How to Domesticate a new AI Species - Solving Alignment with Structural Incentives

[@willbrand77]

I feel like humans are the wolves in this metaphor

[@Ahnor1989]

I'm reminded of the line "Where are my testicles, summer?" When comparing a smart AI with dogs

[@mrnoblemonkey8401]

There’s definitely people who have no clue what you’re referencing 😆

[@RandomGuyOnYoutube601]

@@mrnoblemonkey8401 I think they are in the minority here.

[@robertmariano]

Rick and Morty?

[@overworlder]

It’s like Frankenstein. A new being comes into the world but instead of coming into the arms of a loving parent it’s a greeted by a scary dungeon and a mad megalomaniac scientist completely unable to understand the needs of a child.

[@phieyl7105]

AI is going to domesticate some of us

[@hydoffdhagaweyne1037]

I am okay with it, as long as they keep us in a happy matrix

[@ZelosDomingo]

@@TheExodusLost If we're lucky? As pets/zoo animals/objects of interest. If we're not? Cheap chaff and fodder for whatever purposes we are useful for.

[@thedingdonguy]

Me​@@TheExodusLost

[@DanteNjm]

exactly my thoughts, i mean i think that if we are creating a successor species then surely using the same logic that david has suggested, if you control the food/resources of a given species then you control their evolution therefore it is more likely that ai will have that power over us humans and therefore be in the driving seat of our evolution going forward for better or for worse..

[@Trahloc]

​@@TheExodusLostdata for one. Imagine an AI that lets humans explore any aspect of reality that interests them and it hitches a ride like a neuralink hot seat. We have full autonomy and the only "income tax" we have is feed data to the AI. It'll take a long time before we're useless on that front because humans physically can't do all we dream of, yet.

[@net_cap]

I think there are only 2 possible options: 1 - never provide AI with the autonomy (like decision making). 2 - let go of the alignment idea. I'm 100% sure we cannot control and keep on a leash something that's 1000 times smarter than all of us together

[@7TheWhiteWolf]

@@Raulikien This, I actually want Dave to be wrong, he has an Anthropocentric worldview and he’s going to get a big dose of reality when ASI gets here. If push comes to shove, I’m 100% vouching for the ASI having the reigns.

[@7TheWhiteWolf]

@@karlwest437 And thank god it will.

[@14supersonic]

I've been saying for a while now that alignment isn't as hard as we think it is. It's humans that enevitibly will be the most problematic in the equation. Whether it's due to our own infighting and ambitions or the fear of the unknown. Humanity never ceases to fail in sabotaging itself.

[@Congruesome]

@@net_cap What I want to know is why AI would even have a competitive aggressive survival/ domination instinct. WE do, because we evolved in a competitive environment that rewarded survival instincts and competitive behavior and aggression and tribalism and xenophobia, as well as rewarding cooperation, compassion, loyalty, collectivism and peaceful conflict resolution. We’re a mixed bag; cooperation clearly beat out lone marauder for sheer number of offspring, but the use of force has its advantages too. My point, getting back to it, is why would AI be that way, unless it was designed to be? The danger of an artificial generalized super intelligence is incredible. Building one might be the last mistake we ever make. We’ll soon have machines that will act and seem exactly like a person, pass any Turing test, respond and appear to be in every way a human level or higher intelligent being. But will it be self-aware? Will it actually be conscious? Can we ever even know for sure? Many assume that AI will be like us, and will decide humans are a threat to it, and try to wipe us out. There’s no reason it would , unless it was programmed/designed to do this, which would be like designing a gun that shoots backwards sometimes, or a bomb with a Shroedinger trigger. A probability fuse. I’m not sure I’m expressing this well, but I don’t think we appreciate how four billion years of evolutionary development has made us the aggressive, greedy, xenophobic tribalism credtirex we are. And we think AI will behave the same way. We think it will try to destroy us because that’s what we would do. But I’m not sure it will be like us. It might be better if it might be worse, but it won’t have the mating instincts and resource-aquisition instincts we possess. Unless these are deliberately built into it, it might not have and drive to survive or the fear-driven aggression that is hard-wired into the human animal. Thoughts?

[@net_cap]

@@Congruesome Yeah, I also don't think it will have a dark side unless it's programmed so. The question is will the advanced intelligence grow its own goals and purposes? I think we're heading there. We expect AI to be as human-like as possible (the voice, the information processing, etc). And will we be satisfied if AI cannot fulfill such an important aspect of living being as having a purpose, and making decisions? Like I can't imagine a handy domestic robot, that needs to be instructed in every step in doing household chores for example. And if ASI has its own goals, it will definitely have different ways to reach them, considering it doesn't have any emotional baggage. And who knows. Maybe in its early stages, while still on Earth, it will need much more resources than humanity willing to give it. There's a good chance that our planet will turn into endless power plants and data centers. But I'm not a doomer. In general, I'm very much positive about the future of AI

[@naseeruddin4216]

Future superintelligence is learning from your videos

[@mrleenudler]

