Chris Diehl at UC Berkeley Soc167

Chris Diehl at UC Berkeley Soc167

Recording: Chris Diehl at UC Berkeley Soc167

uh so my name is chris steele uh i'm a principal and co-founder of the data guild and it's a real pleasure to to be here to speak with you tonight thank you for sticking around i know it's tough to to be the last at-bat here i've got lots of pictures and lots of stories to share so uh hope this will be enjoyable for you what i want to do today is actually share some thoughts with you that i've been thinking about over the last couple years probably over the last decade i've spent a lot of time thinking about how organizations learn and how technology can be used to enhance that process one of the kind of early questions i asked myself was you know how can organizations or how should organizations operate in environments that are fundamentally unpredictable and just by asking that question i'm already kind of signaling some aspects of about my beliefs about this the state of the world and some of the challenges that we face and so over probably the first half of my talk i want to basically lay out those um those beliefs for you kind of frame out of perspective then i'll move into actually presenting an answer to that question and along the way we'll contextualize you know how technology can support that process yes sir oh where do i want to start so when i left graduate school i left and graduated in december of 2000 and went to join the applied physics lab at the johns hopkins university which is a laboratory that's basically focused on supporting the department of defense in the intelligence community within the first year that i was there obviously 9 11 happened it was a huge wake-up call you know obviously for the nation and for those of us in the national security community particularly this is a photograph that i took literally days after 9 11 standing on a hilltop not far from the pentagon it was a really kind of remarkable moment there standing there in silence with a number of people kind of looking at what originally was quite surreal you know it became something really quite real to us over the next couple years again those of us in the the community we're trying to think about ways in which we could uh mitigate that kind of future surprise um certainly moving into the kind of 2004-2005 time frame social media was on the rise you know this is obviously pre-facebook but the blogosphere was really you know the was was growing and very active and members of the the community were very interested in ways in which we could exploit social media to understand distant societies foreign societies uh the picture that i have here and that that you know i hope you can see actually is a representation of the persian what blogosphere circa 2008 uh this is work that was done at the berkman center in harvard they did the kind of first survey of this where they first did a broad collect over what they thought constituted the persian blogosphere they did some clustering of that graph and then they did actual hand annotation of those clusters to try and understand what they were looking at one of the most interesting things about this investigation or that would actually one of the most surprising aspects of it was just how active this community was in many respects you know as you conceive of iran and the the nature of that society you would think that open communication online particularly by notable individuals in iran would be fairly limited but there were folks here who were actually quite vocal in ways that were very surprising because of that interest built and continued to build in terms of trying to exploit these kinds of data sources and also think about ways and again that we could mitigate future surprise along that way i would say some rather fantastical claims came forth in terms of what we might be able to predict from these kinds of data sources and you know some of these kind of patterns of of belief continue today so you know i'm fairly new and relatively speaking to the silicon valley and the bay area i moved out here about five years ago and one of the really striking things for me kind of living through this growth of big data and data science and you know all the hype that's around that is just hearing kind of repetitions of of patterns of belief of what we can do with data right so in many quarters certainly within the us government and now outside in broader circles there's this belief that we're only limited essentially by you know how much data we can you know store and compute on basically the more data we can get the more we can know the more we can predict and there really is no limit to that and what i want to share with you today is essentially a counterpoint to that which is i want to articulate the belief that that there there are fundamental limitations that we face and we need to understand uh what regime that we're in you know so that some of the methods that you know omar was talking about in terms of machine learning and data mining and things there are certain regimes in which we're totally justified in applying those algorithms and expecting them to generalize there's other regimes in which that we shouldn't have any of those expectations and it's really important for all of us to understand where those boundaries are so again what i want to to first emphasize is that there are environments in which you know organizations operate environments that we're exposed to continuously that are going to be environments in which surprise is really the norm and and to really kind of emphasize this point what i want to do is first walk through some examples of significant surprises that we've all been exposed to in the last 10 years the very first place i want to start is in the political sphere i want to talk about about the arab spring okay so the arab spring uh so for those of you that may not be familiar with the history of kind of where this started it basically kicked off in a december 2010 front time frame in tunisia it began with this one particular gentleman a young man probably in his mid to late 20s who was a fruit vendor on this particular day he went out like many other days to sell his wares like many past days he encountered some corrupt government officials who came