Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. =Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. = # Digital Chokepoints Authors: Michael Porcellacchia, Christoph Trebesch, Benjamin Wache Discussant: Chenzi Xu Video: https://www.youtube.com/watch?v=U_ssYRB6uPs&t=16411s ## Talk (04:33:31 – 04:59:13) [04:33:32] we're gonna get started with the afternoon here [04:33:34] Michael are you ready to go [04:33:40] alright we are very excited [04:33:42] to kick off the afternoon [04:33:44] with digital choke points Michael you have 30 minutes [04:33:48] hello oh is this on [04:33:50] right hi so thank you [04:33:52] very much for having us in the program [04:33:54] it's a pleasure to be here of course [04:33:56] so today I'm presenting on going [04:33:58] an ongoing project with Christoph and Benjamin [04:34:00] who are here in the room with us [04:34:02] and the project is called digital choke points [04:34:04] so [04:34:06] the project is motivated by [04:34:08] the criticality [04:34:10] of data flows to the modern digital [04:34:12] economy indeed it's hard to overstate [04:34:14] how much things like [04:34:16] AI logistics finance [04:34:18] depend on the speedy and uninterrupted [04:34:20] flow of data [04:34:22] and of course [04:34:24] there is a large literature showing that [04:34:26] countries at all stages of development [04:34:28] do in fact get high welfare gains [04:34:30] from having high speedy [04:34:32] connectivity and large data flows [04:34:34] this project though [04:34:36] is focusing on the physical [04:34:38] infrastructure that carries this data [04:34:40] so I'm talking about the network of cables and satellites [04:34:42] that makes these flow of data [04:34:44] possible across borders [04:34:46] and in particular [04:34:48] this is [04:34:50] surprisingly concentrated [04:34:52] kind of infrastructure both geographically [04:34:54] and in terms of ownership [04:34:56] and concentration of course is [04:34:58] often associated with risks [04:35:00] and this is something that policymakers in Europe [04:35:02] and Asia worry a lot about [04:35:04] especially thinking about the cable cuts [04:35:06] that we see in the Baltic or in [04:35:08] the Thailand Strait [04:35:10] and also in general they worry about how [04:35:12] much concentration in ownership and control [04:35:14] can give the possibility to countries [04:35:16] to weaponize this network [04:35:18] to course them [04:35:20] this paper is [04:35:22] trying exactly to study this [04:35:24] we are quantifying the dependency [04:35:26] and vulnerabilities that are associated [04:35:28] with this global telecom network [04:35:30] in other words we're studying how much [04:35:32] countries and in general the global [04:35:34] economy would be affected [04:35:36] by disruptions in this [04:35:38] critical set of infrastructure [04:35:40] so let me [04:35:44] convince you that indeed this [04:35:46] is a vulnerable [04:35:48] network so [04:35:50] why is it so vulnerable [04:35:52] well first of all for three reasons [04:35:54] because [04:35:56] these cables are very easy to damage [04:35:58] and very hard to defend [04:36:00] so we [04:36:02] a ship dropping an anchor can accidentally [04:36:04] sever a cable [04:36:06] and of course monitoring 300 million kilometers [04:36:08] squared of a seabed [04:36:10] is very very hard to us [04:36:12] second [04:36:14] infrastructure as I said is very rich [04:36:16] there are cables and satellites but really [04:36:18] it's about cables because [04:36:20] capacity means that satellites are no [04:36:22] substitute for cables indeed [04:36:24] all of Starlink so 7000 [04:36:26] satellites from Starlink have the same capacity [04:36:28] as a single submarine cable [04:36:30] built in recent years [04:36:32] and this not only [04:36:34] this is not accounting for the fact that [04:36:36] satellites themselves can be jumped [04:36:38] and tampered with [04:36:40] and as I was saying there are only [04:36:42] 10 cables in the seabed carrying [04:36:44] international data flows [04:36:46] but the geographic [04:36:48] concentration is even more sharp [04:36:50] when you look at single countries [04:36:52] just keep in mind that South Korea for example [04:36:54] is connected by just 10 cables [04:36:56] to the rest of the global internet [04:36:58] Taiwan 14 and Japan 28 [04:37:00] so these are very [04:37:02] very concentrated kind of [04:37:04] infrastructure so these [04:37:06] three facts make cables [04:37:08] a clear and prototypical [04:37:10] choke point of the global [04:37:12] modern economy and [04:37:14] this is the first paper to study [04:37:16] them in this light [04:37:18] specifically we make [04:37:20] four contributions [04:37:22] first of all we provide a public good [04:37:24] we provide a new database [04:37:26] covering 175 years [04:37:28] of telecom infrastructure [04:37:30] so we start from the very first telegraph line [04:37:32] built from France to England [04:37:34] under the channel and we collected [04:37:36] all sorts of cables and satellites and radio towers [04:37:38] that were built since then [04:37:40] across oceans [04:37:42] and in the sky [04:37:44] so this is a dataset called Globetel [04:37:46] will be of course made public [04:37:48] well soon [04:37:50] and then [04:37:52] we use this dataset to [04:37:54] answer the questions I was mentioning before [04:37:56] so first of all we estimate [04:37:58] we propose a new model [04:38:00] and estimate it [04:38:02] to try to link this physical [04:38:04] physicality of this network to the welfare of countries [04:38:06] specifically we [04:38:08] I will show you the model very briefly but it will be [04:38:10] a traffic assignment model [04:38:12] a county for cables and satellites and data flows in them [04:38:14] that will link [04:38:16] how the topography of the network [04:38:18] meaning its disruption or new buildings [04:38:20] is going to affect the welfare of countries [04:38:22] then we use this model [04:38:24] to do counter-functional analysis [04:38:26] and to quantify the risks associated with [04:38:28] potential warlike scenarios [04:38:30] I will show you today a potential [04:38:32] conflict in near Taiwan [04:38:34] or a more broad one in the south China sea [04:38:36] what consequences that would have [04:38:38] for the global internet [04:38:40] and finally a fourth contribution that unfortunately [04:38:42] I won't have much time to talk about today [04:38:44] is we study responses to these risks [04:38:46] in terms of cable building [04:38:48] so that is where history really comes into play [04:38:50] in this project we show that [04:38:52] historically whenever there were increases in risks [04:38:54] conflict wise [04:38:56] or geopolitical risk more specifically [04:38:58] countries and especially great power would engage in cable building [04:39:00] to [04:39:02] self-insure against cable cutting [04:39:04] and to become less dependent [04:39:06] on the current hegemon [04:39:08] but as I said today [04:39:10] unfortunately I will just focus on [04:39:12] 1, 2, 3 for time [04:39:14] ok [04:39:16] so let me start [04:39:18] by briefly introducing you to [04:39:20] this new database [04:39:22] this is a bit of a pity because this was [04:39:24] four years of work and it's going to be two slides [04:39:26] but [04:39:28] so this is the global data set [04:39:30] and we have a very comprehensive new infrastructure [04:39:32] we put together sources [04:39:34] from archival sources to modern sources that are on the internet [04:39:36] we [04:39:38] I think we really did a good job at [04:39:40] merging all these sources together [04:39:42] and getting a truly dynamic picture [04:39:44] of how this network evolved over time [04:39:46] there are [04:39:48] available [04:39:50] there are websites that propose [04:39:52] a snapshot in time of how [04:39:54] the network looks like today [04:39:56] we really focus on getting the dynamics right [04:39:58] and [04:40:00] well from 1850 [04:40:02] so [04:40:04] this data set is not only about the geography [04:40:06] of these cables it's very broad [04:40:08] coverage so we hope it will be interesting [04:40:10] for many people because it covers [04:40:12] not only the geography but the ownership, the dynamics of ownership [04:40:14] and also the capacity [04:40:16] and the technological feature of this cable [04:40:18] which as you might expect [04:40:20] in two under the almost years of history [04:40:22] have developed quite a bit [04:40:24] as an addendum to this data set we also [04:40:26] propose a history [04:40:28] an archive