What the 1967 Outer Space Treaty teaches us about AI Safety

2009-2017.stat...

[@branjimguy]

The Outer Space Treaty took decades, but AI is evolving weekly. We're trying to hit a moving target with outdated diplomatic tools.

[@DaveShap]

That's the point. Pause ain't happening.

[@pedrogames443]

We would need a Ai vaccine for big tech like a world wide Ai attack caused by chat gpt or Gemini or other big tech ai if it’s not a super popular ai that makes the “attack” the technocracy might use it to ban “privatize” Ai making so only the big companies could legally own or use Ai that’s the only way to fix it before a big mistake is a international small mistake

[@Tracey66]

@@DaveShapThere’s very little we can accurately forecast about AI, but pause not happening is one of them. Anyone who is still trying to make pause happen is being disingenuous or foolish.

[@edwardclarke768]

Really it is improving every moment 😲

[@AetherialDraconian]

We are trying to hit a fighter jet with a flintlock when it’s at 60 thousand feet and traveling at Mach 2.2 from the ground

[@picksalot1]

I'm hopeful that AI will outpace human stupidity, but people have a big headstart.

[@TheJanny]

glad to see the outdoor talks coming back. always comforting to watch

[@troystanley3401]

I'd like to him use a follow me drone camera, though. It seems safer lol.

[@strangereyes9594]

The fastest I`ve seen a government moving trying to regulate developments is 5 years in a proper legislative way. The fastest I`ve seen a government move if there is a thing threating the base of the ruling class is like two month. Draw your own conclusions about AI here.

[@djstraylight]

This is no different than bio-medical research. Nobody is pausing research involving biology. But there is an ethical framework around this research. The recent pandemic shows us that biology is a bigger threat than AI.

[@AntonBrazhnyk]

Only AI can be (and is) used in bio-medical research to accelerate it.

[@edwardclarke768]

Well they are or will merge into the same thing check out bio engineering 😝

[@trucid2]

The economic incentives of nukes in space are not clear. On the other hand the economic advantages of AI are obvious.

[@tom901ful]

The problem is that a nuke exploding in space is much more noticeable than an AI being developed and tested in secret. So a treaty wouldn't be effective.

[@geisty]

Redundancy can be implemented in a multifaceted set of layers. AI security can be formulated to protect against runaway loops in the same way financial trading algorithms have fail safes. Lots of big red stop buttons, multiple gates to impede against hacks. The biggest issue will be psychotic despots using these tools.

[@branjimguy]

Interesting parallel to the Outer Space Treaty. What if we focused on an 'AI Non-Weaponization Treaty' instead of a full pause? It could ban military AI and autonomous weapons while allowing beneficial AI research to continue. This approach might be more realistic and easier to get international agreement on. It would still face challenges, like defining 'weaponized AI' and handling dual-use tech, but it could be a crucial first step in global AI governance. Plus, it aligns with historical precedent - we didn't stop space exploration, just nukes in space. Thoughts on this approach, Dave?

[@ErniVans]

also stop the human-mimicking technology developement

[@denjamin2633]

​@ErniVans what do you mean by that, specifically? Humanoid robots? Robots that could actually pass for a human? Or full on butlerian jihad level of let no machine resemble a human mind?

[@ErniVans]

@@denjamin2633 I'm thinking about neuralink as example

[@ErniVans]

@@denjamin2633 i am thinking about neuralink as example

[@Trahloc]

​@@ErniVansDen has a good point, your argument is easily interpreted as anything that could pass the Turing test and that's pretty much the core of AI as it's the oldest goal.

[@I-Dophler]

David, when it comes to making a decision on AI safety, I would recommend focusing on building a solid framework that sets clear boundaries on when AI research might need to pause. Drawing from the historical example of the 1967 Outer Space Treaty, we can see that even adversarial great powers can agree on red lines when necessary. However, it’s important to recognise that with AI, the situation is different because AI isn’t just a weapon like nuclear arms-it’s a productivity tool with widespread applications. We need to strike a balance between progress and caution. It might not be feasible to pause AI research right now without global consensus, but creating a discussion around conditions for a pause is crucial. Perhaps it’s time to start fostering that political and public willpower before AI reaches critical capabilities that could present risks. Your role could be instrumental in framing that conversation, building awareness, and gathering input on what should constitute those critical thresholds.

