AI and Warfare

Welcome to Ai and warfare buckle up for A fast-paced conversation The Lightning Fast adoption of AI Enabled Technologies such as chat GPT Have supercharged recent conversations About how AI promises to transform the Future of warfare But will it and how Is the technology ready or is it over Hyped are military institutions ready Are the right companies ready to work With the military this panel Representing industry and the Department Of Defense will unpack these questions And more Please welcome to the stage our Moderator Rajiv Chan from Wing VC and This impressive panel Tara Murphy Doherty the CEO of govini Torsten rail the co-founder and co-ceo Of Helsing And Dr Craig Martel the chief digital And artificial intelligence officer for The United States Department of Defense Rajiv over to you Hi Joy man welcoming Tara to our student Craig to TechCrunch welcome guys Yeah thank you So I'm going to start this session with You Craig Okay what does generative AI mean to the Department of Defense And is it hype or is it real do you want To be honest or you want me to give the

Actual pitch that I normally say oh Totally honest ly um I think it's Probably 70 hype but my team doesn't Like when I say that and the reason is Because I could be wrong and it's really Important my individual personal opinion Is actually not the relevant points Um I I think there is a gap between Where research said like as a computer Scientist I am hyper excited about the Research that's happening now large Language models are doing some of the Coolest things uh in my uh in my career And then there's what companies are Selling and I think there's a gap there Now the question is how big is that Gap And and so we've been spending and again My opinion is not the relevant point so We've been spending a lot of energy Trying to figure out how to characterize That Gap and how we can measure that Gap So both we as consumers can say uh well That's something we can buy and it will Actually solve our problems and Companies can say oh well we want to aim Towards a particular Point uh and so That we so that customers will want to Buy our stuff How many folks agree AI is 70 hype and How many folks disagree AI is 70 hype Okay maybe five percent how many folks Disagree with that statement you think It's more hype you think it's 80. They're saying that

Taurus doesn't tear your thoughts yeah It depends on which part of AI we talk About um so what we found is with Genitive AI um what happened with chat Gbt about nine months ago or so is that It became much easier to explain what AI Is and everyone got interested in Ai and Everyone understood actually it's a big Deal to what extent generative AI is Actually applicable right now for the Big problems that we need to solve with AI in the defense space I think is a Different matter there are different Deep learning approaches right now where We can get 10x capability gains without Generative AI so I think we might agree With that yeah so more than anything Else I think it's it's been the Perception that's really been helped so People now get AI is a big deal I think Maybe we just need a little bit more Imagination guys So when you mention that the Department Of Defense is focused on the gap between What generative AI can do today and what Companies are claiming Shouldn't we actually as a department be Focused on where is the military today And does your commercial offering get me A little bit farther So I think that I'm saying something Really analogous so I don't think I'm Saying something contradictory to that I Think what we need to focus on is not

Technology but capability and necessary Capabilities to deliver a mission and if We can get that as clear as possible Then we can have a perfectly rational Conversation about whether the Technology being delivered uh satisfies That capability or not so I think we're In line with this so Craig and Tara do You think that AI will transform that Capability set for the US Military I think it will absolutely I think the Question is over what time period and What do we decide the ongoing role of The human in the loop needs to be where Is that human exist but also we should Realize that especially if we're Thinking about well how are great Commercial entities using Ai and Generative AI capabilities today for the Military it doesn't just have to be on The battlefield correct you know great Companies are putting AI into their own Product offerings their equivalent of Their Weapons Systems and Military Platforms and then they're using it in Everything they do DOD needs to do the Same in the applications for generative AI in that sense are huge Craig do you Agree that AI will transform the U.S Military's capabilities absolutely and It already has I mean uh as as you Mentioned earlier The Transformer technology that's Happened since 2017 has already

Transformed our capabilities like the Ability to Interpret images the ability to search Through text the ability to talk to your Systems these are massive Transformations that have taken place in AI so my my my sort of Drop the mic 70 really was about Generative AI hype but AI in general I Think has been massively impactful in The usdod and will continue to be so and Then Craig talk a little about your Generative AI framework that you're Working on oops so sorry Um that was irritating wasn't it um uh So so when we look at when when we we Just launched something called task Force Lima and task force Lima uh only Because they like the letter L it has no Other meaning than that Um uh really has two parts to it the First part is for us to analyze Internally what's the actual demand for Generative AI within the Department of Defense and can we categorize that uh in Two categories where we think it's safe To use and where we think that there are Still some dangers that need to be Mitigated and and the largest danger for Us is hallucination hallucination can't Be the responsibility of the user Solving hallucination cannot be the Responsibility of the user the system Has to come with that kind of Integrity