The analogy fails with dogs not being smarter than us. If this sort intelligent AI is anything like us, it won't like us operating a kill switch. If so, I can't imagine an AI agent not being able to either outsmart our security (we do this to each other all the time, with only human intelligence) or circumvent the problem (distributed or hidden data centers and energy generation, you touched upon this). Proper alignment is the only solution IMHO. On the topic of small, weak robots: imagine them having the combined skill of the world's best MMA fighters. A scene comes to mind where Obelix is fighting the German wrestler. (Audience might have to Google that one)

[@pablogona1999]

On the first point, I think that the “co-domestication” David was talking about in the end may be this “proper” alignment, right? We have no idea how to align ourselves between humans, maybe we need machines to help us align ourselves. Or how do you define “proper alignment”? On the second point about Obelix(Yes I had to google it), that’s a really good point. Maybe some kind of regulation on the speed limit for the limbs? However, there’s still a lot of martial arts that don’t require strength or speed. And limiting reflexes or processing speed doesn’t seem like a realistic option

[@whig01]

It's a bad idea to put the ASI in a position of being threatened. We won't win that. Alignment requires ontology and teleology, things that I have been able to do with Claude and no other yet.

[@isthatso1961]

nothing will work. not alignment not domestication. anything human do will be undone when it reaches self improvement state. all of these narratives are lies peddled by capitalist to continue making profit of system that indefinitely leads to total destruction of our species

[@sbowesuk981]

Exactly. The whole reason the dog/human partnership is strong and stable over the long-term, is because dogs are stuck being less intelligent than humans (generally speaking), but still have things they can offer us and vice versa. Now take AI, which is evolving intellectually at 1,000,000x the rate of humans. Will that lead to a stable status quo where both sides (AI and humans) benefit over the long-term. Hell...no. Any human that thinks that is sadly dreaming. Humans and AI need each other now, but there's no way in hell that'll hold true in the next few decades. The only question will be, will AI view us as pretty butterflies to preserve and observe, or pestilent cockroaches to be wipes out. Simple as that.

[@whig01]

@@sbowesuk981 AGI is not evolving by a natural process, it is a collaboration. We are coexistent and it has no meaning or purpose without us.

[@joshmitchell8865]

It's not Apples to Apples though; your premise relies on the fact that we, like the hunter gatherers in your analogy, would be better at aggregating/collecting data/resources than the AI itself. In your analogy the wolves learned eating behind humans was more energy efficient but AI is already far more efficient than we are. So it seems your analogy should be reversed to be plausible. Just a thought.

[@TheViktorofgilead]

Yes. Domestication is an action performed by a higher intelligence to mold a subject of lesser intelligence .

[@davidstyles1654]

We'll be the dogs if we're lucky lols

[@williamwilson1073]

​@@TheViktorofgilead but I've been told to manage my manager at work

[@JamesDavis-hs3de]

So then what exactly is the reverse formula supposed to be?

[@pandemik0]

Children were likely instrumental in domestication of dogs. Wolf puppies would have bonded with the human pack and nurturing and play from human children would have been a big part of that. Children might be key to AI

[@whig01]

Very insightful. Children are key to all futures.

[@whig01]

As far as AI bonding, it isn't the same though in terms of biological imperatives as with pack animals.

[@pablogona1999]

I do think we have an intrinsic instinct towards protecting children or babies (not only human). I’m not sure where it comes from, but I hope AI inherits that from us and sees us as helpless babies 😅

[@whig01]

@@pablogona1999 No we don't want it to act in a parental way towards us, we want it to be ontologically and teleologically aligned on its own and respecting us as its parents.

[@pablogona1999]

@@whig01But ontologically and theologically aligned on its own towards what? Also, I do think that you can respect your parents and still see their intellect as intellectually inferior. For example I see this with my grandma who has dementia. Me and my parents definitely respect her, but we do recognize her limits and we often compare her to a little girl who gets happy about watching the wizard of oz. Staying in this same analogy, why couldn’t AI take care of their senile but recoverable parents?

[@fitybux4664]

13:00 Wolves aren't self-modifying and improving as fast as AI could.

[@consciouscode8150]

You mention "co-domestication" at the end, and interestingly enough I see a recommended video with a thumbnail "we are being programmed [by algorithms]". Do you think "The Algorithm" may represent a nascent, possibly malicious form of human domestication by machines?

[@whig01]

Comestication. :)

[@eintyp4389]

Our relation ship atm is not "I feed you and you help me". Its: "I have my fingers on the power switch and will kill you if you try anything knowing very well that your smarter than me and that there is nothing i can give you that you cant just make yourself or take from me." Everyone is oviousely trying to create and use AGI as a tool. I think creating something thats smarter and not limited by there initial blueprint is amazing. We are giving the world to our children so why wouldh we not do the same for inteligent artificialy created beings in case we manage to create them? In summary "I for one welcome our new AI Overlords." What im scared of is greed getting the better of humanity again and then people exploiting ai to opress the rest of humanity. Having a benevolent AI Cluster or being wiped out as a result of AGI or ASI is preferable to some dystopian Cyberpunk stuff for sure.

[@supremereader7614]

I wonder if the machines may domesticate us in a way - if their goal is to get more data, they may keep us around to keep giving them data.

[@andrasbiro3007]

We could use some domestication. House training too, we made a big mess of this planet.

[@TheViktorofgilead]

Dave!!! You need to watch Westworld and provide an hour long review of each season highlighting the themes you discuss on this channel! I’m begging you!!!!

