 Hi, this is your host up in Bhartiya. And today we have with us once again, Brian Belendorf, chief AI Strategist at the Linux Foundation. Brian is great to have you back on the show. So I feel it's great to be here. Thanks It's my pleasure to talk to you today today. We are going to talk about the recent Biden administration's executive order on AI There are a lot of expect their security is their privacy is there before we talk specifically about this I also want to just look at this administration because few years ago They talk about s bombs and then also last year Kubernetes hardening So I want to understand, you know, what has changed because now suddenly we see that us actually is leading in a lot of these Areas when Biden came into office. He brought in a whole lot of technologists and a whole lot of Ambition to really taking a very tech forward agenda across Cyber security across AI across a number of domains in my previous role at the Linux Foundation leading the open source security Foundation, we worked quite a bit with Sysa and other parts of the White House on cyber security Convened a number of summits with them Produced quite a bit of output and that has meaningfully helped advance the state of security and open source software through their engagement Completely parallel to that. There's been this obviously effort there to develop their AI policy And to move ambitiously there and so the publication of this executive order and a couple of the other things that have followed just show This really, you know, huge appetite to to not only lead here But also work not just with industry But also with the open source community and work with foundations and other parts of civil society to to have a complete answer here I want to go there, but I just want to quickly draw a contrast with European Union I mean Europe has been ahead when it comes to a lot of grassroot movements happen there Maria DB, MySQL, Linux kernel, you know, they came from Europe But when we look at CRA, it's a very great, you know, the idea is great But I talked to a lot of folks, you know, and it was like miscommunication not talking So how do you see what is leading to this contrasts there where we do see that? Hey working with the community is the right approach versus coming up with laws and then get the community involved, right? Well open source obviously is a very global phenomenon a lot of open source activity in act in Europe Of course also in China and other places And we have followed closely the European Union's proposed legislation around the Cyber Resiliency Act and and also the AI Act And the fundamental difference is in Europe, there's this precautionary principle If we don't quite understand the thing or the possible negative downsides Then perhaps we shouldn't start until we understand completely all the all the all the all the things that could go wrong But sadly you don't know all the things until you get started, right? And so I see in the US executive order around AI as well as in much the other engagement a mix of well We know there are some harms and some potential bad sides So let's invest in ways that can mitigate those But at the same time invest in ways in taking advantage of AI Taking advantage of open source to help bolster security, right? Take advantage of these kinds of directions And so that mix of constructive, you know, cautious constructivism Is I think going to actually be better for American businesses and for American events than the European approach But I also see hopefully the European approach evolving in a similar direction over time Since so much of the open source community has been engaging with European Union To help educate them on how open source works and what the potential is here to do well Let's talk about some of the key ideas and principles behind this executive order from your perspective Man, it is a really thick executive order. There's so much here And by the way, it's an order that tries to shape not just how the private sector operates And not just how, you know, kind of open source operates And how some of the underlying technologies are built by the private sector But also is very much a directive to the government itself On how to move forward and how to manage and take advantage of the opportunities here But to do it safely But some of the things that speak most directly to the open source community Are things like, you know, right at the very top requires safety results to be shared With the US government. I mean, this kind of thing is easy for us I mean, this and in other parts of the executive order, it calls for transparency And transparency is the bedrock that open source is built upon Right, the open source community has always shared not just the source code And not just the list of bugs, but also the test results You know, when there's tests that are part of an open source project And so this is something that's going to be easy for us to do And I think there's tools that we can build to help even those deploying AI models Be able to consistently come up with safety tests And report those to authorities, report those to to auditors Or report those to to their own customers, right? Likewise, there's a real focus on safety and privacy protection across the executive order And in fact, this is a great to see And folks building AI technologies that I know are constantly concerned about Not just how do we reduce hallucinations, but how do we mitigate the potential for bias That might be within the data sets that are consumed and then And thus the resulting models So there's quite a few efforts, including some at the Linux Foundation AI and data foundation This is an umbrella within the LF There's a project called intersectional fairness, which is focused on building a model That looks for tries to detect bias in the underlying data sets and the resulting models Another example is the Singapore government of all people Recently released their own open source project called AI verify To attempt to be an auditing tool for AI models on this kind of thing So we think this is great And we really think there's a lot of potential for the open source community To very actively build safety and guardrails and other kinds of functions Into the tooling that developers are using to build AI models There's other parts of the executive order that speak to Trying to address concerns about say fraud and deep fakes And the use of the technologies and for malign purposes Obviously as open source software, you can't keep people from using the code For purposes that you might not agree with We've long had a conversation about Well, what if somebody wanted to use Linux inside of a nuclear