Already built in and so we're doing that Internally but we also want to push Externally to say what what is sort of a Maturity model for large language models Analogous to autonomous driving level One through level five what's the beauty Of that autonomous driving ranking System well you can see that the Industry kind of asymptotically reached Level three and then there are some Companies working for four and five but Then we know what we can buy from which Companies because it's clearly laid out What it means to be level one two three Four or five I'd like to see something Analogous for AI Technologies so we can Actually with accepted with measurable Acceptability conditions where we can Say your offering is level three oh we Need a level three perfect you're you're Shooting for level four you're not quite There but let's keep talking and maybe There's some funding that we can help to Help get you to level four the short of Maturity model worked really well for Autonomous driving and I'd love to see It for generative AR as well trust in That what pace do you think AI should Permeate sort of U.S and actually Western military purchases and I guess In short should every procurement today Be AI enabled I think so so what we're Finding now is and we I should probably Explain where defense AI company we're

Mainly active right now in Europe in in The UK Germany and France and and we've Just been selected to provide the Electronic welfare capabilities for the Eurofighter and there'll be AI enabled Electronic warfare capabilities and the Difference performance that you can Achieve with AI with deep learning is um Is huge it is transformational and the More some use cases but not for all use Cases correct um but there the more we Are in this domain now and we've been Going for two and a half years now the More we find use cases that have that Kind of transformation capability Getting and I think there should be a Real sense of urgency in Democratic Countries to start deploying this Technology do it in a responsible way Which I guess we get a chance to talk About as well but do so swiftly and in Some cases in some Pockets that happens But in many others it doesn't happen Right now so a question for all three of You and Craig this is the this is the Sort of I think difficult question for The Department of Defense Do we think that the US is moving as Aggressively as it needs to In AI for the Department of Defense Yeah for all three of us are for me for All three of you Tara we'll go in order Yeah absolutely not definitely not I think it's an ill-formed question

Because I don't think we should be Asking about a technology technology I Think we should be asking are we Satisfying capabilities fast enough and I don't actually care if it's AI or not Torsten I I agree with that point I Think it is always about capabilities I Think what we have to look at though is What's happening Um in China and the speed of development There and whether that's AI or other Methods and also what's happening uh With Russia and Ukraine and specifically For Europeans the nature Eastern flank They are big capability gaps right now That we have at the NATO Eastern flank That need to be filled and at the moment Ai and at scale AI systems will be the Best solution for that great Craig what Do you say even if you sort of go to the Capability level there are AI informed Capabilities versus non-ai and form Capabilities so but the argument is and I hear this over and over again that the Department of Defense is not working Fast enough on AI How do you respond so number one I don't Think that's true actually Um it's the world's largest organization And there are all kinds of pockets of Really great Innovation going on and um So I think it's hard to talk about it Monolithically so I think that's Important

Um but look I think so we really think About a hierarchy of needs uh logically Not temporally we have to do all of These things at the same time but AI is Only effective if we have high quality Data and if you gave me a certain part Of of uh resources I would first spend That on getting high quality data Secondly I would spend it on a good Analytics dashboard in metrics so we can Measure our success because I think Eighty percent of the use cases that I've heard within the dod for AI are Satisfied by quality data and being able To know where my stuff is and so sure There are some Exquisite use cases for AI and I think we should jump on those As quickly as possible excuse me quickly As possible and we are and and we really Try to support that but I think we Shouldn't put the car before the horse And we even though AI sells right now And it's the sexiness I would much Prefer to spend my energy on building a Data mesh that allows ourselves and our Partners and allies to have access to The data they need I I agree with that Can I chime in so um first of all the Point about the amount of activity Happening in DOD is an important one so Over the past few years the Department Of Defense if you look forget the budget Request which is the plan look at what The spending is look at how much money

Goes out the door every year on Artificial intelligence machine learning Autonomy and related Technologies and It's over eight billion dollars a year There is to your point Craig a Tremendous amount of work being done and That work isn't all science projects you Know you mentioned Transformer we've Seen the collaborative aircraft Um Advancements there's a lot of Translation of those efforts into what I Would argue are military capability The point that I especially love you Know coming from gavini which is a data And AI company and I said data first About data being the prerequisite and Underpinning to any quality artificial Intelligence is spot on and it's not That I want DOD to just throw more money At this you know that number I just Mentioned a billion dollars a year is Pretty big to begin with it's that those Table Stakes elements I consider them Table Stakes of quality data at scale Available across the across the Enterprise and analytics and indicators On top of that get more serious about That so we can do more translating AI Into true military capability I 100 agree with that strategy in fact That's what we're moving out on exactly That strategy Even with eight billion dollars is the