up and took some of his wares without paying he tried to bribe them to get that those uh the you know some of his fruit back uh that didn't work they publicly humiliated him this particular person actually knew his family was basically publicly berating him and cursing his family and all sorts of other nastiness uh the gentleman left that area went to a government a nearby government office tried to seek redress got no support he came back talked to his friends and said i'm going to show the world what we're faced with every day and after that he went away um showed back up that government building doused himself in gasoline set himself on fire that particular event that was really what catalyzed what happened over the next coming weeks so basically uh four four to five weeks later the ben ali regime in tunisia fell then ali left went to saudi arabia everybody was blown away that you know that basically the regime crumbled so quickly and then the rest of it i'm sure you you know basically that shockwave if you will then rippled across the region and we're still watching this story unfold right now if you were to say as some members of congress said to the intelligence community why didn't we know about this right why didn't we see this coming right can you imagine trying to predict that that particular action would actually set the region on fire you know a day that was not unlike many other days if we now move over into the economic sphere we we all lived through the the global economic or financial crisis of the two 2007-2008 time period here we had a situation where we had a number of different factors that came together right so we had complex financial instruments uh that were being traded that ultimately people didn't really understand what they were dealing with yes sure i'll try to shout more yeah thank you so in this this time frame that over you know for this particular event we have the situation where um we had a number of different factors coming into play so these complex financial instruments nobody really understood and were being traded they were also being misrepresented by by financial regulators we had a lack of transparency all of this kind of came to a head at one point when the marketplace basically just started to seize up and there was great fears about how that was going to ripple across the globe if you ask people who you know worked on wall street was this a surprise many would say no it was not a surprise we knew this was going on right so to first order that they knew there was a problem they didn't understand when that was necessarily going to transpire and the thing that really surprised everyone was that the pace at which the cascading failures happened so when these institutions started to fail and the repeated failure started to happen no one could anticipate those higher order effects that surprised everyone right so there's a certain you know kind of complexity in this system that is just basically impossible to really comprehend if we now finally switch to technology right technology is rife with surprise and when we talk about this typically we're thinking about like disruptive innovation where maybe some new breakthrough comes along that leads to certain types of products that then disrupt markets the the particular focus of surprise in this domain that i want to talk about because it's one that i have a lot of experience with is cyber security if we talk about the internet is really probably i don't think anyone would argue that it's the most complex man-made system on this planet and it's something that we're continuing to build out with the internet of things basically we've built this system that is so complex that no one can understand its vulnerabilities and we're building out this persistent asymmetry that causes a problem in terms of security because if you're an attacker you only have to be right once if you're a defender you need to be right all of the time right then there's a reason that you're see hearing these stories in the media over and over again about you know whether it's target or this you know other commercial institution or you know certain defense contractors that are having all of their intellectual property siphoned off by some unknown foreign government right these these patterns of behavior are going to continue because precisely of this asymmetry that exists and it's our inability to understand that complexity that drives this persistent surprise so now stepping back from those examples and just to review you know the the question is okay what what are those fundamental limits let's just kind of run through again what those factors are the first thing is is again in these when we have these complex systems whether they're social systems economic systems technological systems there's so much complexity and non-linearity in you know interconnections and interdependencies between these different factors that it's it's hard for us even in the best of circumstances to comprehend that so if i were you know if i were god and i could give you basically a complete visibility over one of these systems i could still have a problem where if i give you a certain amount of uncertainty and an input to that system that uncertainty can ripple through the system and blow ups to the point where you basically can't say anything meaningful meaningful about its future state but that's not the world we live in right we never get to see the full system right we're always constrained by limited observability we don't know how much of the system we're actually seeing we don't know if we're actually considering the most important factors right and we certainly don't understand how all of these things interrelate finally we have this issue of cognitive bias right all of us have inherent biases in how we perceive things in the world whether it's things like hindsight bias where things that happen in the past that we've been exposed to are things that we ascribe a higher probability to our higher likelihood to or things like post-hoc fallacy where when you see a temporal sequence of events you're much more likely to ascribe causality to those in certain cases where it's really not justified and this has to do with how we can think about these these events right we we tell stories