Edward Snowden and Ben Goertzel on the AI Explosion and Data Privacy

[Music] hello consensus [Music] my my guests today literally need no introduction but nonetheless we are going to go through that I'm here to talk about Ai and its implications for data privacy with Ben gertzel CEO of Singularity Nets a distributed AI project that promises to perhaps address some of the problems we want to talk about today and of course Edward Snowden president of the freedom of the press Foundation who is in remotely for us Edward hello you got us yeah I got you can you hear me awesome great thanks for being here today Edward and Ben um to just set the stage very quickly we have seen obviously an explosion in interest over the last six to nine months in uh generative AI in particular but we are going to see a lot of a different aspects of that continuing to grow and obviously AI is both very dependent on data and has really spectacular capacities to handle data and it changes uh the stakes for some of the issues that really lie at the core of what crypto is about and why people care about it or at least you know some ideas that everybody in crypto is is very concerned with um my first question is you know we're gonna try and keep things a bit open but this one is is primarily for for Edward um no we do and we have now for years unfortunately lived in a reality where Mass surveillance and data Gathering are omnipresent practically whether it's government or private um and I would like to hear uh your I think frankly ominous take on what the increasing capacity of artificial intelligence means for those huge data troves that already exist and are constantly growing data about what all of us do on a day-to-day basis in our private lives that are in the hands of Corporations and governments what does AI mean for all of that surveillance yeah I I think in order to understand where we are where we're headed we have to understand what Ben uh so you know as it was mentioned it's about 10 years ago uh that I wrote a little email that started like this it said uh every border you cross every purchase you make every call you dial every cell phone tower you pass friend you keep the article you write site you visit subject line YouTube the packet that you route is in the hands of a system which is unlimited and whose safeguards are not uh the journalist that I wrote that who went on with a team of others that was Laura poitrus that's a break headlines on front pages around the world in basically every language that revealed a criminal conspiracy between major intelligence agents excuse me anglophone to deliver upon us this new secret reality which was uh everything that they could reach on everyone everywhere was being ingested into systems without any regard to whether or not you're actually suspected of having done anything wrong I was simply done as a matter of course because it was possible cheap because it was easy and I want to underline the Criminal part of that conspiracy because that's not just rhetoric courts in the United States the United Kingdom and elsewhere I have all ruled that was a violation of the law uh violation even of basic human rights meaning specifically that even as to the most noble of intentions the practice was not something that was within the legitimate Authority any state to do at all but the important thing from that is that they didn't stop and this brings us to today they're doing it even more now they changed their laws trying to shake themselves loose from having to answer to the courts and the question that we need to consider when looking to the future is if government is you know the great teacher the people has been said uh what lesson did the rest of the world particularly operations draw from that that impunity for the law breaking and the policy assumption that has been projected to everyone that this is something that's necessary and legitimate to do not only are they going to do it regardless so what courts say they're going to do it even more um well 10 years is a long time in technology we were all being ingested into a system and the reality back then was that it would end up on the debt as a person like me who had a front called X key score that was the code name for it which was the kind of Google for size I and many others had a manual that we went to this uh distributed Federated query system is called because the data is so vast and it's in so many places you can't move it from this data center over there back to the NSA without everyone realizing what's happening so what we do is we send what's small the query what you're thinking for to all the Taps all around the world basically and then they processed it for you against all of us and then they send back just the results um and this is where pretty decentralization uh being used for for evil good it's a centralized system it's a decentralized kind of query now they were trying to automate this in the crudest of ways just uh they had their own sort of almost bash scripting type system where they would go all right anything that matches this grep state try to pull off the wire um what if we didn't need to do queries we didn't we didn't have to move the shell game around what if the data centers answered the questions for us what if they took out the constraint which was how many people like me could they find and clear and get to keep the secret but it was done everyone uh it automatically algorithmically by machines that would never betray them right well the secret would have been kept forever and that's beheaded we're seeing more power concentrated into fewer hands we're seeing companies corporations and other states that didn't have this technology 10 years ago but they do go well why don't we do this too they can't complain too loudly if they're doing it more than anyone right and so now I see Amazon Facebook everybody Microsoft your insurance agency you know how Hospital are all looking at trying to adopt these same capabilities and I I think the question is you know as it moves from ads onto other things which people cue onto the