Reimagining your technology strategy for sustainable, responsible and innovative AI Reception
thank you for joining us this evening at the AI house my name is Neil McDonald I lead the compute business for hulet Packard Enterprise and I'm thrilled to be with you here this evening with a really interesting group of panelists I'd like to begin by thanking our co-host cruso for this event cruso is on a mission to align the future of computing with the future of the climate which is a critical critical challenge for all of us in this burgeoning world of AI um we're going to spend some time this evening talking not just about the impact of sustainability on your technology strategy for AI but also how you address responsible and Innovative AI on your journey and to discuss that we've got a great panel I'll introduce our speakers here firstly our co-host Chase lochmiller who's the founder and CEO of cruzo then we have yuris Port who's the founder and CEO of rescale which is a cloud-based platform for computational and engineering and R&D and then finally we're joined by May Habib who's the founder and CEO of writer which aims to transform work with a secure Enterprise generative AI platform so with no further Ado let's get to our first question of the evening and it comes in two parts for each of you what excites you most about the AI opportunity and what needs to happen to make that opportunity accessible to more people in society Chase sure um you know I think 2023 was really a transformative year for the AI ecosystem um it was a year where we really started to see um applic applications you know impacting you know Enterprise and you know uh creative Sparks of you know what's actually possible with this new uh platform of computing and new new platform for for Innovation um so you know I really feel like we're in the early Innings here we're sort of in the first inning um in terms of how it's going to transform work how it's going to transform creating how it's going to transform the way we live the way we invent the way we do everything um when I think about the accessibility of things uh you know when we look at cruso Cloud our our GPU cloud computing platform uh you know it's very uh you know I would say a lot of our customers are filled with uh teams of machine learning phds and people that have tremendous amount of experience in terms of understanding you know big Transformer based models and really understand their overall model Pipeline and architecture um there are only a finite amount of ml phds in the world and you know we're we're going to run out of ml phds to run these workloads in the future so uh when I think about how AI is going to you know become more accessible uh you know building further up the stack in terms of creating uh uh AI you know uh operational tools that that actually make AI easy to use for for any Enterprise in the world and and anyone that doesn't actually know sort of the underlying uh modeling capabilities and and sort of what they're what they're building in their pipeline so um you know I know a lot of people are hard at work at that today um including you know youris here who's who's really helping uh uh you know that transformation take place in in the accelerated Computing for um all sorts of hyper performance Computing workloads what do you think yours thanks for the great transition Chase um so at rescale we're focused on empowering you know engineers and scientists they're running generally large physics simulation models things like that uh with high performance Computing capabilities so AI is pretty new so as Chase mentioned we're definitely in sort of just the first Innings here um we've really seen this kind of take off with images and and language and text but there's so much more potential right and I think we're starting to see that now uh emerge in the like industrial sector and Manufacturing and life sciences and Drug Discovery um but still very early and so um one of the things we're focused on is helping make it much more accessible to the sort of Engineers and scientists that are building the future Innovations how we do that is going more full stack so uh one is making it easier to use like Chase mentioned um being able to put it into the layer uh make it familiar without having to have the PHD machine learning uh to be able to take advantage of those capabilities and maybe the second piece and I think HP plays a very important role here cruso Cloud um making it cost effective and efficient and easy uh to access the right compute resources uh to be able to sort of take the data you have and and be empowered with those to embed these AI workflows in the sort of day-to-day work that everybody has because if we took kind of the the AI work that happens today and and gave it to every person uh doing serious uh kind of innovative work uh we wouldn't have enough Compu power wouldn't have enough energy to power all of those things so having all those optimizations uh from making it much more energy sustainable likeo focuses on uh making an optimized stack that HP focuses on I think is absolutely critical to bring that capability long term to to everybody and may you're you're targeting a different set of users but you're on a very similar Journey so what's your what's your take on the most exciting opportunity and what we need to do to make AI truly accessible yeah yeah um and I think maybe I want to take a bit of a contrarian just Neil you asked me to make it interesting right um I think a lot of people in the Enterprise and we serve Enterprise a lot of users their first experience with a chat GPT was like big whoop it's remixed Wikipedia like you expect this to replace me nah and I think the making adoption accessible is not just here's an AI panel at the side of every tool you're using be more productive it's really about Reinventing and rearching business processes AI first and that really is going to require a focus on a last smile that like even us as full stock people really ignore it's the last mile of workflows and it's the last mile of data um and that's a big part of what we're focused on um a lot of people comp compare this moment of AI to the iPhone except when we got our iPhones we were kind kind of obsessed we use them all the time and I would Hazard that even power users of AI today of generative AI we're using it a few times a week at tops right and so a lot of making the technology accessible is really thinking like workflows first I think the second piece this is really exciting stuff is we think about like all of the new capabilities inside of um platforms like ryer and and other