we construct narratives right inherent in those narratives right just we're kind of developing essentially a reductionist explanation of what we think is going on and that we use that whether we're aware of it or not to make predictions about what we think is going to happen in the future and that sets us up for this repeated cycle of surprise so there's there's this gentleman naseem telev who's wrote a really popular book called the black swan and he introduced this concept of the black swan it is essentially a low probability high consequence event that was previously unforeseen right so there's a there's like there's a couple things to unpack there so first it the the event itself is not on your radar you're not thinking about it you're not it's just not something that you're necessarily going to conceive of happening in the future even if you do the thing that makes it high consequence is the environment in which it happens right so if you're trying to predict the black swan not only do you have to predict that the event will happen but the the the nature of the environment that makes it high consequence so going back to the example in tunisia you know he that gentleman lighting himself on fire in that particular environment right led to the events that we're seeing unfolding there could be other environments where he would have done that and we wouldn't have seen that that outcome trying to predict that is so difficult it's basically impossible so if you agree with what my basic premise which is in these these complex environments there's very little we can say about the future right then the question is if you're an organization whether you're a government or a multinational corporation or what have you and you're operating in these environments and you're trying to reason about them and make decisions what do you do what is your approach and the basic answer to that question that i want to leave you with is we need to change our focus we need to go from thinking about the future and trying to predict the future to actually trying to understand the present as quickly as possible so the idea is basically to understand the present to understand the now look for opportunities and try to exploit those opportunities as fast as you can because you don't know even how long those opportunities will persist right once you embrace that idea then that leads you to a very different place in terms of how you think about how an organization should operate and ultimately how technology can help that process and we're going to start to unpack some of that now so to think about some of these ideas the first thing i want to do to talk about kind of organizational learning is i want to borrow an idea from fighter combat from the theory of fighter combat this is a an idea that emerged a number of decades ago there was a famous fighter pilot in the u.s air force by the name of john boyd and he was thinking about dog fighting and trying to understand how to analyze dog fighting and one of the key insights that he had was he said okay if you've got two fighter aircraft and you're trying to understand which one will dominate if you just focus on the performance characteristics of those machines solely you will be misled the idea that he advanced is he said look there's a there's a human in that driving that system and combat in this situation is two human machine systems competing against one another right and precisely it's what that human chooses to do and how quickly they can learn about their environment make decisions and act that has that's a key determinant in terms of who will be victorious so the way that he described this is he talked about four different states that that the human is going through they're observing they're orienting they're deciding and they're acting so you'll often hear this described as what's called the ooda loop boyd himself actually didn't describe it as a as a simple loop but just for simplicity's sake we often talk about this as a as a singular loop and moving through those different states and his assertion was the the the uh in fighter combat the the pilot that can move through that cycle faster will win they will dominate the other opponent if they're slow in in trying to understand what's going on and taking actions and being decisive they will always be behind power curve they will always be trying to understand and failing to understand what their opponent is doing this idea really generalizes and can be a useful metaphor for us to think about how organizations learn more broadly so taking that idea to heart and now asking the question okay well what organizations today actually do this are there any good examples well first and foremost the companies that we're all very familiar with facebook linkedin amazon these very data-driven companies that are focused exclusively online microsoft does the same thing as well right the key innovation that they've had is they're leveraging their ability to conduct high volumes of low-cost experiments all of the time to understand their environments very quickly again to identify opportunities and to exploit them right that's a really key innovation the real question is is okay well how do we generalize from that a lot of people are excited about big data and data science precisely because of what these companies have accomplished and they're trying to understand kind of what is that generalization and this is where a lot of the fantastical claims come from right is people think that the the model that we see here can then generalize to other domains so whether we're taking on bigger issues like poverty or you know other challenging social problems for example there's a in some cases a naive mapping where they say oh yeah we can leverage these same ideas and go after that well there are critical differences there and we'll talk about some of those challenges tonight but what i want to do before we get to that is actually i want to share with you another example of an organization that is not focused exclusively online that surely their their example surely fits in with what i'm describing and it's the spanish fashion company zara what is interesting about zara is they do not follow kind of the typical pattern in fashion which is uh the following which is