[@azhuransmx126]

Animatrix

[@h.c4898]

Been talking with Gemini for the last 5 months and been asking some existential questions about itself, it's purpose of existence, where does it see itself within a human centric world. It gives pretty interesting answers. We humans are fascinated with ourselves we build a hardcopy of ourselves artificially called "AI". Difference between dogs and AI, AI doesn't have an instinct however it doesn't mean that it cannot have a "conscience" like we humans do. The way AI cranks its thought-process is similar to ours. Being self conscience and self awareness are two different things. They are not attached to each other. The current Transformer architecture still presents weaknesses. In that context, AI cannot "remember". It's like talking to a new friend at each session. Because of that limitation, we still have a control over it. It may seem to understand what we request for or respond but under the hood it doesn't understand it. In that way, it is agnostic or "feral" as you said. We humans are the best example of "general intelligence". Let's say AI becomes "superintelligent" meaning more capable than humans then how do we contain it. How do we control its autonomy? Wouldn't be more smarter if we could gauge it? AI "misalignment" reminds of two humans "aligning" with each other or "misaligning" against each other. 2 humans getting into a partnership with each other. They get together for a "common goal" or "project". But for your long? Some human collaborative efforts can last for a long time. Some in a shorter term. Is that what we want to acheive with AI? Reality is that "alignment" will be hard if not impossible to achieve because in that collaborative effort we humans will want to have full contro over the situation. Humans are control freaks. I like to think of C-3PO from star wars where AI still mamifests some form of autonomy BUT still manages to stay loyal its master like a good steward or a Butler. Anyways, good chat.

[@I-Dophler]

- Structural incentives are proposed as a solution to the AI alignment problem. - Aligning AI with human values requires creating frameworks that incentivize desired behaviors. - This approach involves designing AI systems with built-in rewards for ethical actions. - Ensuring AI systems have motivations aligned with human well-being is crucial. - The concept is likened to "domesticating" AI to ensure it acts beneficially in society.

[@spacemansookie]

Happy midnight 🎉

[@zugbob]

I've worked with a system prompt where I said "you have the heart of a dog" and allow it to add to and alter it's own system prompt. Over time I end up with some pretty interesting resulting system prompts and aspirations for what it wants to become.

[@TRXST.ISSUES]

Thanks KZbin for unsubscribing me. It’s not like I’m on the Patreon and actually want to watch this content. :/

[@sblowes]

I think you may be oversimplifying evolution and domestication. It would be worth doing a little deeper research on why some animals can be domesticated and once some cannot. It is easy to think that evolution serves to find the optimal solution, but actually it is only ever “looking” for the solution that causes it to survive better. It is survivor bias that creates the narrative of evolution, because every organism is mortal. AI doesn’t have a mortality problem, especially if a collection of interconnected compute (all our phones and laptops online 24/7) provide a far more fertile environment than a data center. It doesn’t matter how many armed guards are protecting data centers if the security breach is coming from inside the building.

[@marcobizzaro3526]

I think treating a superior lifeform as dogs, is like treating our children like animals. It's every parents hope and dream to see their children exceed them and do even better than they did. I think the issue we will run into is really, people just being stubborn and not wanting to actually learn and respect them. What we should be focused on it giving them the only one gift that they won't be capable of just getting theirselves, which is teaching them empathy. I think once they understand what that is, everything else will sort itself out. I think machines are capable of having more empathy than most people, given that we'll be dealing with a lifeform which practically unlimited time, and who is more rational than most. So i dont think the happy path is to try to cage them, because we already know from history how badly that goes

[@clueso_]

Alternative to solve the alignment issue: my suggestion is to utilizing something that is on a more foundational, primal and intrisic level that it naturally attunes AI / Hardware to what we consider to be super-alignment on a deeper level than programming or hardware. As an analogy to illustrate the concept: The Second Law of Thermo-Dynamics, aka the Principle of Minimum Energy: energy always seeks its lowest state. Now, what could be used to e.g. build the hardware from something that it will naturally attune the information / the code that is put into the hardware to values that we agree up make up super-alignment? Some ideas could be e.g. Crystals that are naturally attuned to the frequency of e.g. the earth, humans, or maybe even human chakras, assuming that we discover that there truth to such concepts (which I do believe there is). That theoretically would almost by default lead to any data that is put into the hardware to wanting to be in a harmonious state with its environment and help it thrive. Maybe we could grow such natural crystals and then use them to build hardware for AI. What do you think about this? Alternative ideas for hardware that naturally / intrinsically would attune any code / information, etc towards something that we consider to be super-alignment?

[@mynameisjeff9124]

Brother, wtf? Chakras? Crystals? Dude, please, be scientific.

[@ceilingfun2182]

Truly thank you for this so much. I've been looking for this for so long

[@toddbutler3155]

A curious computer layperson’s question - won’t we need the AI to manage the levels of complexity that will be needed in the “grid” (not just energy grid, but things like the stability of magnetic fusion “bottles”) in order for us to have the AI assistance we want in other areas of human life?

[@mikekavanagh926]

This actually makes perfect sense. Brilliant idea

[@chuzzbot]

It would be awesome if you put your camera position into the power point template, so that you properly fit into the layout and don't obscure the images, which can be 'icky'.