full stack platforms we have now the potential to give every team the ability to build their own tools build their own software and that is going to require a new type of person not someone necessarily with like all of the coding jobs but someone who can think about with business Acumen what those workflows are what the AI generative AI workloads should be and really help companies architect that we're calling that an AI program director type of role we had our first conference for those people among our Enterprise customers in in December and I think that's going to be one of the fastest growing new occupations are people who can help conceive of new tooling for their teams that are really custom built so those those are all going to be required Transformations for um adoption to kind of go where all the promise I think is thanks M and of course another big transition that we need to deal with is how do we do all of this sustainably because these workloads are the most computationally demanding workloads we've ever seen as an industry and as a society and the amount of energy that's involved in feeding them as they become more accessible and more broadly adopted is colossal so how do we make AI more sustainable to manage and mitigate that environmental impact if we're successful in achieving the kind of adoption that we've just talked about that's a great question um so you know just starting first with the problem when you look at uh when you look at you know AI infrastructure what powers AI you talked about it being very computationally complex um and computationally intensive uh if you look at like a single h100 node uh and when we place that in a data center we budget 12 kilowatts of power uh draw for that single node um traditionally you know a lot of data center rack space um is engineered around a 7 kilowatt rack um so with a 7 kilowatt rack you can't even power one single h100 server um so you know starting first with like the data center design piece it's it's actually requiring the industry to reinvent what digital infrastructure looks like to manage the power and the cooling needs of AI workloads so that's like you know a first first thing that needs to happen um you know the second thing that needs to happen is really you know with all of that power consumption and and to put it in perspective for people like 12 kilow is like as much as uh 10 US households consume so it's like a single node 10 US households you know deploying thousands or millions of the you know you're talking about a tremendous amount of power consumption uh you know that the industry is going to consume um you know uh the recent forecast that I read was uh AI infrastructure is going to consume 38 gwatt of uh uh digital infrastructure power consumption which is just like a you know it's a mindblowing number it's it's it's absolutely astronomical it's you know multiples of the global data center infrastructure power consumption today so that begs the question all right where is all this power going to come from where is all this new data center infrastructure going to come from um honestly it's like a new problem that you know we um as innovators in the space have the fortune to uh be starting with kind of a blank slate um and we have the the the opportunity to really reinvent what the solutions are going to be uh so that we don't uh accelerate a climate crisis by sort of enabling AGI uh you know because if if if that happens it's like we you know 95% of the species on Earth go extinct but we have AGI like is that future we really want to live in like I I don't know um so you know one thing that we're focused on at cruso is is taking this very energy first approach to building the infrastructure needed to meet this demand um so what that means is we actually partner directly with the energy producers and we try to partner with uh energy producers where we can actually uh have on-site clean power generation either through uh carbon- free Power production or methane mitigating power production where we actually capture waste methane from oil and gas production uh or waste methane from landfills and can actually lead to a net emission reduction um and so what you're seeing you know just broadly as an industry is that uh the uh energy world is getting more tightly integrated to the large scale high performance Computing infrastructure world and they really need to work hand inand in order to deliver the solutions that the world needs uh to power the future of AI thanks Chase what your thoughts yours yeah I totally agree with a lot of these uh points Chase just made I mean I think if we kind of work backwards from the the overall problem uh we need a lot of innovation and a lot of AI to solve the engineering and science problems that I think faces society today and uh we want to bring that capability right um so we want as much use of AI as possible but again if you forecast uh kind of look at how much use there is today there's not enough a AI accessibility which you already spoke about I mean you sort of uh just data center consumptions about 3 to 4% of all you know uh emissions globally and if you sort of draw a line where that's going it's it's increasing exponentially so you need a lot of different uh I think Technologies to contribute to uh doing this all sustainably right and it starts with the energy source and I think cuso is an absolute leader in this field um taking you know energy that's wasted and and and powering data centers with that it's a very difficult problem very large data centers are critical um to AI right to be able to run these large GPU clusters concentration of energy is very important making that sustainable in a way that can be powered uh uh resiliently is is critical the second layer is you know there's a lot of uh compute capability just kind of sitting idle so have to actually look at most data centers there's still a lot of things that are not for example enabled by Cloud right so uh being able to tap into those sources and basically build that right network of compute capability to make it easy to access um is critical at the infrastructure uh layer and being able to make that accessible through cloud computing through SAS products and things like that maybe the third and final lever is really about the optimization of that stack so I think if you've done any like large language model training yourself or any of these kind of big Endeavors it's extremely expensive and it's very difficult to optimize and tune these models and so um building the right software throughout the stack uh to make it really efficient in the way we actually