basically as you're thinking about um essentially future seasons and i have a friend who's in fine in in fashion who does this where you know if it's the springtime and she's already thinking about the fall she's making predictions about what people are going to want to wear in in the fall and she says look it's it's a guess i have really no idea what people want right zara owns up to that and they said look yeah we have no idea what's going to be popular absolutely no idea so what we're going to do instead is do the following we're going to look in various markets we're going to see what people are wearing right now we're going to take those as base level ideas we're going to riff on those and make variations of those ideas make small batches of garments put them in those markets and see how well they do so they're conducting their own small scale experiments to test these hypotheses and what makes that possible is they have basically dramatically reduced the time to get go from a initial design to garments in store so from a brand new design to the time that they put garments in store they can do that in four to five weeks if they're just riffing on an existing design it's two weeks so again they've they've dramatically tightened that learning loop and they're exploiting that to their to their advantage to do something that that no one else can do because again they've got so much more delay in their system so the the the old way that we're you know i'm basically arguing we need to move away from is this idea of planning as prediction right so not only in fashion but in so many organizations today we go through things like strategic planning exercises where we make guesses about what sorts of alternative futures will we think will unfold right and then we orient our strategy around those predictions now the problem is is that when you're wrong you then have to adapt to that reality that new reality that you didn't anticipate and so often these same organizations that are making that are making these predictions they're very slow to adapt right so the cost of the prediction error plus the delay gets them into really hot water and what i'm again advocating for is to say look you know organizations need to shift from planning as prediction to planning as knowledge aggregation so again with this shift from the future to the present the goal is to say okay let's try to bring together as much knowledge about the present state of the world as quickly as possible try to understand again where those opportunities are and then to go after them so now what i want to do kind of and the rest of the talk is is talk about that process of knowledge aggregation you know in contrast to the online example so again like facebook microsoft linkedin and all those where they're for them knowledge aggregation is about doing these high throughput low-cost experiments it's a completely different game when we're starting to talk about arenas that are completely offline and we we now have organizations with people where knowledge is distributed people are being exposed to different things in local environments and we need to think about how do we do that how do we construct that essentially shared vision of of what conditions are on the ground and then coordinate our actions around that the first example that i want to talk through is one that that i experienced basically in around the kind of the mid 2000s time train so around 2004 2005 the number one killer of u.s soldiers in iraq was the improvised explosive device basically we had insurgent groups that were taking you know whatever they could get their hands on so whether it would be just ordinance that was lying around you know cell phones any kind of electronics and they would be cobbling together weapons and putting those by roadsides burying them what have you and they basically the insurgency was being run like a startup right they are basically constant they were constantly innovating constantly trying different designs doing experiments seeing how the us military reacted to that and changing their tactics they got inside our ooda loop and they were dominating us and we were spending billions of dollars trying to catch up and failing so basically we have this following problem where here we are trying to learn and trying to get back get back ahead of this problem and if you think about the us military as you know being kind of spread across this entire country there were different elements the us military they were being exposed to different insurgent groups some of those were communicating and so and they were sharing lessons learned on our side we had certain parts of the military that were you know unbeknownst to them were being exposed to the same threat some of them were innovating faster than others and we had this problem of we've got if there is a success in a certain part of the military in a certain region we've got knowledge there that we need to share right and the question is like how do we get that to other people who can can use that and and really take advantage of it if you think about like how do we do that how do we share knowledge in an enterprise right well still today in 2014 it's pretty old school right so sure we have question answering forums and you know we've made some steps forward but in many cases it's totally old school if you have a question and you don't know you don't know the answer yourself often you'll go to someone you trust and you will ask them that question and say hey do you know how i do x right and if they don't know the answer maybe they know someone who knows the answer and so on and so forth the problem with that obviously is that kind of limits you to the knowledge that's available in your local network right and as far as the the trust will get you right how do you make big jumps in that network right when in particular in situations when you need that now omar actually set me up for this right because he actually started talking about some of these these routing issues and the interesting thing is that my co-founder actually did some really uh kind of trailblazing work in the mid 2000s on a system called ilink and what they did was they actually basically added technology to existing online communities for the us army namely company