let's call it of [04:40:30] digital coercion events [04:40:32] from cable cutting events to sabotage to embargoes [04:40:34] okay here is [04:40:36] the mandatory picture of how [04:40:38] the original sources look like [04:40:40] and this is what [04:40:42] and this is what we did with it [04:40:44] okay so here you can see [04:40:46] the reconstructed network for a specific year in time [04:40:48] 2022 [04:40:50] it looks like a mess because it is [04:40:52] so the black lines are [04:40:54] the red dots are [04:40:56] the red dots are [04:40:58] the small ones are starlink [04:41:00] and the other the bigger ones [04:41:02] are other types of satellites [04:41:04] that have a different kind of orbital [04:41:06] properties compared to starlink [04:41:08] but anyway this is how the physical [04:41:10] network looks like [04:41:12] in 2022 and we can do this picture [04:41:14] since 1850 with this data [04:41:16] alright [04:41:18] so now I told you that we can [04:41:20] measure and observe this network [04:41:22] now how do we link it to welfare [04:41:24] and how do we quantify the costs [04:41:26] related to disruption in this network [04:41:28] for that we build a model [04:41:30] of data flows on a physical network [04:41:32] so this model is fairly simple [04:41:34] and I will try to present it in one slide only [04:41:36] the idea that we borrow [04:41:38] from the knowledge pilover [04:41:40] knowledge externality literature is that there is a factor of production [04:41:42] that we can call you can think of it as [04:41:44] information or knowledge [04:41:46] that increases the productivity of firms [04:41:48] the difference we have here [04:41:50] is that this factor of production is not [04:41:52] just something that arrives to you [04:41:54] you have to source it and the way you do it [04:41:56] is that you connect to the source of this knowledge [04:41:58] via data flow [04:42:00] so you import data [04:42:02] from the source of knowledge and that data [04:42:04] is what [04:42:06] these X and J is going to be [04:42:08] composed by a CES aggregator [04:42:10] into this measure of productivity [04:42:12] that affects firms so that's the link between [04:42:14] data flows and productivity [04:42:16] but as I said [04:42:18] these data flows are not [04:42:20] just have to go through [04:42:22] this physical network and for that we simply [04:42:24] assume that [04:42:26] these flows have two costs [04:42:28] one you have to pay the price [04:42:30] to a telecom carrier to [04:42:32] actually establish the connection between London [04:42:34] and Sao Paulo and second [04:42:36] there is a shadow cost [04:42:38] let's say which is that [04:42:40] the slower is the connection [04:42:42] the longer you have to wait [04:42:44] for a response so the less efficient it will be [04:42:46] so these are the two costs associated [04:42:48] with importing data flows [04:42:50] the cost you have to pay directly [04:42:52] and the workers cost of [04:42:54] waiting for the data to arrive [04:42:56] which is what in this literature is called latency [04:43:00] so with this idea we both of these [04:43:02] features are linked to the distance [04:43:04] across the network between the source of knowledge [04:43:06] and the position of the firm [04:43:08] and therefore we can derive a simple gravity demand [04:43:10] for data flows based on this [04:43:12] distance or latency [04:43:14] and given the importance of congestion [04:43:16] in this network we cannot just do [04:43:18] shortest path assignment but we have to [04:43:20] employ a model of [04:43:22] that takes into account of congestion [04:43:24] think about the last time you were in a concert [04:43:26] or in a soccer stadium [04:43:28] your phone wouldn't work that's congestion [04:43:30] right there for you so the idea is that [04:43:34] the cost of [04:43:36] data flows for example the time it takes [04:43:38] for the data to reach you from Sao Paulo [04:43:40] to London will depend on how busy [04:43:42] it is the route to compare to how capable [04:43:44] the cables are on that route [04:43:46] okay so with these features [04:43:48] we characterize the full model [04:43:50] where we just have an equilibrium where [04:43:52] demand for data flows [04:43:54] the routing on the network and therefore [04:43:56] latencies will all be jointly determined [04:44:00] and this finally gives us a link [04:44:02] that I promised between the welfare of [04:44:04] countries and the characteristic of the network [04:44:06] because the characteristic of the network give us latency [04:44:08] and costs fantastic so this was [04:44:10] the model now we [04:44:12] take the model to the data and we [04:44:14] estimate some of its key parameters [04:44:16] so first of all we have to say what do we [04:44:18] need for welfare or counterfactuals [04:44:20] well as you all know we need bilateral demand flows [04:44:22] which you can observe we have data on [04:44:24] that and we need some parameters [04:44:26] on demand function and on congestion [04:44:28] now I'm not going to bother you [04:44:30] with how we estimate the elasticity of [04:44:32] production or the elasticity of substitution [04:44:34] we use standard IO and trade methods [04:44:36] but the more interesting part I think [04:44:38] was to estimate the congestion effects [04:44:40] for that we simply use [04:44:42] so what are the congestion effects [04:44:44] is the relation between how much [04:44:46] when capacity fills up latency [04:44:48] goes up so how much cost increase [04:44:50] when capacity increases when capacity [04:44:52] is filled up utilization increases [04:44:54] and here we use [04:44:56] intraday within path variation [04:44:58] because what you can see what you can [04:45:00] observe is that as [04:45:02] counter is during the day and night cycle [04:45:04] counter will use more or less [04:45:06] data flows but capacity is fixed [04:45:08] so what you can observe is that [04:45:10] these rhythm of the day night [04:45:12] in capacity utilization [04:45:14] reflects the rhythm in latency [04:45:16] that you observe across the same path [04:45:18] and we use that to [04:45:20] estimate these congestion parameters [04:45:22] right so armed with all of this [04:45:24] we also can validate the model [04:45:26] I'm not going to bother you with that but the model [04:45:28] predicts for the few cases and years [04:45:30] in which we can observe routes [04:45:32] actually taken by data flows [04:45:34] so it's fairly good at predicting [04:45:36] where these routes are going compared to other models [04:45:38] of like geography where random walks [04:45:40] and things like that [04:45:42] so yeah we end not only that but the model [04:45:44] is fairly solidly [04:45:46] reproducing [04:45:48] what we observe in the data [04:45:50] okay now let me tell you about the findings [04:45:52] now that I told you about the data [04:45:54] and the model so the finding number one [04:45:56] is that global data flows are [04:45:58] very very highly concentrated [04:46:00] specifically in maritime [04:46:02] and international maritime choke points [04:46:04] so here you see [04:46:06] just the model prediction [04:46:08] where traffic goes [04:46:10] overlaid on the actual cables [04:46:12] you can see that there are some cables [04:46:14] across the Atlantic and the Pacific [04:46:16] as you would expect that capture lots of these flows [04:46:18] because there is lots of demand between the US [04:46:20] and Europe [04:46:22] and this is very [04:46:24] very concentrated as I was saying at the beginning [04:46:26] 10 cables only [04:46:28] carry 25% of all global [04:46:30] traffic [04:46:32] another way to quantify this concentration [04:46:34] is to look at this picture [04:46:36] which shows instead rather than the single piece [04:46:38] of infrastructure it looks at [04:46:40] in every grid cell of [04:46:42] 2x2 latitude longitude degrees [04:46:44] which is the size of [04:46:46] Maryland or Sicily [04:46:48] if you prefer [04:46:50] it captures how much data flows on that grid cell [04:46:52] so this captures a more geographic [04:46:54] kind of measure of concentration [04:46:56] and you can see here there are certain areas [04:46:58] that light up especially in the English channel [04:47:00] you can see the Suez [04:47:02] and the approach to Suez and the Malacca [04:47:04] rates are dark red in fact [04:47:06] 20% of world data [04:47:08] flows in the English channel only [04:47:10] okay so of course [04:47:12] these are just traffic numbers [04:47:14] these are the model prediction of where traffic