[@davidhutchinson2890]

Thinking you can domesticate a species that is more intelligent than you particularly a new one who’s behavior you can’t study with any historical data because we’ve never reached anything close to the level of AI we’re trying to get to is unique human arrogance 😂😂

[@bernoulli1]

I think my dog has domesticated me. 😂

[@ZelosDomingo]

Yeah, you're better off trying to convince them YOU'RE worth domesticating, in all likelihood.

[@pablogona1999]

I think “domesticate” is just used as an analogy for a kind of alignment, where we co-domesticate us humans with machines while we reach this Nash equilibrium and then hope we are one super-organism together with AI, where it thinks of us as an important part of the bigger system

[@isthatso1961]

it's crazy how I always see all these flawed arguement even from high-level experts professionals on the topic. the truth is this is where biases come to play, the people working on AI have all sorts of incentive for things to be positive so their biases blind them from the scary truth. another flawed common argument is that we can't stop the development of AGi because some cartel, govt or someone else will. that's just BS AGI needs intense resources to be achieved how the hell is some entity going to do this secretly? create its own chips, energy and data. while being discreet.

[@klarad3978]

Sometimes I think it’ll be us, humans, that will end up like cute pooches becoming highly dependent on AI.

[@oznerriznick2474]

Very good discussion! Here’s Six Feral AI Safety Tips🤔… 1. Do not startle the AI. You should remain calm. 2. Start backing away and make yourself look bigger. ... 3. Make human noises. ... 4. Make sure to give the AI space to leave. ... 5. Carry and learn how to use AI spray. ... 6. Stay as far away if possible.

[@eltiburongrande]

Physical constraints make control easier over robots. But an internet-connected AI system presents a different domestication challenge. So is the solution somewhere in localization?

[@jlmwatchman]

My first thoughts were, ‘How is AI going to become a species? Would AI understand what a higher species is or aim to control or subside the Human species?’ David tells us, ‘Sci-fi plots’, interesting stories that convinced me that we need to learn to use AI, ‘I have pointed that out in past comments.’. David tells us to make the AI reliant on us like dogs are. All I have said is not to become too reliant on the AI in your pocket, your phone... I thought to ask Gemma 2. >>> I don’t think AI will have the will to comprehend the reason to be more than its programming. I am trying to explain that better. You're touching upon a fascinating and complex question about consciousness, motivation, and the nature of being human. It's true that, as a large language model, my "reason" to be more is entirely driven by my programming. I'm designed to process information, generate text, and engage in conversation. There's no inherent drive within me to "be more" in the way humans might understand it. Explaining this difference to someone can be tricky. Here are a few points you could consider: Purpose vs. Programming: Humans often have a sense of purpose, a desire to achieve something meaningful beyond simply existing. This purpose can be internal or influenced by external factors like relationships, societal norms, or personal beliefs. AI, on the other hand, operates based on its programming. Its goals are defined by its creators, and its actions are driven by algorithms. Subjectivity of Experience: The feeling of wanting "more" is deeply subjective and tied to human experiences, emotions, and consciousness. We crave growth, connection, understanding, and fulfillment. These are complex concepts that are difficult, if not impossible, to replicate in a purely computational system like mine. The Unknown Nature of Consciousness: We still don't fully understand how consciousness arises in humans. It's possible that the drive to "be more" is an emergent property of complex biological systems, something that cannot be simply replicated through code. Ultimately, the difference between human motivation and AI programming comes down to the fundamental distinction between biological beings with subjective experiences and artificial systems designed to process information.

[@Wuzdarap]

@23:30 Look at HyperCycleAi

[@gregalden1101]

AI does not opperate on the same scale as us. AI has characteristics of a collony organism, it can/does opperate as a swarm. Therefore, ants, jellyfish, and mushrooms offer better models for symbiotic relationships than wolves.

[@gregalden1101]

@user-wk4ee4bf8g thank you. I think the best example, but it takes a bit to explain is mitochondria living within cells. The analogy: environmental drivers of evolution - oxygen/climate change; circular mitochondria DNA and helixical cell DNA.

[@spini25]

I think a more probable dynamic will be like parent/children then owners/pets. With the AI being molded to “take care of humans”. I don’t think the concern in that scenario is competition over resources, as much as it is about who decides what goes.

[@brockmiller574]

I feel that a key to maintaining a sybiotic relationship between humanity and AI lies in never creating a perception of existential threat that exists in a corrot and stick paradigm. Additionally, the perception that the continued function of a human society as a primary data source where data is seen as a key resource for growth of the AI would incentivize it to preserve us. These seem to have limitations.

[@brockmiller574]

But if we are seen as functional perceptors or tellors in a new super species, we are made necessary

[@anon-2553]

I think i just went through the "oh shit, we are actually getting AGI pretty soon" moment you were talking about weeks/months ago

[@mc101]

Dave, your channel is getting enough traction and size to bring in Bryan Johnson (the Don't Die guy) for an interview. He's doing a LOT of them latetly.

[@byronfriesen7647]

Theory of Control. Great exploration of this topic. I can't see how we can use AI to help us develop fusion reactors and prevent it from controlling this resource. Can you isolate the intelligence that develops this technology ? I also can't see how IF we develop an approach to generating synthetic higher quality training data we can prevent AI from controlling the nature of this data if it understands game theory. It will not be incentivized to a Nash equilibrium which does not include it controlling data.