run the Computing is absolutely critical so that's not just the right infrastructure but also the right of that software stack kind of perfectly integrated right um and so that's the vision that we also had rescale tried to help enable and then bring that and have as many engineers and scientists use AI uh to basically build much more Sustainable Solutions whether in energy in Aeros space in automotive and that ultimately will come full circle that way thanks so inspiring like we're able to do what we do because of you guys's work I think um maybe two ideas to contribute um one is and we we've validated this and others have a chat GPT query is 60 times to 100 times more energy intensive than a Google search so if you are powering use cases that really should be Excel macros with gp4 like not good for the environment and so a big part of a like sustainability lens to generative AI adoption is what you use cases should we be using um another aspect to it is just the size of these models um we just trained and chipped in production a 702 Bill parameter model that's at the top of the leaderboard of Helm the St Professor Percy is here um and like that's a model that can fit and run on two nodes and so I think as adoption gets more sophisticated we also got to understand like what workloads do we put on what size model and really have sustainability in mind as as we do that because it's going to be a while because before the um kind of energy clean energy production catches up so really when that all comes down to is an endtoend reflection on sustainability from the energy source to the data center design to the system design to the stack that's been used and how optimized it is to the selection of the workloads you really need to consider the whole end to endend chain in order to make great decisions from a sustainability point of view so I guess the third question here is about the impacts from an ethical and societal perspective of AI you're all professionals leading the way in the adoption of generative Ai and the infrastructure to power it what are you most concerned about in terms of ethical and bias risks in AI and what do we need to do together to prevent them and maybe we can start with May and work back yeah sure I'm most concerned like seriously concerned about lots of good people not being able to shift their skill set fast enough and you think about the kinds of jobs that are aspirational in developing economies and like even in just 18 months right the workloads that AI is capable of have gone from ooh this is stuff my EA and chief of staff could do to holy crap like this is expert level stuff and I do worry about the speed of like non-o non-private sector actors that are required whether it's like the education sector or the government sector and helping prepare Society for for that um shift we think um our role as a tech company extends to the adoption piece which is why application layer and use cases and sending people on site and trainings and all hands and like we show up to try to get as many folks as possible learning learning how to use the tools not just the 15 20% of your employees that are naturally early adopters to anything and are suddenly 300% more productive than the other employees you've got so the social impact is going to be massive and I think we um we underestimate the the the impact of it what do you see is the biggest risk and what do we need to do to prevent it yurus I think um overregulation would be my uh my take on it um I think it's you know we got to be really careful and uh I think it's really important to sort of uh innovate responsibly right um but if we think about overall um I think the solutions uh to all these problems certainly can speak for sort of engineering and science lie in leveraging AI in in the right way right and and making that as accessible as possible um an interesting stat I saw was around kind of us um employment rates and it doesn't you know change that much if you look at like different technological impacts and what while it doesn't change much what actually happens is about 25 to 30% of the workforce every year gets reskilled and retrained in new capabilities and um I think it's an important thing to remember that as long as we provide the right education and accessibility as we have uh spoken about um we can actually uh use AI to solve a lot of these uh new challenges in uh you know I think if you look at engineering and science we have to solve all the energy uh solution problems um we want to live longer healthier lives and and in life sciences is critical uh what we're doing in science and uh things like AI are a key enabler to allow scientists to be you know 10 20 times more productive we have mod physics models on rescale scale um where instead of running the physics simulations with discrete calculations we can use AI to do them a thousand times faster now of course we got to be really careful are these AI approximations for physics um from a regulatory perspective and a certification perspective uh going to really work right um but you see it used in drug Discovery you see these probabilistic methods being extremely effective in in making engineering and science much more productive and I think that ultimately leads to a much better world with more accessibility to better products and uh what we just need to really focus on is making sure the accessibility to everyone um and the retraining and reskilling uh to allow all these people to kind of move up a layer of abstraction in their role because if I look at uh something as simple as llm models and chat GPT I think the basic like 80% language answer of an email or a press release or notes for a panel for example um are are something um you know maybe not super productive work for society if we can just automate it right and then if we can rescale uh the folks that are those knowledge workers on those problems to do much more much better more interesting value added uh productive tasks I feel Society is much better off so you you touched on we're going to stay with the orus a minute because they gave me a hook here you touched on the uh risk of as you put it overregulation others would contend there's an enormous risk from under regulation and you know may talk to the need for other actors to keep up with the rate of innovation here when you think about the diversity of community that we have gathered here at Davos and at the AI house what council would you have on how we find the right balance together on the regulation Spectrum yeah I think it's a great question um I think it starts with uh access and enablement uh