goes [04:47:16] it doesn't tell us anything about the cost [04:47:18] of disruptions [04:47:20] for that we need welfare exercises [04:47:22] and we need to allow for rerouting because [04:47:24] if there are 2 cables very close by [04:47:26] we can see that traffic shift it around [04:47:28] and go around it [04:47:30] and in fact this intuition [04:47:32] gets some traction if we look at the picture on the left [04:47:34] on the picture of the left the exercise we do [04:47:36] is we remove a single cable [04:47:38] and we see the effect on global welfare [04:47:40] and you can see a fully [04:47:42] wide picture and this is very intuitive [04:47:44] with what computer scientists say [04:47:46] which is that the network is actually [04:47:48] fairly robust to accidental cable faults [04:47:50] all of the time like there are 200 cable faults [04:47:52] a year [04:47:54] and the network just [04:47:56] routes around them and nobody notices [04:47:58] however [04:48:00] if we look instead [04:48:02] at the choke point kind of view [04:48:04] the geographic grid kind of view [04:48:06] we see of course that these impacts [04:48:08] can be very large [04:48:10] and especially if we look at the [04:48:12] typical maritime choke points that you might be thinking about [04:48:14] okay now [04:48:19] I show you [04:48:21] the two views on geographic [04:48:23] concentration and their impact on welfare [04:48:25] and let me go to more [04:48:27] let's say doomsday kind of scenarios [04:48:29] let's look at warlike scenarios [04:48:31] and before I just make assumption [04:48:33] what a warlike scenario looks like [04:48:35] I want to show you some historical examples [04:48:37] so this is Cuba [04:48:39] these are cables [04:48:41] drawn according to [04:48:43] the historical part of our data set [04:48:45] and you can see that Cuba is connected to [04:48:47] Spain with the cables going all the way [04:48:49] up north and also to neighboring [04:48:51] countries this is Cuba in 1898 [04:48:53] now [04:48:55] this is what happened [04:48:57] what the US did [04:48:59] at the breakout [04:49:01] of the Spanish American war [04:49:03] the objective of the US was to isolate Cuba [04:49:05] and the Spanish garisons that were located in Cuba [04:49:07] from receiving orders from the motherland [04:49:09] and they did that by just severing all cables [04:49:11] that could allow Cuba to communicate [04:49:13] directly or indirectly with the motherland [04:49:15] yes [04:49:17] a more dramatic episode [04:49:19] of cable wars [04:49:21] is World War I [04:49:23] and World War II is not very different [04:49:25] but here you see [04:49:27] an episode of all out cable [04:49:29] telecom wars between [04:49:31] Britain and Germany [04:49:33] there were more than 61 cable cuts [04:49:35] during World War I [04:49:37] and World War II is the same magnitude [04:49:39] and here you can see that they mostly focused [04:49:41] on the English channel [04:49:43] and the Baltic as they try to [04:49:45] isolate each other from the rest of the world [04:49:47] and force each other to use their own technology [04:49:49] so that they could monitor and surveil [04:49:51] this war was won [04:49:53] the cable war was won by Britain [04:49:55] and as you know this played a crucial role [04:49:57] in both world wars actually [04:49:59] as it allowed Britain to [04:50:01] to intercept communications [04:50:03] as in the famous Zimmermann telegram episodes [04:50:05] that you might be aware of [04:50:07] okay [04:50:09] so we have seen [04:50:11] what could happen in history [04:50:13] and this is the ship that made it possible [04:50:15] this is the British C.S. Alert ship [04:50:17] it was the first to cut the five German cables in one go [04:50:19] and of course [04:50:21] technology has only improved since then [04:50:23] and here you can see [04:50:25] media rumors of [04:50:27] what cable cutting capable drone [04:50:29] submarine drone [04:50:31] would look like this Chinese [04:50:33] prototype [04:50:35] okay [04:50:37] so inspired by these [04:50:39] green episodes in history [04:50:41] we do two scenarios one is a full digital [04:50:43] blockade of Taiwan in the spirit of what we have seen [04:50:45] happening to Cuba and you can see here [04:50:47] we are cutting I think [04:50:49] the 