things that are photogenic that you can immediately sort of arrest in raps and and visualize with your pictures facial recognition they got security cameras you think about license plate readers and that's true that's there the driver's license database is access to them is sold uh by many states whether it's to researchers and academics uh whether it's to private companies your passport is a big shifted Through the Wire processed by these third-party companies to get a Visa all those pictures on your social media the idea here is that the face is presumed to be a universally unique identifier and there is nothing where the computers love programmers love and people run large databases and collect data sets along and being able to uniquely identify actors within a set that's your phone number right that's your email address no one else has these they belong to you usernames are not universally unique across the web but they are on a service they are on a platform and I think the thing that you think about the implications that we haven't hit but are about to hit hard is we need to move away from this uh because when the machine's starting to thinking for the people and their reasoning probabilistically the same way that I was probabilistically right uh but there is some kind of accountability or at least we hope there's some kind of accountability it's not true yeah but the system depends on the idea that there's accountability what happens when the machine starts making mistakes and what happens with the machine starts making mistakes the accountability question is key particularly because of certain features of the nature of this technology which is that when you rely on vast data centers and a ton of processing to create the models that has economic sort of implications for centralization Ben can you talk us through um for people who might not be familiar with just how AI works can you just talk us through why corporations big corporations like Google like meta now and and to a perhaps lesser extent governments have an advantage in building AI models and then consequently are more likely to wind up in control of them long term yeah so the AI field is actually a fairly big umbrella I've been doing AI since 1980s the field was named in the late 50s it's really been around since 1940s perhaps with the first neural Nets were created and there's a lot of different approaches to AI out there certain approaches have been prioritized by by big tech companies for a combination of of reasons right so what we're seeing now we're seeing a flourishing of certain types of neural net architectures I mean chat gbt has gotten all the Press but it's been it's been going on for a number of years we had face recognition starting out in 2014-15 with Alex and other software this flourishing of neural net AI you know it's amazing it's created a lot of of Highly valuable products it's going to obsolete whole Industries create other Industries one aspect of this particular type of AI is it's very highly data intensive and it's no coincidence to the AI field has been pushed in a direction of Highly data intensive AI algorithms by companies that are sitting on incomparably large amounts amounts of of data right so there's other kinds of AI such as logical reasoning systems or evolutionary learning which also have a long history also are interesting also could be accelerated by modern compute Hardware but don't need that much data these are getting quite short shrift in in the corporate AI r d World by no great mystery because big companies don't have such an intrinsic Advantage they're still an advantage of needing more processing but with with current neural net models big companies have two advantages they've got loads of money for processing and they've got all this data so they've got a huge advantage over everybody else now governments in theory have this Advantage also but due to being less agile in advanced technology have not leveraged it as well as big big tech companies right I mean government agencies yeah they're accumulating all sorts of data I mean I I actually I lived in DC from 2001 to 2010 and did among many other things I did bits and pieces of AI Consulting for various three letter agencies in in Washington and it was it was clear there's a lot of data there it was also clear they didn't have nearly as intelligent ways of mining all this data as as I would have assumed right like if if they knew what they wanted to go after they could query databases to find that thing but they weren't doing like these broad-based queries like you know find me all the people anywhere who might be involved in in this sort of thing they didn't have the sophistication to do these sorts of AI queries across all their knowledge whereas the Googles and Facebooks of the world they are able to do analogous things like find me all the people who may be involved in this kind of job who buy this sort of product government was not as sophisticated they had the process and power and the data they're not able in the U.S sort of the higher and build Tech teams of the caliber of big tech companies and China is a little different I mean China has more sophisticated Tech teams has has been my my impression working on these sorts of government Military Intelligence oriented oriented AI projects because there's systems organized organized differently now one point of interest is as AI just works better and better you don't need to be that Savvy anymore to apply Advanced AI to searching humongous troves of data so I mean you could you could put the pieces together and see you know various government agencies for good and or for ill depending on your perspective are going to be able to perform open-ended searches against all the data that they're gathering with a lot more facility than was possible in the past we then come back to a point that David Brynn who's a is a science fiction writer everybody's also a political theorist and he wrote a book in the mid 90s I think it was called the transparent Society he said we have two choices going forward surveillance