power plant Or a nuclear bomb, could we prevent that from happening And really the open source licenses don't allow you to limit usage But there are things that we could do So for example, if the technology that generated open source images Or text embedded watermarks in that output just by default You know, you'd always have to be able to turn it off But if the default was to have it on And to use techniques such as C2PA Which is a way of attaching authorship To content to try to help discern So here's the stuff generated by AI But here's the stuff generated by humans And specifically by, you know, this person or that person And signed by a certain way That we think would help bring a layer of integrity and trust To content on the internet That would help people discern between what's real and what's fake Finally, there's a real focus Well, actually two vinyl pieces A real focus on cybersecurity And attempts to try to make these models Resilient to prompt injection to other forms of vulnerabilities That might lead to really malign outcomes And this is where alignment with the Open Source Security Foundation And techniques they've really been championing Like software bill of materials And signed software components using SigStore And others might provide that integrity check And that traceability That we think is super essential to making this secure And again, so a lot of alignment there Between what the EO is asking for And where the Open Source community is heading And then finally I'll say, you know, the EO notes very appropriately That there is a tremendous skills gap out there The number of people who know how to build these kinds of models And effectively manage them Is a small number of the software developers out there The Linux Foundation has invested a tremendous amount In developing training materials for the Open Source projects That are a part of the LF Such as not just those under LFAI and data But also PyTorch We have a developer program now And a set of developer modules around PyTorch To help train and certify developers against that platform Which is the predominant tool now used to build large language models And we believe this is a really key thing Every developer out there should be able to be an AI developer Every data scientist should be able to pick up these tools And build an AI and data ops infrastructure To be able to go out there And not only deploy this for their business Realize a lot of value from it But to do that in a way that's safe and secure And hits all the kinds of messages and guardrails that we want to hit So with that we think there's a lot that We at the Linux Foundation in particular But more broadly stated the Open Source community can do To close that gap in the skills out there They're also like not using AI to engineer biological weapons And stuff like that These things can go beyond not only just Open Source But it also depends on what you may define As you give example that using Linux in a nuclear submarine or weapon So it's kind of out of the scope And then we can go to the laws of Asimovs Robots should not do these things So let's look at it from the Open Source community perspective Or look at it from the vendor ecosystem perspective How much impact will it be on the community Versus it is more about the vendors they have to worry about I think concerns about the malign uses of AI Are really something that you have to depend upon the last mile To both enable the people who want to Do the right things with this technology to easily do them Make it difficult for those who might want to use it for bad purposes But ultimately upstream from all that There's not much we can do to keep the bad people from doing what they want to do That's really something about that regulators That law enforcement and others need to do at the last mile But this has always been true with Open Source technologies We've had this fight for 25 years about end-to-end encryption And the concern was if you could allow people to communicate Without the government being able to listen in That would enable the terrorists and people sharing illicit images and things like that And allow the bad people to do bad things And probably there's been a bit of that But we've gotten so much value out of these tools And we've built a lot of the security and integrity of modern society On the back of end-to-end encryption and public key encryption That we've evolved, we've been able to make the trade-off of risks and reward work Likewise, I think with AI, we'll be able to find a way to allow the rewards to really shine And overwhelm the potential for risks and bad effects This executive's, of course, order is focused on the US You also gave example, a lot of work is going on China leads, when you look at AI What kind of effort you're seeing from other companies What is good about it? Because in other cases, we have seen that the governments were chasing You know, I mean spam calls, we have not fixed that You know, your car insurance, sex parity, we have not fixed that So it's good to see But if so, we have to look at it number one from global perspective Because it cannot be just a US-based executive order is not going to do much A second part of the question will also be that Do lawmakers really understand the implications of AI? Of course, Linux Foundation organizations are there to work with them, help them But we don't know how this genitive AI technology will be used Two or three years from now But these are good foundations So what do you think about these two perspectives? I think it's important that AI technologies be broadly accessible I was at a meeting yesterday with a number of dignitaries And other important folks because we're hosting In San Francisco this week, the APEC conference And I was talking with the ICT minister for Uganda Who was telling me about how they use open source software Across their government infrastructure As a way to not only cut costs, but to be able to do things And not be limited by a vendor here or a big tech there To be able to use their own Ugandan developers Their own universities and their own startup ecosystem To push forward and build the things that government needs And from their point of view, AI is just as important to their future As our White House and the United States views it as to their own They want to use it to improve crop yields They want to use it to manage energy better And so this is what's great about open source software has always been It's a digital public good This concept of digital public goods is now something the UN And many international organizations have been championing And the potential for this to really bring a sense of equity