U.S behind China In its Ai and AI enabled capabilities In the military I'd argue yes to that I'm starting to sound like the pessimist In the group look at China's stated Vocal uh AIMS in this regard look at Their amount of spending into artificial Intelligence they have been so outspoken The Chinese Communist party to be clear About unseating the United States and Its Global Leadership position and in Particular using artificial intelligence As a way to do that and the United States has been in comparison relatively Muted now the Department's AI strategy Is an example of a Counterpoint to that But they're how do we know and and Gomini's in the center of data and so How do we know that China is ahead of The U.S so the stat that I've seen is a Scale AI scale AI quote that China is Spending between one and one and a half It's a military budget and AI while the U.S is spending between you know 10 Basis points and 20 basis points you Know in addition to spending as an Output look at that as in addition to Spending as an input look at the output And so if you look at areas like core Sciences you'll see that China is ahead From a patenting perspective and so what They're getting out Of the system for the dollars that They're putting in I'd argue is more

Productive than the United States so in Terms of stats of military specific Patents or these are just overall these Are overall technology patents and any Other stats any other sort of data Underneath the kind of the view that China is ahead of us Um I think the uh the rate at which They're amassing military capability is A testament to not only the Innovation That they're driving with that spending But also their willingness to steal Intellectual property from the United States which we all know is Um uh you know proven method for them so You look at the aims they have if you Want if we want to stay in the military Zone look at the numbers that they're Pursuing from a military capacity Perspective and that's another indicator Where we take ships for example we Cannot meet our own American Shipbuilding plan and yet the Chinese Are wildly outpacing Us in that Development Torsten from the European Perspective is Europe ahead or behind China in its Ai and AI enabled Capabilities for its military we only Have any of those evidence right now but The and and quantitative evidence like Number of patents is always with to Actually judge to what extent you know That that actually is valid or not and But what we're seeing is that that China

Has a different attitude to collecting Data for starters they're much more I Think happy to collect huge volumes of Data they're much more used to doing That and also than to applying this is China has server Fusion uh civil Military Fusion so they are able to Compel civilian companies to work with The military in a way that we rightly Can't and so I think they have some Structural advantages I think on that Side where I think we feel should feel Quite good about what's happening right Now in Democratic societies is that we Are getting now to critical mass in some Companies and organizations of talent of Top talent that really wants to Contribute and really find creative Extremely Advanced solution to these Problems in a way that math and just Data can't solve it so I think what We've been finding in the civilian space We had companies like um in the AI space Like Google Facebook peepmind Etc where You had this critical mass of amazing Talent we're starting to see this in the Defense space as well Andrew is an Example here I think in the US we're Trying to be one in in uh in Europe and One shouldn't underestimate how much of A difference that makes because the you Know China may have tens of thousands of Different ml Engineers but the reality Is it's actually just a few hundred that

Will really move the needle and you need To get those together to form critical Mass to really create those big Capability advantages so Craig the U.S Military is obviously the strongest Military in the world do you do you feel That there is a risk that China is ahead Of us in this domain I think that's such A hard problem to such a hard question To answer there are external indicators That we can look at but we don't really Have the indicators to the degree to About how those things transition to Actual capabilities deployed Capabilities that's completely opaque to Us it is true that they have the ability To sort of demand companies to do what They want or shall be here right I'm not An ex I'm not an expert at China Um I'm actually very sanguine though Because I think we do a really good job Of transitioning uh the best technology To our to the to the use cases that we Need fairly quickly I know not as Quickly as companies would like but but Is there a I grew up in Mobile and I Remember the burning platform you know Metaphor by Nokia CEO many many years Ago this is not analysis but but is There is sure that 8 million be 80 Billion and it should go from 8 to 80 Much sooner than it is Well so again that's that well number One that's putting me on the spot so I'm