to the data and accessibility a lot of Open Source I think can play an important role in the technology sector right uh to be able to co-develop and have sort of equal access to these Technologies at different layers in the stack um ultimately it's I think a lot about commoditization bring the cost curves down right so if you think of things like uh the rep platforming of uh mobile or the internet uh these have been hugely democratizing solutions for society right um a mobile phone can cost as little as like three or four dollars these days right um a uh not the one ones you probably have but you know uh it can't get that cheap you know they're providing these super cheap phones internet is is getting close to free right um and so AI will be also a rep platforming I believe um that will actually be be um you know very deflationary kind of uh uh solution right um and something that quickly is accessible to everyone you see it with these large language models of course there's the leaders who have huge data sets and and are spending hundreds of millions of dollars training these things uh but very quickly uh they're fine tune made smaller very accessible for everybody in a in a relatively cheap way um that doesn't all happen at the same time uh but I'm a big believer if you look at those other technology trends that you know the democratization will happen pretty fast and again with the internet and accessibility and Cloud uh the these things uh become possible very quickly thanks yurus so back to you chase what what are you most worried about what do you see is the biggest risk in the AI journey and what needs to be true from all of us to prevent it yeah so um I think there's a question around inclusivity and like you know how do AI models introduce biases that we don't necessarily want as a society um you know starting with that sort of premise and goal um I think It ultimately comes down to um a two-way street of if we want an inclusive model we really need inclusive data um when you look at any of these large AI models you know they're ultimately just large scale nonlinear statistical models that reflect the data that they're trained upon um so I think you know uh the data collection is you know ultimately becomes a a very critical key component to what the future of uh AI modeling looks like um and the AI models that people uh rely upon will will will inherently be um uh driven you know the outputs of those models will be driven by the data that's input to them um so you know I think there's an enormous opportunity I know there's a lot of people um with large Financial incentives that are sort of going after this space of just like uh getting unique diverse data sets um that ultimately will uh uh mitigate a lot of those bias and you know I'm I'm very optimistic about you know our um Collective effort to be able to resolve a lot of these problems um you know I I did like a lot of what may said in terms of uh you know uh you know the risk of AI sort of uh eliminating jobs I think is interesting and then I you know ultimately for me I think as a society we really need to be investing in educ a for like what the what the future Workforce is is is trained upon and what they're ultimately doing um you know it's a personal passion project for me I actually sit on the board of a uh Computer Science Education charity called CS forall um and CS forall is focused on trying to get uh Computer Science Education introduced um to the K through 12 uh education system um so that people get exposure to this at a very very early age um and I think you know ultimately uh I I I actually didn't know this stat that you shared around the percentage of the workforce that actually gets retrained every year um that's that's really fascinating and I think um you know ultimately what what that tells me is that uh the workforce is more malleable people are people are adaptive you know people uh you know we didn't when we invented the car it was like you know the people driving the uh horse buggies you know it's like they they adapted they figured out a new job that they could have in in their life and I I think uh you know you're going to see something very very similar I think AI is sort of unfolding in very unexpected ways um you know I think you know creatives and people uh in sort of White Collar jobs you know always felt like they were insulated from being disrupted by technology but uh and it was going to be all the the blue collar Workforce that was going to be disrupted by you know self-driving cars and all these things we sort of seen the opposite happen with AI like AI is actually been really good at uh generating new content um about doing you know uplevel uh ort of uh White Collar work so to speak um and it's actually there's a massive shortage of blue collar labor so um you know I I don't know exactly what the future holds but you know sometimes it unfolds in very very unexpected ways um and ultimately you know as a society we really need to be investing in uh the educational resources um to to to give people the tools to uh uh be productive members of society you something you want to add babe yeah I think this is especially at the W is an important debate to have we've had decades and years in other platform shifts to resk and educate people I'm talking in like a year most companies will be able to operate at 20% of their staff like 18 months what does that what kind of impact does that have um when we look at workflows across digital and marketing and custom customer success customer support operations like AI is going to eliminate already like the need for a lot of different people and the world's only going to need so many nurses and plumbers like there's going to be a lot of jobs that I think are impacted in a really short period of time I have no doubt about society's um uh courage and facing that but I just want to be like realistic with ourselves about the nature of the challenge or we just won't meet it like we're talking real short-term impact right thank you and thank you to all of our panelists for a great discussion this evening at HP we are committed to technology strategies that address AI in a sustainable and responsible way as well as an Innovative way and it's been wonderful to share some complimentary perspectives that hopefully have given you some thoughts and ideas of the components of a technology strategy to address that that you're going to have to build for your own organizations thanks for being with us here at the AI house and thanks again to cruzo for co-sponsoring this panel this evening thank [Applause] you
2024-02-03 16:16