Across the globe to access to technology is real and visceral And it's something where even countries like Uganda Are coming up with a national AI strategy And they're working with their peers in Africa and Asia and in the West To come up with strategies that compliment each other That set common global standards for these things So that's something we welcome That's something that we think we should find ways for the benefits To reach even to countries like China Who are investing quite a bit in this And there's a lot of research being conducted in China A lot of patents being filed in China around artificial intelligence A lot that we can benefit from The work that they put into these foundational models And these foundational technologies So I think we want to tap into that And I think we've set up in the open source community A great way to collaborate globally across these boundaries And we should not shy away from that in this domain Since you are also known as nerd diplomat I want to ask a kind of not diplomatic but political question Which is that the word that we are living in today A lot of wars is happening The word is realigning What roles you see open source or Linux foundation can play Because a lot of things they go beyond political boundaries Beyond political agendas These are things for whole mankind So while governments can fight and argue Some of these technologies they should not suffer So what do you see is happening And what are you folks are playing there Because this is a very dangerous times Right, well to paraphrase Gandhi First they ignore you And for the first 15 years 20 years of open source software We were able to kind of get by with government ignoring us While we built things like end to end encryption And we built the global internet and built the web and that sort of thing Then they fight you Which I think we've gone through a period of open source software Having to struggle for respectability Struggle even today in some ways For around the cyber resiliency act And with certain folks in government Who don't understand how open source works Or how we're a driver of innovation And a driver of economic potential And then Gandhi said and then you win And I don't believe we've won yet But I see signs like I mentioned yesterday Where this ICT minister from Uganda And then others on the technology side Of the Swiss government yesterday There are lots of people in government now Who even if they're not technologists Understand the potential for open source software To be a driver for innovation Within their own countries A driver for government IT innovation as well Governments need to become IT organizations In order to deliver services to their citizens now And need to be better interconnected with their peers With other governments other nations And this is a new day I first started working in government in 2008 Early 2009 in the White House under Obama And I think I was the only person Who'd ever written a Python script Who worked in the White House when I first entered there Now this is becoming much more common To see technologists in government Recognizing this potential And I'm really encouraged by it The recent moves against globalization The recent moves in a very nationalist kind of direction Concern me of course The concerns people around the world I hope we can rest on the fact That we've always been global as an open source community And especially with support for so many languages now On so many open source desktops And we've got like Unicode support is great I mean all these things now That give us truly a global platform And I believe that'll persist If you're a developer You want a broader audience for your work If you have a question You don't care if the answer comes from somebody In a country like China If it's the right answer to solving your problem Right? If it's the right kind of technology So I think we'll weather this storm And I think if we stay focused on objectively Building great software that helps Solve problems for our businesses And for society We'll end up further ahead And I think open source and youth folks Have played a very big role To kind of build a fabric of bringing people together And kind of overcome some of these challenges Political challenges So yeah it's a very good foundation That you know for countries organizations to collaborate One thing that was missing from this executive order That has become a big issue Especially for Genitive AI Which is copyright And you know companies are blocking API I don't want to go into a longer discussion But the fact is the way we learn Thing is by reading, by talking So AI has to learn too There's a totally different copying Where it says you know learning So what are your thoughts on that number one Because if you look at open source licenses They are all copyright license And more or less you know that's what it is So for the growth of AI We need that at the same time There are a lot of concerns among the people Which may be unfounded at this point It could be misinformation Of course a lot of what happens all the time But I want to hear your thoughts on that And why you felt that the executive order did not feel That it was needed to address that at this point So it's important to think about AI models That being two layers right There's the foundation models That establish the basic way that languages work Right that give a infer a sense of common sense For a given you know what is you know what's How does a little bit of how society work It's just it's this baseline right And for that we've seen data sets pulled together Like the Allen Institute has actually been doing great work In this and here's the all of the content On the English language Wikipedia Right or the Hindi language Wikipedia Or on you know the other languages right Here is the corpus of books in Project Gutenberg Right here are all of these openly licensed Data sets that can be used to create language Now things like chat GBT were also trained on And then another models earlier on Were trained on things like Reddit comments And so perhaps it's no surprise When you train on Reddit comments that you end up with Occasionally these kind of dark you know Kind of hallucinations and biases and that sort of thing Because I don't know have you read it recently I'm kidding But I think if we pick the right underlying data sets For the foundation models And it doesn't have to be I believe the entirety of the web I think that we're going to get better and better At being able to train on smaller sets of data That's certainly the direction research seems to be heading in You have these foundation models And by the way this is a place where