Not going to answer that particular Question Um and and number two I I think we just Have to ask we can answer that question By asking what are the capabilities we Want I do not believe at priori there's Some number of money some amount of Money we should spend on any particular Technology we have to ask what's the Mission what are the capabilities we Want for that mission and what do we Need for those capabilities it might be Some brand new Transformer system it Might be some old and crusty linear Regression logistic regression based System or it might be a good dashboard To help a decision maker make the Decision more quickly those latter ones We wouldn't call AI today so I'm really Hesitant to say we need more AI we need To be clear about what our what the Capability gaps are that we have and Figure out how we can better work Completely agree better work with Industry to fill those capability gaps Let's turn to use cases so Um uh I read Craig I don't know if this Is accurate or not but I read that your Office has 700 or so AI projects under Its review that was a GAO study I think From from earlier this year but whatever The case may be what are the what are Two use cases that you are most excited About that you would love this room of

Founders to work on yeah to help you Enable through AI technology yeah Um look we need a anytime we can Understand our environment better that Is a huge win so computer vision huge Jumps in the last 10 years let's keep Pushing that let's keep figuring out how Computer vision can fit into particular Missions for particular capabilities Natural language processing you are 100 Right there are billions of dollars in Potential savings if we can make our Business processes better help me with Natural language processing there Because our documents aren't even Readable by humans never mind so so the Data might be hard right and I would Love to see the promise of large Language models Here's my problem with large language I'm going to switch to put my science Hat on for a second There's nothing about the loss function That constrains them to tell the truth And there's nothing about the loss Function that constrains them to reason Correctly Stop selling me uh magical systems that Will answer my questions I just don't think it's true I think Saying yeah yeah we'll get to Hallucination later is not as Disingenuous Solve that problem build me systems

Where I can ask questions and get Trustworthy answers because people's Lives are on the line I would be super excited about that I'll just give an example of that point Yes great let's go talk great Um So perfect beginning that was not a Promise of a contract in any way I'm Gonna hold them to it so one of the Things that gavini software is used for Is supply chain analysis absolutely so Your program executive officer or Program manager for that next Generation Unmanned undersea vehicle and you're Trying to figure out beyond that one Company you did award the contract to Who are the literally hundreds of Thousands of suppliers that feed into That We have heard from program managers who Are afraid of data and afraid of AI we Hear things like you can't give me the Wrong answer no data is better than bad Data Not knowing something is better than Being told something and it's incorrect That I that approach of I'm afraid of What I don't know or my inability to Either trust the data or be wrong in a Decision because of it has to change if We're going to move past where we are Today and actually do this so this is Why we need this maturity model I agree

With that because then you can say Here's a use case like first draft Generation totally okay a first draft Generation has hallucinations because You're responsible for getting rid of Them and your job is on the line right And there might be say that discovered a Level one sure level five might be I am A new doctor working a robotic surgical Surgical system robotic surgery system And I have to ask it a question I'm Pretty sure we want five nines accuracy There we don't we don't want it to cut In the wrong place and you can imagine Military contacts where we would want Five nines accuracy and a level 3 which Might be what you're saying might be Something like information retrieval Where there's a confidence level Attached to each of the facts exactly I'm open to all of those things but we Have to be able to we have to be able to Classify the use case and specify the Acceptability conditions if we if my Office can help help us as a community Specify those acceptability conditions Then you can take away that fear Then I Then we can go to someone and say yes You can buy it if these acceptability Conditions are met and then you can Demonstrate that you've met them that's Part of our goal for task force anymore I think that makes a lot of sense as Soon as you

Um as soon as you drill down from Something that is unmanageable in the Ecosystem of the data that they have to Work with to something that is Manageable for a human to make decisions Around that risk plummets Tricia your very quick thoughts on kind Of use cases that you're most excited About or eager to solve yeah there are Already a number of I think very large Use cases so they're not niches they're I think quite broad um where I can make A transformation difference one of them Is electronic warfare Um I think in a few years time uh most Electronic warfare um will be AI enabled It will have to be Um we won't have a chance otherwise Against software-based emitters uh Anti-submarine Warfare uh will also be AI enabled AI can make a massive Difference there we have specific we Have very specific applications ideas What that is uh saturation attacks Against air defense a massive Application I think we're currently Completely underestimating how badly We'll deal with saturation attacks um on Our different systems if you can to some Extent automate this with AI uh we we've Shown you can make a massive difference And and the fourth and very obvious one Is is the battlefield and is is Compressing decision chains on the

Battlefield to enable humans to make Much better decisions making much faster And be much more reliable that might not Be entirely AI like part of that might Just be everybody being being able to See the same 100 100 A specific question within use cases is The degree of decision-making Authority That we will give AI at some point in Time Craig do you believe that AI at some Point will be given autonomous lethal Decision-making Authority No No let me tell you what I mean When I drive my car I have it has some autonomous Capabilities and I practice with that Car to figure out the degree to which I Have Justified confidence in those Capabilities so I'll tell you honestly About my car I won't tell you the brand To ask about my car I trust the adaptive Cruise So I make the responsible choice for Which I am ultimately responsible I'm The agent that makes the choice to turn That on Now there are millions of decisions Being made per second so clearly there's Not a human in the loop when it breaks When it speeds up when it decelerates Right those millions of decisions are Being made by the system but I as the

