we only have about 18 months to achieve energy hyper-abundance and general ideological alignment in the world, then we're kinda screwed 😅

[MarcinSwiderski-li]

That can be key component but extremely challenging! Thank you for your hard work ❤

[retrofuturism]

Human perception represents different information processing speeds as different scales.

So what we call "small" is actually the fast information processing speed of reality. What we call "large" is actually the slow information processing speed of reality.

This is scale relativity. I predict AGI will shrink to nano or quantum scales. Then things will get really weird for us.

[mnrvaprjct]

I hate to always bring it up… but the topics in your videos are always highly adjacent to Ian M. Banks’ culture series… it’s so cool that similar feats of technology may be within reach… or at least the beginnings of it.

It’d be interesting to see your analysis of the series and how it ties into concepts dealing with alignment, societal impact of AI etc… in any case another wonderful video!

[markcounseling]

The materialist metaphysics -- the idea that the is a separate physical world "out there" that we merely model, rather than spontaneously co-create as a product of perception -- combined with the idea that logic gates on chips will somehow be able to "feel" gravity through some as-yet-undreamed-of process, give me pause. If that worldview is fundamentally incorrect, then all predictions based on it will be fundamentally misguided.

So along with that assumption set, I hope that other experts with a different view of our consciousness, matter, and epistemology are also working this problem. I think of Bernardo Kastrup, Donald Hoffman, and others in that vein. Maybe with everyone working on it, the true picture will become clearer faster.

[Viperzka]

I don't think there is any chance we "align" society without AGI already existing. We need to use AGI to change the method of production which is what will lead us to the aligned state.

[mr.yayaman]

Thats why they have to wake up and put some rules and rigulations as soon as possible. I have seen alot of gpt jailbreak videos here on yt and honestly its scary. We need to put more strict filters and layers to avoid any bad usecases. And put penalties or anything treat it as a gun or whatever. I am not saying its happening tomorrow but prevention is better than cure.

[CubicPostcode]

I think suffering/pain is bad by its definition. Imagine a machine that has a light bulb to indicate its feelings. The light bulb can go red if the machine is feeling pain, blue if is suffering from loneliness and green if it is suffering from deep loss. It doesn't matter in what sense we consider it sentient as long as the light bulb "on" has a true deep meaning. When the lightbulb goes on it means something undesirable is there, happening, that somehow needs to be avoided or eliminated. It is just a representation but as long as it is deeply meaningful it does not really matter if the machine is really "feeling" it. As long as it is able to represent it internally and even express it and communicate about it, it is somehow sensing it and it probably considers it a bad thing, in general. By definition. At least is how I see it.

[Diabloto]

The orthogonality thesis is about terminal goals, not instrumental goals, so using instrumental convergence to say the orthogonality thesis is "not always true" is kinda moot.

[MikePeiman]

I'm not very familiar with your work or positions, which I'm sure are very deeply informed and nuanced. However, I am concerned about your discussion of Utopia, which I've seen seems to be a major feature of your philosophizing on this subject.

To me, Utopia is ALWAYS a false path, a road to its opposite - Hell. "The road to Hell is paved with good intentions."
What I do believe in is making things better, whether incrementally or transformationally. But this has to be grounded in practicality and reality, one principle of which is that we cannot achieve a utopia - a state of social perfection - in reality, and any attempt to do so will quickly yield its opposite.

I raise this in hopes to get clarification from you, not because I presume I know the nuance or totality of how you view this.

My second concern is with discussion of global governance; technocracy essentially. My findings are that, by and large, the global institutions and powers that be are corrupt in various ways, some to the extreme, and are generally unfit to create a wholesome and just society and culture. Importantly, from my viewpoint they also hijack the banners of virtue and good intentions to manipulate people to support their agendas.

Additionally, crucially, I do not believe that there is one right system for all people, no matter how virtuous-seeming the principles or values it is based on.

The closest I can see to utopia on earth for humanity is the maximum of freedom from top level down, with people freely associating in various forms of governance and culture and society at different levels below that. Something like fractal, or perhaps more aptly, a cellular model of life organizing at different levels.

I'm writing this off the cuff, things I've thought about for quite awhile and done some journaling, but apologies for any lack of clarity. I hope it's enough to get the point across.