14 segments that Taiwan [04:50:51] that lands in Taiwan you can see [04:50:53] that they are highlighted in black in the picture on the left [04:50:55] these two pictures show welfare effects [04:50:57] on the left and internet speed on the right [04:50:59] so you can see the welfare effects is barely [04:51:01] touching anyone but Taiwan [04:51:03] Taiwan is completely isolated from cables [04:51:05] so they have to rely on Starlink [04:51:07] which by the way here we allow use of Starlink [04:51:09] but in the real world Starlink is [04:51:11] negotiating with Taiwan [04:51:13] and it is not done yet [04:51:15] for granting access it is not done yet [04:51:17] anyway here we see that Taiwan would be very severely affected [04:51:19] by just relying on satellites [04:51:21] because they have to shrink their demand [04:51:23] to squeeze it into the Starlink's capacities [04:51:25] and you can see below [04:51:27] a comment that [04:51:29] without satellites actually the welfare losses would be [04:51:31] double even if satellites only carry [04:51:33] 7% of [04:51:35] Taiwan baseline traffic [04:51:37] they do absorb quite a bit of the shock [04:51:39] and [04:51:41] so a more interesting [04:51:43] more global exercise [04:51:45] is to think instead of a more [04:51:47] one-like scenario [04:51:49] of cable cutting in the South China Sea [04:51:51] here again you see on the left welfare effects [04:51:53] on the right internet speed [04:51:55] you can see welfare effects being very high [04:51:57] in many countries [04:51:59] you can see Myanmar and India [04:52:01] and the Philippines having effects between 10 [04:52:03] and 15% of their welfare [04:52:05] here many things are at play [04:52:07] including the problem of [04:52:09] congestion all of this demand being [04:52:11] displaced and congesting everyone else [04:52:13] and in fact these effects are global [04:52:15] are not only regional [04:52:17] we can show that global welfare shrinks by 3.4% [04:52:19] the US is around 2% [04:52:21] for an episode of this magnitude [04:52:23] and speed which is maybe more relatable [04:52:25] has a kind of measure [04:52:27] would fall by one third and in the US [04:52:29] a half [04:52:35] so [04:52:37] finding number three [04:52:39] we talked about geographic [04:52:41] and infrastructure [04:52:43] we talked about world-like scenarios [04:52:45] so let's talk about ownership for a bit [04:52:47] and [04:52:49] another type of concentration I mentioned [04:52:51] at the beginning is that these [04:52:53] infrastructure is controlled by [04:52:55] by very few players mostly from the same [04:52:57] countries [04:52:59] and this is true also historically [04:53:01] you can see here [04:53:03] the map again with our data [04:53:05] going from 1850 to today [04:53:07] where we show the distribution [04:53:09] of ownership [04:53:11] of this infrastructure [04:53:13] where it's scaled by kilometers [04:53:15] length so that a transatlantic cable [04:53:17] counts more than a small domestic one [04:53:19] and you can see the incredible [04:53:21] dominance of British firms [04:53:23] and the government in the telegraph [04:53:25] era [04:53:27] and you can see that in the late 90s [04:53:29] with the fiber optic revolution [04:53:31] and the dotcom bubble the US is taking up [04:53:33] a much bigger role [04:53:35] you see also lots of other emerging [04:53:37] countries and that's because the [04:53:39] kind of market structure changed [04:53:41] now it's rare that one firm owns a whole [04:53:43] cable they typically have consortia [04:53:45] that bring together [04:53:47] local actors [04:53:49] it's a way to obtain access [04:53:51] to landing points [04:53:53] but the idea here is if we just zoom in [04:53:55] and look at great powers it's a [04:53:57] story of displacement where the US [04:53:59] became the dominant player [04:54:01] and this figure is dramatic for satellites [04:54:03] I don't need to show it [04:54:05] but I'm sure that the US is essentially a monopoly [04:54:07] on satellite technology at the moment [04:54:09] and here what we can do [04:54:11] with the paper, with the model [04:54:13] is to do a scenario, an embargo [04:54:15] scenario where we take [04:54:17] here every country, imagine that every [04:54:19] country is targeted by