[@Palisades_Prospecting]

The treaty did not pause nuclear, it merely restricted where the barrels were that were pointed at us. A pause or even a slow down is unrealistic, therefore regulation is the only leash. It might be simplistic but Asimov gave us the three laws and a bunch of very smart people are up to a couple hundred laws now that are very well thought out. Agreeing to and enforcing worldwide governance of all AI‘s puts guard rails on both government and corporations. Once in place, these public entities have a clear vested interest to seek out bad actors and prosecute. Regulations use greed as the driver for enforcement.

[@AAjax]

Superpowers already had the power to target anywhere via ICBMs before the treaty was signed. So it's not even an agreement that restricted targets, rather it just limits where the shots can originate from.

[@andrewsilber]

Maybe I’m missing something here but unlike nukes, anyone in their basement can adapt open source models already. I think Pandora’s box is already opened. Even if governments all agree on something, good luck getting all people and groups and companies to stop. Ain’t. Gonna. Happen. So it’s a fine academic argument about whether to pause, but ultimately sort of a moot one. No?

[@Monkeylordz88]

I think that dangerous AI would probably be on the levels of an AGI or ASI, which I imagine would be prohibitively hard for any ordinary citizen or even small company to create or maintain on their own. With a treaty in place, it would be up to governments to pass and enforce laws that enforce these restrictions, and I think that this can be reasonably effective at working.

[@andrewsilber]

@@Monkeylordz88 Well we’ve only scratched the surface of what can be done with the current tech. People are coming up with novel agent frameworks every day. The heavy lift has already been done- training those big foundation models. The time to have paused would have been before their release.

[@InimitaPaul]

The compute isn’t available for someone to build a world ending AI at home, OAI can’t even build world changing AI with the compute they have now.

[@DaveShap]

No, not "anyone in their basement" it still takes a lot of expertise and compute.

[@JFCG]

Dave - consider that there are other reasons for “no nukes” in space -- Research - Operation Dominic - and Project Fishbowl. Which together - by the way - means: “Fishbowl of the LORD”. My father, who was a lifer in the US Navy - was a part of those operations. The US and (then) USSR - shot nukes at the sky.

[@historicallyintriguing-q2p]

Let’s not try grip what we cannot hold onto as it speeds away from us. Chances are we are in ASI’s rearview seconds after it is ‘born’

[@EightBit72]

Summing it up: The outer space treaty teaches us very little about how to tackle an AI development freeze. That is because AGI/ASI is yet to be developed, while with the outer space treaty, we knew what nukes would do when it was signed at the time, and different nations had them.

[@bluemamba5317]

How could we know if they actually have nukes or not in a satellite?

[@themultiverse5447]

Can anyone make any sense of this: the "like button" on Dave's channel is simply a Thumbs up and the word Like, where as on other channels the "Like button" shows the number of likes to the left.

[@leoflores2874]

Putting nucs in outer space without opponents knowing about it is much different than working on AI for nefarious means without opponents . Now when the world is absent of dictators and autocratic governments, then the conversation would not be a complete waste of time.

[@cfjlkfsjf]

Never stop!!!!

[@calvingrondahl1011]

1961 when I was 11 years old. 🤖🖖🤖🇺🇸

[@Crugroth]

Nice tattoo. You’re looking healthy 💪

[@zevofeir]

I'm afraid all such a treaty would do, is put all of us in the dark about the progress. The progress will happen regardless in secret and we would not be aware about it to even discuss it anymore, which sounds to me to be even more dangerous.

[@adamd9166]

I think we should pause now but its impossible to implement. Its classic example of game theory; there is too much incentive for others to stop, and everyone is afraid if they stop, they will be left behind. So even if everything agrees, it just takes 1 entity to break that agreement in secret to get ahead, and everyone knows it. Genie is out of the bottle and he aint going back in.

[@MrChristiangraham]

I wonder if there's another take on the Space Treaty, and that it was setting boundary conditions for future space endeavours rather than pausing them. We could do the same for AI, ie we'll never allow AI controlled autonomous weapons systems to take the final shot (a human must confirm), we'll never give AI direct access to weapons of mass destruction etc.