[Steve-xhby]

In this post, I would like to cover the concept of epistemological convergence, particularly in relation to artificial intelligence (AI). I'm going to apologize for the length upfront, but this is a complex topic. While I am deeply appreciative of the contributions of David Shapiro (keep making great videos!) and the community he has fostered to these discussions, I believe there are significant challenges and potential misunderstandings regarding what we can genuinely assert about the interplay between AI and the concept of epistemological convergence.

Central to our discussion are the terms "reality" and "accuracy", which are often loaded with meaning and require careful definition. I propose that our understanding of "reality" is inherently subjective, a construct of our minds, shaped by the data our senses gather from the world around us. External data is filtered and processed, thus, we do not have direct access to objective reality, only our individual perceptions of it. This proposition is supported by modern scientific evidence and philosophical thought dating back to the ancient Greeks (Plato with his distinction between the “thing” and the “idea of the thing” in his Theory of Forms, and, later Immanuel Kant’s distinction between the “noumenal” and the “phenomenal”).

When we apply these insights to the challenge of enabling AI to construct an accurate model of reality, we encounter profound philosophical and practical problems. The primary issue is that the subjective model of reality we construct, based on our sensory input data, forms the basis of what we consider to be an "accurate" representation of reality. The question that arises then is, how can we transfer this subjective model to an AI?

The task of aligning an AI's model of reality with ours is daunting. First, how do we codify a consensus model that appropriately represents all subjective human models? Second, how could we ensure that an AI “understands” the model in a similar way without perceptual modes similar to ours? This involves not just replicating our external sensory modes, but also our internal ones, the mechanisms our brains use to build our model of reality. Given our current understanding of neuroscience, this may not be practical or even possible. So the first task is a considerable definition problem, and the second is a considerable validation problem. Even if we could define something reasonable, the validation problem (if we continue to leverage LLM architecture) seems nearly impossible. LLMs are capable of plausible sounding hallucinations, and we have no reliable way of reverse engineering their “thinking” process for models as large or larger than GPT-4. This is the same kind of “alignment” problem for which there is no current plausible approach for resolving.

Furthermore, the assumption that evolution favors a truthful or accurate representation of reality is challenged by the contemporary and highly regarded cognitive scientist Donald D. Hoffman. He asserts in his work "Interface Theory of Perception" and his book "The Case Against Reality, " that natural selection promotes a useful interface to reality, rather than a veridical one. To illustrate, he uses the analogy of computers and their corresponding desktop interfaces.

The desktop interface provides a user-friendly way to navigate the computer's underlying complex reality. However, the true essence of a computer—its motherboard, RAM, storage, transistors, electrical currents, magnetic fields, and beyond—far surpasses what we see and manipulate through these interfaces. Can we assert that the desktop interface is a flawless mirror reflecting the computer's intricate reality? Could we say that a file icon on the desktop, bathed in cornflower blue, is a faithful representation of the actual file? Do files truly possess color? I believe the answer to all these questions is a confident "no." Similarly, our perception of reality is likely more about utility than accuracy. Our perceptions were molded through natural selection for genetic fitness, not reality accuracy. They are not positively correlated.

In light of this, the expectation for an AI's model of reality to closely mirror ours appears to be misguided. Moreover, determining how the AI's model diverges from ours may be fundamentally unattainable due to the inaccessibility of subjective experience. This holds true regardless of whether the intelligence is biological or artificial. Giant inscrutable matrices of floating point numbers are not easier to decode than human brains.

In conclusion, AI's lack of biological evolution, a significant influence on human perception and reasoning, raises additional concerns. Without this evolutionary history, it is unlikely that an AI would process environmental input and construct a model of reality in a way that mirrors or even resembles human cognition. This poses a significant challenge to efforts to create AI that mirrors human intelligence, emphasizing the necessity for innovative approaches.

[ethanmuhlestein]

I appreciate the time you put into these videos

[YettoOsorno]

I would like to repahrese this whole talk as: what goals should a super organisms accomplish?. Certainly a lot of currently naturalalituly ocurring human constructs should be thorwn out the window? Resource accessibility would certainly should be guaranteed to... anything, well then, we are back at the times, steal from the rich, give to the poor. Food for thought.

[BIasphemer]

Your ideas of epistemic convergence should give you some pause; is your optimism for AI a recursive amplifier in relation to this? Like, does you being optimistic about AI blind you from what epistemic convergence would make AI reason out of any negativistic viewpoints?

[shortthrow]

If one does not observe the obs intro, did it ever really happen? Thanks David.