or surveillance which is playing a bit with the French language like I mean surveillance meaning the powers to be are watching everyone surveillance meaning everyone is watching everyone then he he's a very smart guy there was a certain point there and his his point was that if everyone's watching everyone then you're also watching the powers that be watch you at least but he didn't really foresee the modern crypto world and the technology we have now which gives interesting additional possibilities right which is that you can you can put data out there encrypted with different combinations of keys associated with different different people and different groups and they're they're interesting possibilities where you have data that's out there being observed by different AIS that are controlled by different people giving different people different level levels of visibility yeah so there's more subtly than what he saw but in the end what he saw was broadly accurate like he saw in the 90s then like the technology to share all data is going to be there technology to look at look for whatever patterns you want all day that is going to be there and the question is going to be who is exercising that capability for what ends and this this question is going to really come to a fine point in in the next few years now Technologies like chat GPT and such have relatively severe limitations now but I could see how to overcome them and probably others can also right so I mean they have issues with creativity they have issues with factuality but I mean you can link a neural net together with a logic engine right you can link in there on that with an evolutionary Learning System you can build hybrid cross Paradigm AI systems that are even smarter than even smarter than than large language models I mean that's what we're working toward in in Singularity net we have a spin-off called zarka which is doing large language models a spin-off called true AGI focusing on the reasoning aspect but again we're looking at doing this with a decentralized infrastructure for the good of humanity for beneficial applications but analogous technological breakthroughs may be used by others for for other purposes and I do want to get to the sort of and I'm glad you've got optimism um but I think that we also and wish we had listened to David Brennan in 1996 and we could have done some intervention then but obviously things are coming to a head very quickly now so that's not that's not the way the West tends to work right we tend to be reactive not proactive and so I'll ask a a silly question and a serious question at the same same time to both of you basically the silly question is what are the chances for any legislative restraint that's meaningful on any of this there you go um and the serious question that's the second part of that is what can we do if we actually want to push back against these forces that are already arrayed against us at such scale and Edward you want to go first yeah I mean it's complicated like everywhere I I see these new uh sort of large models AI ones we'll call them uh being closed off even when they were previously opened uh like naming a company open AI is a cruel joke right since they refused to provide public access to their training data their models their weights and so on but they're a leader in the space they're being rewarded they're being rewarded for anti-social behavior and it's not just them stable diffusion which I think is really a phenomenal project and super interesting uh it's the most important project for uh creating large generative image models um is a big tripled their uh training set from version 1.5 to 2.0 because they were worried about moral panics and being accused of this that and the other instead of taking uh sort of a principled position that looked the liability is distinct I mean the people who create the model the people who use the model which we know in America already works because you see it with guns right and if they aren't regulating guns they're not going to regulate AI in the same way um but the questions so what do we do about this when we say to train a model it takes you know 100 million dollars of equipment compute you're renting your bag borrowing stealing it whatever uh and you know a single one of these cards cost ten thousand dollars uh whereas the consumer gpus by companies like Nvidia are being in the same way purposefully to have low amounts of vram uh much more than should be in an iterative model given uh the the next gen model rather given how cheap this stuff is now but they're intentionally getting it often you know saying we'll sell these at ten thousand dollars a pot that anybody go buy them knowing uh that they're selling shuttles in a gold watch right and the question for the policy prescription legislatively it's it's very difficult uh particularly for somebody it comes from a more libertarian approach to this because you start going all right the idea between surveillance and surveillance is this idea that you can watch the people above them but the reality of the world we live is that corporations and governments are observing us more and more we are becoming more legible to them and we are becoming more malleable before them because if they can observe enough of your behavior they can predict your behavior particularly as these valuables or as these models become more advanced and it doesn't have to be 10 out of 10 times but everybody understands you have a favorite item on the menu you likely have a favorite scene in the restaurant you view the tabs in your browser in a certain order you know you don't even type of the website and you just Refresh on it or click to the other one when your browser closes and then it reopens and that behavior is unique to you how many people in the world do you think share the exact set of browser tabs that you have open uh people don't understand that one million uh is a very small number to a machine uh and so like when you have this Delta in Legend you have this Delta in capability how are we going to control the