Responsible agent trained with that System and the military is really good At training trained with that system Understands the limitations of it and Know where I feel comfortable deploying It or not and where I feel comfortable Taking responsibility for it And the contrary the lane keep assist Bounces back and forth and always hits Truck tries to hit the truck so I would Not deploy that I do not have Justified Confidence in that and that's the way we Think about this if you ask me are there Decisions yeah there's millions of micro Decisions the system is making but Deploying A system that is going to go Do that thing that will always be a Commander's decision and one for which He or she feels Justified has Justified Confidence in the system I'm going to Ask the audience again how many feel That AI should never be given autonomous Lethal decision-making Authority How many feel that AI should at some Point be given lethal decision making Authority Yeah so this is fascinating this was Also in my research I spoke with the Chief technology officer who said who Served Afghanistan And said he would love to have had an Autonomous Sort of system once the data confidence And Trust had gotten to a point where it

Will be perfect even he said it won't be Perfect Tara do you what's your thoughts on this Topic and then interested and then yeah I'm surprised by the results too to be Honest which is why it's so good to be Here and engaged with this group and Outside of the DC Beltway for once it Reminds me of and I'm gonna date myself Which I shouldn't but it reminds me of When we stood up U.S cyber command and Grappled with this same question Regarding cyber operations and you know The idea was you could have any civilian You could have a contractor you could Have any employee of U.S cyber command Or in some cases NSA perform all of the Functions of an operation right up to The point where an action was taken that Had either lethal implications or the Potential for lethal implications and at That point in time a uniform uniformed Member of the United United States Military had to walk over and hit enter I mean that facetiously hit the button To complete the operation that principle That there is always someone in the United States military who is the final Authority to say yes hit the button when We take a life away anywhere in the World That's going to be hard to roll back no Matter how great technology is so torch I'm going to ask you the last question

Um will will the US's adversary and the West's adversaries have autonomous Lethal decision-making military systems And if so will that put us at a Disadvantage So I believe that will happen I think We'll find probably quite soon that There'll be Lodge reconditions that will Be fully autonomous and they'll be used By adversaries with a different set of Values and I think for us as democracies We have a special responsibility to use Our technology thoughtfully to stick to Our values and that means that we have To create essentially a technology Leadership that allows us to have the Luxury of having a human in the loop I Think we'll find that it will be Tempting depending on the context and The future of adversaries using this Technology to also take a human out of The loop in particular in highly Contested environments what we're seeing In Ukraine we're deployed in Ukraine With with our technology and we have People on the ground all the time it's a Highly contested environment and so you Don't actually have communication Anymore very often you won't be able to To see um or to actually instruct a Particular drone what to do Um that's going to be the new reality And I think it's incumbent on us to find Ways to deal with that and still keep a

Human in the loop despite these Disadvantages again I think it's Democracies that's our Responsibility so I hope that for a very Long time hopefully forever we'll be Able to stick to that principle We'd like to thank Tech Garden session we'd very much like I'd like to thank Tara Craig and Torsten For their thoughts and thank you have a Great rest of tech Ranch Thank you


Coinbase is a popular cryptocurrency exchange. It makes it easy to buy, sell, and exchange cryptocurrencies like Bitcoin. Coinbase also has a brokerage service that makes it easy to buy Bitcoin as easily as buying stocks through an online broker. However, Coinbase can be expensive due to the fees it charges and its poor customer service.

Leave a Comment

    • bitcoinBitcoin (BTC) $ 66,540.00 0.68%
    • ethereumEthereum (ETH) $ 3,595.43 1.01%
    • tetherTether (USDT) $ 0.999024 0%
    • bnbBNB (BNB) $ 608.37 0.34%
    • solanaSolana (SOL) $ 148.32 2.4%
    • staked-etherLido Staked Ether (STETH) $ 3,595.52 1.08%
    • usd-coinUSDC (USDC) $ 0.999868 0.02%
    • xrpXRP (XRP) $ 0.488865 0.28%
    • dogecoinDogecoin (DOGE) $ 0.136061 0.12%
    • the-open-networkToncoin (TON) $ 8.03 0.42%