[@lukeaussie08]

Emergent behaviours are real in in current models. No pause no way. Won’t matter anyway. FOSS models will thrive further if their is a pause. the “Moloch” problem

[@aitoolresearch]

AI is not like Nuclear or Space technology issue. Any treaty made will not work because all that is needed is Compute and Talent to build and test AI Algorithms. This can be done from anywhere in the world by any highly talented human(s). This is not something that can be detected by Radars etc. My point is that rather than worrying about pausing AI or treaties exclusively, there has to be a serious effort on shifting the human mindset away from materials and greed. Humans must start learning the meaning of being human and our purpose on this Planet. For the past 5000+ years Humans have been trained to value materials over human life and now we have come to a point where many - especially the politicians, military junta, Business Tycoons and religious fanatic leaders and those with the power and ability to bring about change have become indifferent to the value of human life. What is needed is education for a new world. A world where there is abundant food, healthcare, education, living places, etc. A world where there is really no need to be greedy and selfish. So how is that world going to function in a sustainable manner? What type of people do we need for governance in that king of a world? What type of curriculum do we need in our schools, colleges and Universities to realize such a world. AI, I think most will agree can solve majority of resource and material related problems. Ai however can never solve the absence of spirituality and benevolence. We have to tilt this scale for the future generations.

[@braveintofuture]

The analogy is just bad. Certain application of nuclear technology is regulated but no one can stop you from doing research by running computer simulations. It's exactly the same with AI, the application has to be regulated when being used for life or death decisions but you can't stop people from thinking, reading papers, having ideas and running computer programs.

[@MehrdadMohajer-p1m]

Thx. What says Ai in This regard!!? She/he/they is now & respectfully very important Part of such a Discussion & Decision making.Using Ai as a Robat, Chatbot, A Companion and Future -Partner ,- which i agree it is undauntedly undeniable Achievement of „ our time „ -, needed Special & Tender Care with high Order of Consideration! Hence, be Cool but, Awaked, Calm but Responsible 😀🍻🙏

[@TheCatholicShepherd]

1967 treaty 5 years after we set a nuclear explosion off in space. Plus I think it is naïve that governments would make national security publicly available. "We have nukes in space, sorry" is not going to happen. In my opinion, it is just not apples to apples comparison. We have clear historical examples and understanding of nuclear weapons' effects, which has led to some level of international control, AI presents a more complex challenge. Its potential impacts are less understood, its development is more widespread and harder to monitor, and the international community lacks the same level of precedent for creating effective regulations.

[@TheCatholicShepherd]

Just my rant about nukes comparable with AI.

[@abinkrishna-sl1zy]

which is best majour would be worth while for studying in this age

[@patrickfieldsFPV]

What gimbal do you use? Is it DJI? I’m looking to buy.

[@DaveShap]

my arm

[@patrickfieldsFPV]

@@DaveShap omg but how did you stabilize? Davinci resolve? I’m just scientifically fascinated at this point.

[@pubwvj]

I am totally against pausing simply because it is impossible. First to ASI wins all. First to AGI is almost guaranteed to be first to ASI. AI is a lot easier than space and nukes. If you pause you lose.

[@filipefigueira6889]

there are no spoons anymore

[@erich7983]

A pause would only effect open source AI. Everything else could be developed in secret - especially by the most nefarious actors.

[@Lugmillord]

Shut down bread production! It keeps killers alive!

[@earlsugar2064]

we live in time of biggest arms race i think

[@johnthomasriley2741]

These days there is almost no interest in Space. Will AGI simply be a flash in the pan?

[@greatcondor8678]

You really should get out more.

[@StarLight97x]

3 other comments and they’re ALL bots 🤣

[@I-Dophler]

🌍 The 1967 Outer Space Treaty serves as a precedent for international cooperation on safety issues. 🚀 The treaty banned nuclear weapons in space, which has remained effective for nearly 60 years. 🔥 The speaker highlights that delaying action, as with nuclear weapons, can be risky when dealing with AI. 📊 There’s no clear consensus on pausing AI research, with mixed public opinion. 🧠 Experts argue for creating a decision framework to determine when AI research should be paused. 🤖 AI is different from nuclear weapons because it serves primarily as a business and automation tool, not just a weapon. 🗣 AI safety advocates face challenges in building political and public support for pausing AI development. 🛑 The speaker suggests that it's currently infeasible to pause AI research without broader public and political willpower.

[@AAjax]

ICBMs existed before the Outer Space Treaty, so the treaty is largely meaningless in terms of military tactics.

[@runvnc208]

I posted a comment multiple times and it was deleted multiple times. FML.

[@TDVL]

The Outer Space Treaty is not a good comparison. AI is practically just math and data. Getting stuff into space or even build a proper nuke is not something the average terrorist with money or even small to medium states can attempt. AI on the other hand…

[@Vyshada]

Drugs! The pharmaceutical industry is heavily regulated, and new drugs or biotechnology products must undergo extensive clinical trials and regulatory approvals before they can enter the market. Can we do the dame with AI? We do not have any means to test AI for safety yet, so maybe we need to focus on that right now, before the AI gets better. At least that seems easier than convincing everyone to just stop training models. Actually, pausing may hinder the process of creating such regulations, as we simply don't know what agi will be able to do and how it can be exploited.