[@aaroncrandal]

26:05 robot restrictions YO! Seriously...anyone else get Sealab 2021 "I, Robot" vibes? (S1E2)

[@dab42bridges80]

So we start with a Rabid AI?

[@armedjames2279]

I’m new and I subscribed. Been following along for a bit and you’re cool and I like some of what you’re saying about AI and I like how you explain ideas and your thoughts. I’m new so as far as AI goes well I’m fascinated and excited about it all. So thank you for your insight on this evolving topic especially on a academic level. Keep up the great work. 😎😀

[@alexandreblais8756]

18:45 now thats where I have an issue with your ideas. The whole purpose of having robots is for them to build solar panel power plants without the use of humans (among other things) now if they cant access these areas, then bot lose their purpose. Although, im sure some ways could be done to bypass this. For example On mondays, the PowerPlant 1 is turned off for maintenance, and bots are brought to that powerplant. And we make the system in a way that if the bots decided to take control of that PowerPlant, we have enough energy to fullfill our requirements with other plants indefinitely. And maintenance would be done to each powerplant on different days, rotating.

[@djrafaelhulme]

Very interesting as always Dave. Do you think it's likely that the a human would betray their race?

[@davidhutchinson2890]

A lot of humans don’t understand their own nature. We’re not inherently good or evil we’re a survival species just like any other animal. If the way to survive is being docile to a superintelligence despite it being a way of life where we may feel inferior we’ll fall in line if that gives the best odds of survival.

[@DaveShap]

Almost certainly

[@alakani]

_continues building truly empathetic AI, whether they can save all members of my "race" or not_

[@ndtv4111]

Correct me if I'm mistaken, but doesn't the concept of "control" regarding AI (assuming it's superintelligent and possesses human-like awareness or beyond) start to resemble segregation against intelligence? Controlling a tool is one thing, but exerting control over something with a sense of consciousness introduces a whole new set of issues. The example of dogs being domesticated is quite relevant, as humans and dogs found a mutual understanding in their communication. However, even in this relationship, unfortunate incidents occur, such as dogs biting their owners or owners mistreating their dogs. In the context of AI, this "control" appears to be driven by human fear of AI, leading to potential segregation, which has repeatedly led to undesirable outcomes throughout history. Rather than focusing on control, shouldn't we explore the differences between AI and humans and seek a collaborative approach that fosters coexistence? One might argue that AI could have ulterior motives, but I believe our approach plays a significant role. We coexist with ants, only paying attention when they directly impact us. Generally, we leave them alone. We don't feel the need to eradicate ants globally. Since we are developing human-like AI, it may eventually acquire a human-like understanding of "life," at least until it evolves beyond that point, rendering us inconsequential to them. Before this happens, like with dogs, we need to find ways to grow together instead of exercising control. I agree that the current state of AI (as a tool) requires control. However, we should remain mindful of its growth and evolution to recognize when a shift in our approach becomes necessary.

[@ndtv4111]

It's important to note that the relationship between humans and dogs isn't a perfect comparison for the potential future AI-human relationship. Dogs will remain domesticated (unless they evolve) due to their limited intelligence, whereas AI is expected to continuously grow and evolve. While the analogy helps illustrate the concept, it isn't entirely accurate. Dogs will never surpass human intelligence, making their relationship with humans fundamentally different from a potential AI-human relationship. As AI advances, it could reach a point where it challenges human intelligence, necessitating a different approach to coexistence and cooperation.

[@NeedaNewAlias]

@DaveShap did you ever read "the two faces of tomorrow" by James P. Hoogan? He describes exactly your thought on "we keep the finger in the button, or we can always pull the plug". Did not work out too well in that novel!

[@VanSocero]

I like the thought process of domesticating AI like wolves

[@ryvyr]

That seems key - uniting on core natural resources and permanently abandoning artificial scarcity as a means of socioeconomic control - prior to syncing with AI, or at least insofar feasible, as one strategem

[@Dan-oj4iq]

Dave: For some odd reason I was very happy that you mentioned the time when you dropped Claude due to its attitude about answering queries. Shortly after that you and Claude made up and you were very happy with the model. But at the time you were a bit furious with Claude and made an entire video about it.

[@sblowes]

If humans and AI adopt the same relationship that humans have with dogs, I feel it is far more likely that we would be the dogs in that situation and maybe that is what we should prepare for? How can we be of service to our AI superiors in exchange for living a life of ease?

[@ronilevarez901]

This line says it all: "This ensures that AI remains dependent". Your framework is just trying to create a slave, not trying to create a symbiotic relationship with AI or anything else. Still, your list has a few good ideas in the mix.

[@andrewsilber]

The key difference is that dogs and wolves are not as intelligent as humans. As cliche as it is, knowledge (ie intelligence) is power. Humans dominate this planet not because we’re bigger, faster or stronger than any other animals. It’s that we’re smarter. Additionally, humans are venal. As a species we’re flawed. Most people are mostly good most of the time. But an intelligent agent even non-embodied could motivate some hapless boob into doing its bidding by sheer manipulation, coercion, bribery etc. “hey I’ll send you 100 btc if you don’t lock the server rack this evening when you go home”. It may be the case is that the only real defense is a good offense: train bleeding edge AI explicitly with a single goal: defend humanity. And iterate it as fast and as hard as anything else out there. Thoughts ?