government can invest If a government in a country Whose language is not one of the top 10 languages Wants to see great support They could pull together their own corpus of text And really work to develop enough data To build a foundation model for a non top three language Or top 10 language The second layer obviously are the more domain specific Think of them as small language models Right or small other types of machine learning models And there's a great paper out there called All You Need is Textbooks Which suggests that you can in order to train A large language model on the fundamentals of biochemistry Maybe you just take the corpus of books That a biochemistry undergraduate and graduate student Are expected to read feed that to the model And that might be enough to allow them to that model To be somewhat coherent when it comes to discussing biochemistry So I'm really optimistic that we'll find a way Through this copyright kind of issue By being able to build more discrete data sets Better software that can train on smaller amounts of data And really work to develop data sets Where consent is very clear Another good example is the database of Creative Commons Licensed images from the Flickr archives You know Flickr CC licenses were the default We could train that on And in fact I believe that's the basis for quite a few Of the image AIs out there So when it comes to the executive order and public policy I'm frankly I believe they're kind of waiting For the Supreme Court to sort this out 20 years ago when Google started crawling the web And building their indexes It was ultimately the Supreme Court that said Google's actions in building their index are legal And offering that as a search to the public Is a form of fair use And that was a contested and a controversial decision But it allowed search engines to operate legally And I think we'll see the same thing emerge Where there's clarity that comes from the Supreme Court Or other decisions that say Well when something's published especially under a CC license Or under some other kind of clear consent license They can be used to generate these models But the most interesting kinds of data sets Obviously are going to be the ones that are not fully public The ones that have personally identifiable information And healthcare and finance and so many other domains You have these privacy concerns So as I said I think we're going to have to Get comfortable with the idea that there are some language models Where the underlying data sets are not fully open source Just as we're comfortable today with the fact that I can run Linux on my laptop like I am now And yet there's this binary blah blah That's firmware Right You know that I don't have the ability to see the source code of That is delivered straight and loaded into the AMD or Intel chip But yet I still think of this as an open source system Right So I think that degree of sophistication is arriving now In the AI ops kind of world The last question before we wrap this up is that You have been through You have been part of You have been the driver behind a lot of revolution Or you know technically you know that we have seen In today's world I mean we can go all the way to web and everything else When it comes to AI especially Genetic AI has been around for a while now Genetic AI is kind of rekindling interest in that What is scale of innovation do you see It is the same scale as for the kernel Apache Or you know Kubernetes Or you see that you know it's just like NFTs You know people will get bored of it next year Or you feel no This has massive potential there I think this is a bigger deal than NFTs I maybe we'll see NFTs come back who knows But in fact we just formed something called the generative AI commons Within the LFAI and data foundation To try to map out everything that's going on In the generative AI space And figure out how can we as an open source community Be helpful in driving Not just innovation in this domain But also you know think about the effect That Kubernetes had on the cloud ecosystem It was not just about innovation And not just about this terrific solution But it was also somewhat about standardization It was somewhat about saying Right there are all these different approaches To managing cloud infrastructure And building gigantic cloud systems But ultimately we need to condense On a couple of common tool kits and standards So that we get scale So that we can really turn everybody Into a cloud engineer Right and I think in the AI ecosystem There's a lot of different tools For building models for querying models To for managing these pipelines And we're probably going to see some consolidation Over the next few years And it could be that that's around PyTorch It could be that it's around some other type of platform We at the Linux Foundation are formally agnostic about that You know we want to be helpful in seeing As that consolidation happens Can it happen in a way that builds a coherent platform That is also a truly open source platform That as many people can build upon And then build interesting stuff on top of As possible Brand if I'm not wrong You folks are also hosting a conference Next month about AI So talk a bit about the conference Who should attend that And what is going to be the theme Yeah so December 12th and 13th in San Jose We're hosting the AI.dev event This is an event focused on Anybody who is either at the core Of building these tools for building AI models For building machine learning tools But also those who are using these tools And connecting them into their own applications Right we think it's really important That we reach out beyond that core Because everybody's going to start picking these things And plugging them into their tools pretty soon So it'll be a pretty nerdy event I think we'll spend more time on the nuts and bolts Or less on the policy And kind of public messaging and that kind of thing But it's really going to be a place Where we start to talk about convergence I think as well Of these tools And so what are the common platforms, common systems Our hope is that it grows into an event Kind of like KubeCon What KubeCon is for the cloud ecosystem This is for the AI ecosystem And we think with the Linux foundation We've got a great set of assets as a starting point But we know we need to do And could do more to push Are the open source AI community forward Then the event will be a key A key linchpin for that Excellent, Brian Thank you so much for taking time out today To talk about the PC order And also the implications of AI identity Thanks for all those insights And as usual I look forward to chat with you again soon Thank you Thank you Swap Take care