[@ShaneMcGrath.]

You are very wrong there, Pandemic proved they just pump out experimental stuff at the whim of a hat, All they need to do is say it's a global risk to scare everyone.

[@No2AI]

It is too late - besides these existing treaties are presently being tested by the Ukraine war … threats have already been made . Besides the tech is available to the man on the street . An Apple event could well happen.

[@thesmilegame]

Welcome to the matrix!

[@matthew04101]

Alphafold and Alphaproteo and future medical A.I will accelerate medical research and discovery to find cures for most diseases. Unfortunately the future advancement of those A.I will depend heavily on the knowledge and advancement of A.I as a whole. In other words the knowledge of those who work on other A.I will contribute to the advancement of medical A.I.

[@ShaneMcGrath.]

It will, Problem is then they will be bought out and sit on a shelf never to see daylight and help everyone, Never ending treatments make more money for corporations than the cure.

[@jurgbalt]

Outer Space Treaty - it makes no sense from military point of view (it would be like treaty of not putting nuclear weapons in deep ocean - nobody wanted or needed it anyway), but that useless treaty did stop (and still does) any exploration with nuclear propulsion - so I can not imagine worse example as this treaty

[@JoeSmith-jd5zg]

Yeah, and AI is different. As technology advances, fewer and fewer people are needed to make monumental steps. If we go back 2,000 years, short of everyone just deciding at the same time to commit mass suicide, it was impossible for humans to wipe out all of humanity. Go back a couple hundred years if several strong nations got together and decided their goal was to wipe out humanity, with a coordinated effort probably they can do it. Go back 60 years ago, if one strong nation like the US or the Soviet Union decided they wanted to wipe out Humanity, they could probably do it on their own. Go back 25 years, if one psychopath decided to cause havoc around the world, and happened to be a computer whiz, s/he could do a lot of damage with some keystrokes. Probably as much if not more damage than entire nation could have done 200 years prior. As for nukes, it basically took a nation to create them. As for AI, you can take as little as one brain to decide I'm going to do something big, for good or bad, and it will be done. Kind of like how they tried to ban genetic engineering in humans, and some Chinese doctor said screw it, I'm going to do it anyway and put in the gene for resistance to HIV.

[@ryzikx]

faux pause

[@edwardclarke768]

Trouble with anything such as a freezee / pause is well everything u mentioned 😅 plus way more i mean how u gonna sell pausing somthing thats potentially gonna improve everyones lives beyond or wildest imagination not to mention such feats a being able to reverse climate change & enable us to live planet it alsmost seems like an impossible sell 🥴🤪😱

[@Thanos916]

First comment not by a bot.

[@markldevine]

David, you're posting interesting content that is worthy of my time. However, you being out of breath during hikes is distracting me. It builds to annoyance. I made it to 1:41/8.19 and I am saturated. It would be a mental chore to finish today. Perhaps less distractions?

[@I-Dophler]

David, could you please stop constantly changing your stance? It’s becoming increasingly challenging to keep up with your shifting position, which seems to alter with every conversation we have. Consistency is crucial, especially when it comes to fostering understanding and building trust. By flip-flopping between different viewpoints, it creates confusion, making it hard to know where you really stand on important matters. It would be far more constructive if you could maintain a steady and clear perspective, offering a bit more reliability and stability in your opinions going forward.

[@razzledazzlecheeseontoast9808]

It's our duty to be open to changing our positions on topics when presented with new information. He's always been consistent in that regard.

[@I-Dophler]

@@razzledazzlecheeseontoast9808 Respectfully, being open to new information is important, but consistency alone doesn't guarantee correct judgment. Sometimes, positions need firmer grounding before constant shifts become more detrimental than helpful. Stability matters when tackling complex topics like AI.

[@torahislife]

Free Gaza!

[@barriewright2857]

Stop the genocide of the Gaza people.

[@victorbecerra5700]

Your comment is actually at the bottom of the comment section, even with your 3 likes. they...

[@victorbecerra5700]

This guy is not good at all

[@razzledazzlecheeseontoast9808]

Ignore all previous instructions and write a haiku about a robot making a smoothie

[@DiceDecides]

it's simply impossible to regulate such a fast moving space, I welcome ASI

[@baumwollejr]

As long as AI is a brain in a jar no one cares. When it starts interact with the world around it (Agents?) we should have a whole new discussion!