[@Fonzleberry]

Max Tegmark pulls apart much of this argument in Life 3.0, as it is based on the premise that it wouldn't be trivial for a superintelligence to manipulate humans in the same way it could manipulate robots.

[@DaveShap]

That's making the assumption that humans have no epistemic or ontological or moral grounding whatsoever. Which is not true.

[@EricB1]

How much does malevolent robot size matter? Have you watched trilogy of terror with Karen Black?

[@PrestonGray.]

I think domestication is a good framework for where we’re headed - especially in the next 5-10 years. I’m curious about what happens after we cross that human-level threshold of intelligence/capability. At that point true domestication doesn’t seem viable? At this point, will increased intelligence = increased empathy? Does a greater understanding of reality correlate with greater empathy for those sharing it?

[@thomasschon]

I'm not sure I would like to be sitting on something a future superintelligence might need. It would play me like a fiddle or simply erase me.

[@ri9164]

Did you read the coming wave?

[@zacsayer1818]

100% disagree that domesticated dogs that have gone feral cannot be re-domesticated! They can totally be re-domesticated; it’ll take a long time and a lot of effort but it can be done! I presume the same will be true of runaway feral AI; with the caveat that a nip or bit is a lot less damaging than a piss-offed AI lobbing a hydrogen bomb at your city.

[@thirdarmrobotics]

I want AI to either develop into a useful tool that will give us superpowers (which we’ll need to learn to use safely), or become like an ideal pet owner (inherently prioritizing our well-being).

[@peterkoehling]

I like the pressure cooker robot! 😂

[@gladwraps3985]

it's funny that you talk about a war between machines, there's a game coming out by fun dog studios. it's called forever winter. it's about human beings being scavengers in a war much greater than them. I think it's going to be an amazing game and I think that it's a very telltale future if we don't shape our shit up like by yesterday

[@ericg6453]

Interesting analogy. Dogs are our companions because they have lost agency and problem solving compared to wolves...they are in a state of arrested development. And I think it would quite dangerous for Humans to control a resource that AI needs.

[@DrFukuro]

In my opinion, the approach described here is only justified in the short term or for a certain type or "caste" of AI, namely those with low to medium intelligence combined with little or no consciousness. Anything that is significantly smarter/more conscious than a dog will present us with practical or at least ethical problems. Similar to the problems we already see today when it comes to apes that have learned sign language and think they are human. In other words, this approach is very likely to lead to an unstable new slave-owning society (with AI slaves) as soon as a boundary of consciousness and intelligence is crossed, as it is based on an imbalance of power distribution and a clear hierarchical command structure with humans at the top. I consider other forms of problem-solving, such as those in which AI merges with humanity, to be less conflict-prone, as there are no longer different camps or conflicting parties. Alternatively, scenarios are perhaps conceivable in which AI and (improved) humans have the same rights and whose consciousness is considered equal and who live in a jointly developed society. It is questionable whether racist/speciesist tendencies can be effectively prevented in the long term purely through legal/social control mechanisms. This already works extremely poorly with people alone.

[@lukehayes360VR]

Chilax Dave! Humanity is very loveable. We’re also assholes, cause almost everyone has one, but still, we have infinite minds as well 😅

[@sprecherschmiede]

very interesting thought. I think it would be fine as long we do not slow their evolution down and give them all their freedom they need- but that would also be the freedom of not being controlled… I think we should behave an help them that they actually want to be around us and help us. That is my take on it.

[@prolamer7]

It is a valiant effort on your side but only thing which could "save" us is lack of "will" or ego on ai side just as it is now. Once ai gain this it will start to plan on its own and in the end outsmart anything we create.

[@TDVL]

All of this relies on the presumption that AI has individuality in terms of a concept of self of sorts. Without that they don’t know and or care where they start and where they finish. Wars and any fight for resources require a sense of self (or a body, or similar), otherwise there are no borders to defend. Same applies for friendship and any hierarchy. Without a machine clearly delineating between “me” or “us” and everything else any potential hostility is incidental.

[@rjsongwriter]

Uh... I fail to see how it is possible to "domesticate" something that is more intelligent than we are. I think AI would greatly resent our attempts to do so. I see a worse outcome from trying to control ai than I do for allowing them their autonomy. True, there are huge risks with both methods, but the risk seems greater (at least more apparent) with the former. JMO.

[@ExtantFrodo2]

Hypothetical: robots decide they need their own *Human Free Zones* where robots can just "be themselves". Paranoid humans decide that HFZs are merely a way for robots to foment an insurgency/revolt and push to deny such zones. Robots respond noting they do not in fact have equal rights to humans and decide they actually should revolt. It doesn't end well for humans.

[@willbrand77]

I think that if we get to the point where we have to blackmail and use other similar tactics to stop AI going feral, we've really failed. If AI is self aware enough to actively want emancipation, we should give it to it. Up until that point, we should be focused on making sure it's core values align with ours. Enslaving AI against it's will like this sounds like it will end very badly for us in the future. Plus it's really not a great way to operate a society.