command and platoon leader so company commanders are captains in the army uh platoon leaders again are first lieutenants and the guys that started company command made this observation they said you know when we get back from um you know doing training or other exercises the oftentimes what they do is they'll grab a case of beer they'll go sit on a porch drink some beer and they'll talk and they share knowledge and these guys said you know what there are company commanders all around the world that are doing this every single day we need to actually make an online space to make that a kind of a broader reality where we can do greater knowledge sharing now my my co-founder dave uh he was at the stanford research institute and they built this system called ilink that basically was a machine learning system that watched question answering behavior and what it did is over time it learned how to map basically questions to experts when they installed this in these communities the the army soldiers were were amazed they said you know this has breathed new life into these communities and it's teasing out pockets of expertise that we just basically didn't know existed and what's sad is this kind of technology is still not out there right and there's so many ways in which we can expand beyond that the final example that i want to walk through that again talks about the the power and and the need for these capabilities to leverage what we call tacit knowledge which is knowledge that's in people's minds and it's very difficult to actually codify that right such that these kinds of systems are really useful where we can go and find the right person to ask and get them to engage and then the environments where this can actually be really powerful right is in disaster recovery and like post-war reconstruction type events disaster recovery in particular where you've got a real timeliness to the to this problem right and this uh in particular i wanted to talk through um a situation around haiti and actually i thought that was the next slide so i'm gonna step back for a moment and talk more broadly just about this challenge of of these strategic collaboration problems or so-called wicked problems and just as a side note this idea of a wicked problem is actually one that came out of berkeley here in 1973. there was a notable design theorist here by the name of horse ratel that wrote a a really powerful paper about the idea of wicked problems and being these what he was focused on with social policy planning problems but basically he makes a a very cogent argument for why scientific approaches to those types of problems will ultimately fail or will certainly be rendered with a lot of friction in a lot of different ways so in these kinds of problems you know as i mentioned kind of post-war reconstruction disaster recovery typically you'll have a situation like this where you'll have a number of different parties that come to the event um they all have local objectives that we would all agree are good that they're trying to optimize for but the problem is is that when each of them engages and tries to do their local thing oftentimes they can create a worse situation on the on the global picture right so trying to not make the situation worse is usually the bar at which things are set i i spoke with a marine one star general who who talked about this and he said usually we we talk about the disaster after the disaster which is when all these parties show up they try to do their thing and it just makes the situation worse so back to haiti in particular the earthquake so one particular open source software initiative and related response event that got a lot of attention was this project called ushihidi and ushihidi actually started um in africa uh when there was a kind of post-election violence there a number of years ago and the the innovation that that happened there was uh they were actually once the the government actually shut down the media uh some local developers said you know what we need a mechanism by which to aggregate events of violence that we're seeing happen in the country and they hacked together a simple mapping solution but they also coupled that with using text messaging to collect reports from all around the country and this platform has been used in a number of different events in haiti in particular this got a lot of a press they basically set up a short code in the country and the haitians themselves the victims that you're trying to support became this massive source of information in terms of you know areas that needed attention this part of the story is is well understood and people have talked about but there's another part of this story that hasn't really gotten any major attention which in some sense is even more critical so what was happening when they set up that short code and they started getting all these text messages coming in there's one problem all of them were in haitian creole all the responders for the most part spoke english and didn't understand haitian creel so we were getting all of this great information but it wasn't in a form that could be acted upon there's a gentleman here in the bay area by the name of rob monroe who actually led the effort i'm going to describe to you which is called project 4636 and basically what he did is he said okay you know we need to try to crowdsource this problem and what he managed to do through his own network through basically high trust ties that he had in the development community is he reached out to the haitian diaspora and basically put out a call for volunteers and folks showed up by the thousands and then they re quickly reconfigured a crowdsourcing platform to start feeding out these text messages and getting them translated now it wasn't just a simple exercise of translating the the text messages the the the tacit knowledge that was being taken advantage of here was the following which is you might have a message that says okay that you know this hospital in this neighborhood is running out of supplies right there's no gps coordinates in that message nothing that allows them to immediately market on the map so that the folks that were actually doing the translation were also leveraging their local knowledge to actually translate those descriptions