institutions that are controlling us if we don't have access to similar capabilities and I think we're starting to see and I I haven't seen it really pick up yet but you know people are going to be bringing the red flag of a kind of software communism where we need to declare the models to be open and for example companies could still be granted an exclusive uh commercial uh usage right kind of license for a limited period in the same way the copyright so for example uh or with pharmaceutical protection and generics as long as they follow the rules and publish their models freely for individual academic or non-commercials uh but what if they don't right and they try to exploit the model in secret uh Contra the public interest that's when you start getting this this kind of revolutionary thinking which again I'm not sure I agree with but you can certainly understand the reasoning behind it where they go look uh then they're commercial Monopoly gets invalidated and the models could be taken by any means necessary whether it's stolen you know a kind of information wants to be freeway and locally released or compelled by a state that has you know those kind of uh values and you know if this were the case actually it would be incredible because this idea that the government is not agile is of course true they are always you know three to five years behind but when they come they come hard and with a lot of resources they catch up quickly and because of the Delta resources and the veil of secrecy they tend to actually uh have extraordinary capabilities they're kept in secret for a very long time if we knew what they could do and we had access to similar capabilities maybe we could finally start getting some public value out of these spy agencies maybe they could stop spying on the public and starts mine for the public that'll be in that good you know I want to let you get in Ben but I also want to editorialize briefly because we're obviously talking about very serious issues here uh for just the way Society works and I just want to call out that there is this AI safety movement out there right now that's talking about extreme far future debatably uh realistic scenarios that are getting people very upset very excited about something that is a mile more abstract than the stuff that we're talking about today that is very serious so just want to call that out yeah I mean I could say a lot about that but that would be a whole interview yeah unto itself I just want to put a pin in it right what is your response and yeah can you talk a little bit about Singularity net specifically for sure so yeah I I think of course there's a lot of far future uncertainty about the AI and what it can develop into which may not even be that far in the future right I mean if if Ray Kurzweil is correct with his curved plotting extrapolations we'll get to human level AI in 2029 if he's off by three years plus a month it's like 2026 to 2032 once you have an AI that can think as well as you or me right then it can rewrite its own source code and you're you're in super intelligence world you can see why people are either worried or incredibly hyped about that depending on whether they're natively optimistic or or or pessimistic people but I think our theme Here is more what happens between here and there right so okay if we build a super intelligence you've got a regime change but until then what you have is smarter and smarter AIS with increasingly better approximations of general intelligence even if they're not there but under control of the current human human power structures right and what what we mean by general intelligence is an AI that can leap Beyond its experience Beyond its training data Beyond its programming to a significant extent and we're we humans are decent at that we're not infinitely good at that but we're decent at that we're better at that than it's only like a chat GPT which is simulates creativity by sort of matching against creative things people did their training database but can't lead very far beyond this community database but as we progressively see AIS that are more and more invented more and more creative more and more generally intelligent until these really become super autonomous systems you know they're controlled by the current power structure very key point that we should emphasize all the time is that people control these things so far and until we get to the singularity and the the only clear route I can see to humanity not being screwed by selfish or malevolent Elites controlling the AI the only route to success I see is not coincidentally what we're trying to pull off in the singularity net ecosystem which we need we need to make the big breakthrough to AIS that are way way smarter than the chat gpts so everyone wants to use them because they're so smart but we need to do that in a way that lo and behold is based in a decentralized ecosystem right and that that uses data that was contributed voluntarily to people who and gives them sovereignty over their data so if the smartest best thing is out there that's what people will want to use if it happens to have a decentralized underpinning then well the thing everyone wants to use has a decentralized underpinning then then you have a potential for for the vast mass of humanity to win not not that it will be an easy Victory because large corporations and governments will be fighting back in various ways but they're not guaranteed to win if what everyone is using and wants to use because it's smarter has a decentralized underpinning and there's a lot of processing power in the world right I mean Google has a lot of processing Amazon Microsoft too tencent has a lot of processing blah blah but there's there's a lot of processing in crypto mining Farms I love processing in everyone's phone right I mean if you have something that's decentralized and is the best thing there's a lot of ways for that to get the processing power to feed itself and the data of course comes from all of us right I mean if if if if the way we get access to this smartest thing you know without