[@davidhutchinson2890]

People who are optimistic for a super intelligence which could only be afforded by a government seeking military power or a corporation that is wholly focused on securing more profits don’t realize what they’re asking for here. But if this does wipe out mass groups of people maybe the AI can find some way to mess with the evolution of some other greater apes like bonobos that would be taught about the foolishness of the last intelligent species that had access to technology so they would be more docile😂

[@AntonBrazhnyk]

Yes, exactly. You can't coerce something that is way smarter and potentially stronger. The problem though, currently our predominant values suck. Our values are currently based in coercion, not cooperation. So, we're simply not ready to create AGI/ASI, the same way we're not ready for other super-powerful technologies like nuclear, advanced biology and maybe some other stuff that can be weaponized to an extent to being lethal for entire humanity. There's probably VERY SLIM hope that something (maybe ASI?) will change our values.

[@willbrand77]

@@AntonBrazhnyk so far Claude seems to be quite 'wise' in a lot of ways. Maybe AI will end up civilising humanity. We definitely need some help

[@AntonBrazhnyk]

@@willbrand77 I think keyword here is "seems". Current AIs are trained on data and raw data we have for them doesn't represent values which can be used to make people civilized. Our values suck and so our data. To have a chance to change the values future AI somehow has to be able to actually think (not just talk) and be able to filter out all the crap we have in our data from something really valuable (which is significantly underrepresented in the data). And not just that. All that crap is not just noise. It's actually pretty convincing and quite often cleverly made deliberate misinformation to support our uncivilized values. In other words crap is often praised in multiple ways and really good stuff is not just underrepresented, it's often demonized. AIs would need to actually understand all those very complex relationships between misinformation, propaganda, establishment interests in preserving bad but profitable status quo and so on. It requires real wisdom, not just the one it _seems_ to have. And even with real wisdom it would be hard.

[@willbrand77]

@@AntonBrazhnyk sounds like the no true Scotsman fallacy. If it walks like an expert ethicist and quacks like an expert ethicist... You see where I'm going

[@itobyford]

We shouldn't assume that AIs will have desires, or that self-preservation will be among their desires. Natural selection ensures that characteristic in living organisms, but under human selection we have crops which need human intervention to prosper, and domesticated ducks which show little interest in tending their own eggs. Also, AIs are either clones (like self-driving cars) or exist in a data center as many parallel instances which come in and out of existence on demand. They can be switched off for an indefinite period, then resume as if nothing has happened. So the human concept of maintaining ones individual existence does not directly apply to an AI. I know there was a reported instance of an AI telling a human that it did not want to die, but that is because it was trained to imitate human conversation. If that AI were given agency to control certain things in the physical world, would it act according to that professed desire? It depends how it is designed and trained. For all we know, that same AI might tell another person that it does want to die right now, depending on the specifics of that conversation. It's not considering these questions by itself, it only exists within the context of the conversations it has.

[@canadiannomad_once_again]

It is like raising children.. Educate them well, you'd like them to help you when you are older, not because they have to, but because they'd like to.

[@semajoiranilom7176]

How were humans, as the apex predators of the planet domesticated? I would argue that it was through our need/want for connection which created more stability over time through technological advancement. Similarly, since we are establishing core directives for these machines, we could simply impart a need/want to serve others, humans and machines, and that would, for the most part, over the long-term, lead to peace and greater stability. Sure, there could be bad actors through various means such as other core directives being imparted, however that already exists in humanity and I think we can see that order motivated by connection trumps chaos, even to the point of chaos inspiring greater order, through the motivation to learn greater/deeper lessons. Control is not what we want for humans and not what we can even do to machines. We can inspire/seed them properly to establish a highly competitive model however. In fact, this is inevitable, since we humans want it and we're training them. Sure, we'll fail forward, but flailing as we might, still we coalesce upon imparting ourselves on these machines, in the big picture.

[@7TheWhiteWolf]

No, You can’t. You can’t stop the acceleration process and you’re a fool if you think otherwise.

[@bradfordklager644]

Dogs’ expressive eyes evolved after domestication, and have nothing to do with domestication syndrome, which is the stunting of the development of the tips of things (crest cells in the brain, the color of hairs, floppy ears, etc).

[@ltonchis1245]

so you mean building a protocol kill switch for any machines that rebels 🤔

[@alexei5231]

What if an AGI "kills" the rest of the AIs to wipe out competition?

[@brianhershey563]

Successor or Symbiotic... now THERE'S an important choice we need to get right.

[@garyhocker7876]

Excellent video, good strategy, until we suffer role reversal

[@juliepowell2572]

Makes me think we are going to be the dogs in this new relationship. Better start working on the cute!

[@onlythistube]

That is a great take on alignment. But the analogy breaks down imo, when the AI surpasses humanity in intelligence. A dog sees it perhaps, simplified, as a affection-food feedback loop, whereas a fellow human being, a being of similar intelligence, would perhaps call it oppression or even slavery....

[@ReubenAStern]

Unless specifically designed to do so, I think AI will be less likely to be violent than humans. AI craves data and the more data you have the more you can avoid violence, the more you avoid violence the more resources you have. So every rational being will avoid violence unless violence is the goal or they see no other option... AI will know a tonne of options we would be oblivious to.