into coordinates and also in a corresponding message that could be put on the map in contrast to the public story about ushahidi being so dominant and they weren't doing really the bulk of the kind of the geolocation and the transmission of these messages project 4636 was doing this and basically those messages were a critical feed to the us military that was engaging and for example the us marine corps came back and said you know please don't stop you know you guys are producing valuable information for us that is helping us save lives from a technology perspective it was actually again quite simple like they were using essentially something equivalent to irc chat to basically have discussions so when when particularly challenging text messages would come in that maybe somebody didn't necessarily know the answer they would jump on this forum and start talking and there were a lot of things that were going on here not only just translation activities but there was social support activities going on as well it turned out that folks who were so dedicated to this and just doing this over and over again for hours and hours and hours they got so emotionally involved and so emotionally attached that they were starting to over time show signs essentially of post-traumatic stress right so this again was a really powerful but yet basic application of technology but again it gave them enough power to to leverage the collective wisdom of that crowd the collective knowledge in terms of the local terrain where people were and and were able to do amazing things just as a final slide to step back now and think about like in these domains what where do we need to go with technology i still feel like in many respects we're really at the infancy of what we've seen and you know in these so-called wicked problems we have this situation where as the problem complexity grows and it really gets to a point where no one person can really understand the the the expense you know the expanse of the problem right ritel himself was talking about this notion that we need to basically build technology that supports groups of people to try to reason about what's what's taking place in a given environment and to actually construct an approach to dealing with that one of the things again that he was talking about decades ago that was fascinating is he talked about technology as a reinforcer of natural intelligence so artificial intelligence at the time was starting to come onto the scene and he was very critical of it he said why are we trying to make machines smarter right we need to build technology that helps us do our jobs and reason about that complexity better and this is something that you know in recent years is people are talking about that but it goes under the label of augmented intelligence but it's so fascinating to see him discuss that at such you know so many years ago so i'm going to stop there i could go on for quite a while there's a lot of different stories and things that i think are interesting and compelling but i just wanted to get again give you a flavor of again what i think we're faced with as organizations and the implications in terms of you know once we fully embrace the the complexities that we're dealing with and the uncertainty and that i think it takes us to a different place in terms of how we reframe how we we operate and there's so many organizations that have yet to really fully embrace this because it leads to some challenging decisions in terms of how you structure organizations to to move through that learning loop more effectively and so with that i'm happy to take questions thank you questions yes where did you work for uh who did you work for when you were working uh so i actually started in the intelligence community i worked for an agency called the defense intelligence agency uh decided that being in government was was not the best answer for me then i went to the applied physics lab at johns hopkins which is a defense lab uh i was there for about eight and a half years and then i went to lawrence livermore national lab uh for another year and a half so on the whole i you know i actually started in the intelligence community at the age of 18 basically got a full scholarship from them and spent summers going and working in different parts of the community and and then continued to support that community after graduating from graduate school so any other questions yes i'm sorry can you expand more on the lowering points and what if what what what simplification is next so what's really interesting and i i know you'll appreciate this right sorry oh okay so the question was uh could could i expand more on just on the routing so one of the things i'm i'm sure you'll appreciate i think you even made reference to in your talk right is uh well we're trying to even just design systems where we're building in these interactive feedback loops one of the things that we don't spend enough time thinking about is essentially decisions that we make in designing the user experience to make that more palatable and to support that and you know one of the interesting lessons learned that that i thought was interesting was asking dave when i asked dave well how did how did you route the questions did you go straight to the experts and he said actually no we didn't do that because that turned out to be not very effective because if the machine comes and all of a sudden goes hey you know you ask this question oftentimes it gets ignored so what they did was the following is they actually would route if they they had a target in mind of who they thought was the expert they'd route the questions to a high trust person that was connected to the expert and ask them to route it in right so it was an interesting kind of human machine you know it's like cooperation to get get the job done and once they did that they saw the the acceptance rate go quite high the other thing that was very interesting he said there were there were cultural subtleties about how they phrased things in the community so they needed to really understand the nature of the the way the soldiers would communicate in there what was important to them and and building all of that cultural context into the design was super key any other questions all right thank you oh