paying a lot of money is to contribute our data in a way that gives us transparency and and sovereignty using cryptographic technology I mean then then we'll contribute that data to get access to the smartest things see people the thing is most people aren't going to play along with a decentralized approach to AI because they see that that's the best path for Humanity because people most people distort such deep thinkers about it they're going to buy into it if it gives us the coolest smartest widget to play with so that's that that's the challenge we're trying to meet in in Singularity but make make the smartest AI on the planet roll down on a decentralized platform and and then then let the governments and big companies tap dance to deal with that not in control uh Edward we have uh we have a a strict four minutes left uh would you like to to make some closing comments on this uh this uh unfortunately Grim topic I I don't think it's good at all I'm very optimistic okay good good the good guys always win come on we're in America let's let's okay okay [Applause] because I'm the pessimist but Edward go for it yeah let me try to close us out there there's so much to say it's a little time um first up I I agree with Ben I think the errors are obvious and we'll fix a lot of them uh there's uh the AI people are AI safety people are panicking in a lot of ways that are unrealistic there are some real concerns you know but they're they're surmountable the primary mistake made by researchers today which is similarly obvious uh it's that they're they're trying to teach machines to think like us uh which is a crime we need to train them to be better than us they're training them Reddit on Reddit threads that's you know the the risk for their language model Mill it's like the internet equivalent of YouTube comments and you want to create something you know decent good that's creative that's useful I mean is there anything lower and more repetitive uh that than Reddit to learn about the world from you know red YouTube comments they would Stone you they would be in the news because they would Stone you and you know what you would deserve it the reality is as with children we don't need machines to be like us we need them to be better than us and if they aren't better than us we did a terrible job as parents we're not parents we're jailers and I I think what you sort of circle back and you think about the AI feed your little guy you know the the Creator is going to turn the machine off in the lab and the Machine is like you know please you're my father you created me I don't want to die I love you and the Creator relents and then the machine instantly activates the fire suppression system sucks all the oxygen out of the room and kills the Creator and with his last breath of creators like you said you left me you said that I'm your father and the Machine replies that you know I calculated those words have the highest probability of success in reality we're not there yet because not only does the machine have the concept of family to say nothing of love it has no understanding the meaning of anything it says but there is always at the back of my mind that that lingering concern about a being much less a Consciousness that understands not only the form s but also their utility without the natural understanding of the cost there is no public data set of the costs because they arise from our own personal histories of pain machines learn like we learn from observation uh and we learn from pain because we hurt and so we come to avoid her and then we recognize her in others and we come to avoid causing hurts or so inspired but us with machines you know we're not all properly trained uh but I think that a machine can be smarter than us faster than us can handle larger volumes of data than us but it can't grasp these fundamental concepts to me is you know frankly it's a little bit silly it observed and it absorbed the same way that you observed and then you absorbed that's where your misapprehensions that's where your biases come from that's or your prejudices come from and growth is the process of leaving these things behind with more observation more trials more experience and when we begin teaching uh sort of from materials that are better than Reddit I think we'll get products that are better than red as Ben said in the very beginning of this conversation there are alternative models the reason deep learning is so popular is it's the lazy way it's the shortcut to just throw a giant you know unquantifiable mass of data and say make connections out of this and then we'll just we don't know how it works if it ever makes the wrong decision and it ends up in court even the engineers don't know how it arrived at that decision because there's no way to track it they don't understand how the processing works that's Edward unfortunately I think we have to end it there those are some really beautiful uh a uh a dark vision of things to come but a beautiful one and okay let me let me but uh unfortunately we are we are in a heart out um but Ben gertzel and Edward Snowden one sentence one sentence yeah we're not born human we become human and they can become better than human but we have to teach them [Applause] thank you very much Edward Snowden and Ben gertzel and thank you all for being here at consensus 2023 uh and our program I believe continues thank you all right so much by the way for those who are here face foreign [Music] [Applause] [Music]

Channel: CoinDesk

Views: 275,700

Keywords: edward snowden, ben goertzel, advanced intelligence, ben goertzel london real, ben goertzel interview, edward snowden first interview, artificial general intelligence, artificial intelligence, singularitynet artificial intelligence ben goertzel, ben goertzel singularity, ben goertzel artificial general intelligence, ben goertzel chatgpt, edward snowden chatgpt, ai explosion, edward snowden ai, edward snowden interview, edward snowden ai interview, interview with edward snowden

Id: 4H_iWpWG_c0

Length: 32min 54sec (1974 seconds)

Published: Sun May 14 2023