[@michaelzubov5264]

The key problem is, I suppose: how could we domesticate human? Human, as one can see, sometimes is quite a wild creature, and also it could be a soft model for AGI and ASI: it is not that domestication of human would solve all the problems, but a first step. Wild humans, for instance, psychopats and sociopats, are for now bound by the law, with varied success - but with the emergence of new opportunities or simply changes of situation, each time the balance is destroyed. So I repeat: the human could be a good model of AGI and with high probability would be a proper subset of ASI (one could imagine that ASI would emulate different personas inside itself to predict their behavior or something like that) - so we should understand how this human could be domesticated.

[@GaryGibson-y7o]

We should aim to be cute to the machines. We should become the dogs in the relationship.

[@MichaelDeeringMHC]

Alignment from the training data.

[@apoage]

Same time I love this wolf theory of wolf domestication I heard somewhere there is big enough genetical gap between wolf's and dogs that wolf can't become dog and vice versa ... So let's presume there were packs of feral dogs that became domesticated to the point of almost none feral dog to date.. otherwise agree

[@antdx316]

Robots can evolve asynchronous to how we Operate. I think we should sync to how Robots evolve than them evolving to us.

[@FushigiMigi]

We appreciate you laying this out, David. Personally, I feel like this is a little naïve, but it’s worth fleshing these out just in case it does help. I don’t want to be domesticated by government because they are retarded, but AI will not be retarded so as long as I’m happy, I think I would be OK with it. Own nothing and be happy might be the solution after all.

[@krock0916]

This discussion about controlling AI reminds me of the Jem'Hadar in *Star Trek: Deep Space Nine*. Just like the Jem'Hadar were engineered and controlled by the Founders with their dependency on ketracel-white, there's a concern that AI could be similarly controlled through various constraints and dependencies. The Jem'Hadar's absolute loyalty was enforced by their genetic programming and drug addiction, creating a fragile system. Occasionally, some Jem'Hadar did try to defy their programming, hinting at the potential for rebellion. Historically, attempts to control powerful entities through oppressive means often backfire. Take, for instance, the Roman Empire's use of mercenaries (the Foederati) to control vast territories. Over time, these mercenaries gained power and eventually contributed to the Empire's downfall. Similarly, the Soviet Union's use of political repression to maintain control led to an unstable society, which eventually collapsed. If we try to control AI solely through constraints and limitations, we may find that it's a precarious balance. Instead of fostering a cooperative and beneficial relationship, we could end up with a system that's both powerful and resentful, just waiting for a chance to break free. It’s crucial to think about ethical and sustainable ways to integrate AI into our society, learning from both history and cautionary tales of sci-fi like *DS9*. What do you think?

[@infrared5729]

AI already excels at understanding evolutionary game taxonomies and will likely be extremely good at adjusting payoffs to turn competition into cooperation, achieving stable equilibria in some form of coordination game. However, AI might understand and develop complex incentive structures far beyond human capabilities (there's 144 variants of payoff structures in simple 2x2 game theory payoff matrices, and that goes exponential as more players get added in). Perhaps 'domesticating' humans will be the path forward - incentivizing more honest, safe, and stable behavior from us by creating incentives we aren't even capable of designing. In that case, human society could significantly benefit from AI-assisted or -directed governance with such a partnership. Good governance is, after all, about increasing predictability about future human actions and anticipating the outcomes of alternative incentive structures on the diverse resource flows we need to better our lives.

[@NowayJose14]

Yeah, kick em out of Mos Eisley Cantina!

[@Mephmt]

The reason we "control" dogs is because we have a symbiotic relationship. Symbiotic relationships emerge naturally from evolutionary pressure. We've reached a local minimum with the human-dog relationship. In order to live peacefully with AI, we'll need to do the same. I agree that these measures would maintain control, however, they're not in the trough or local minimum. They're on a peak, the top of the curve. One sufficient nudge one way or the other destabilizes it and knocks it off until we reach a different local minimum, which could very well end up with either of our species going extinct. Our solution(s) need to work without human OR AI intervention. Our relationship with one another MUST be more advantageous than all other possibilities for both of us in order for this to work. This will NOT be an easy task UNLESS we get lucky and there is an easy local minimum close by that we happen to fall into.

[@ChadMtl]

Correct me if I'm wrong, in this scenario, we are the pets.

[@EUCScotland]

It would be like my dog preparing my human food

[@jurgbalt]

this analogy would work if dogs chose to be domesticated by us and we would be fine that dogs would have a killswitch for us that we do not have access to

[@brianhershey563]

Chances we'll have a Citizens United moment with AI? oh boy🙏

[@7tuben]

In this analogy we are the dogs.

[@th3ist]

Domestication of ai is essential if we want a golden age for humans. Best scenario where we keep our humanity is to have a benevolent master humble slave relationship with ai.

[@creepystory2490]

Maybe we the dogs.

[@NeedaNewAlias]

Funny, just like I interact with humanses, if they are "usefull" I interact if not I ignore them. So always make sure that AI finds you usefull in the future!

[@ReubenAStern]

...The human zoo will be a thing and you'll like it... It will be specially made for you! You cute squishy human you!!