an embargo [04:54:21] or a sanctioned regime where the US [04:54:23] forbids that country from using their cables [04:54:25] so everyone else can use American technology [04:54:27] just that country, color deer [04:54:29] is targeted and you can see the effects [04:54:31] that these embargo would have on [04:54:33] that country's welfare and you can see [04:54:35] as you would expect that Latin America is [04:54:37] quite heavily affected, especially [04:54:39] Central America, given the [04:54:41] in black you can see the American-owned cables [04:54:43] so you can see that the US has a very [04:54:45] major role in connecting Latin America [04:54:47] and you can see in fact that costs [04:54:49] are highly concentrated there and I [04:54:51] think relatively surprisingly we also see [04:54:53] that China is quite dependent on the US [04:54:55] despite 10 years of [04:54:57] investments in this [04:54:59] technology to try to get a more [04:55:01] independent foothold in the world [04:55:03] okay [04:55:05] so [04:55:07] in conclusion [04:55:09] we [04:55:11] we contribute [04:55:13] to the understanding of this problem [04:55:15] in three main ways, so first of all [04:55:17] we provide this public good, this new [04:55:19] dataset that will allow [04:55:21] everyone and well [04:55:23] us to start with to study [04:55:25] the history of telecommunications [04:55:27] from the very first [04:55:29] telegraph international line to telephones [04:55:31] to satellites and to the internet [04:55:33] systems of today [04:55:35] we then [04:55:37] propose and estimate a new model [04:55:39] that takes seriously the idea of traffic [04:55:41] assignment of routing on this physical network [04:55:43] and of the congestion that might emerge [04:55:45] and we use this model to run counter [04:55:47] factuals and we can indeed [04:55:49] use it to do [04:55:51] any kind of scenarios that policymaker [04:55:53] might be interested in and [04:55:55] the main takeaways [04:55:57] are that the network is [04:55:59] very [04:56:01] resilient and [04:56:03] very [04:56:05] uncomfortable for what concerns the [04:56:07] modern period, I would say [04:56:09] is that these cables are [04:56:11] real choke points of the modern global [04:56:13] economy. [04:56:15] They are geographically very [04:56:17] very concentrated and [04:56:19] while the network is indeed [04:56:21] resilient to small cable faults [04:56:23] it is [04:56:25] very important to have [04:56:27] the network [04:56:29] to be able to [04:56:31] use it to [04:56:33] understand the [04:56:35] the US has as a provider of these services. [04:56:37] I actually hope [04:56:39] I still have the time to open [04:56:41] because I said I wouldn't talk about [04:56:43] the response to these kind of risks [04:56:47] but I really would like to show a picture [04:56:49] how much, how long do they still have? [04:56:51] A few minutes, amazing. [04:56:53] All right. [04:56:55] So here we have the [04:56:57] risks of [04:56:59] cable building. [04:57:01] Here what you can see is the old red line [04:57:03] this is a famous infrastructure [04:57:05] investment that Britain did just [04:57:07] before World War I and the objective [04:57:09] here was to make themself immune [04:57:11] or say insured against [04:57:13] potential attacks from the [04:57:15] German submarines or navy [04:57:17] and the logic here is to [04:57:19] build a network that would allow them [04:57:21] to reach the world by only [04:57:23] a few meters. [04:57:25] So it was much more difficult [04:57:27] to cut cables in the middle of the sea. [04:57:29] You needed to reach land and there are [04:57:31] locations where Britain could just put [04:57:33] the army in front and scare ships away. [04:57:35] So this is for example how Britain dealt [04:57:37] with this major risk. [04:57:39] And [04:57:41] oops, okay. [04:57:43] I have the other [04:57:45] story to tell you. [04:57:47] Okay. [04:57:49] And here you see instead how the [04:57:51] British [04:57:53] had to build a network [04:57:55] that was [04:57:57] the most important in the world. [04:57:59] So as I told you at the beginning [04:58:01] Britain was the undisputed monopolist [04:58:03] of cables in the 19th century. [04:58:05] And here we see [04:58:07] when that monopoly started being [04:58:09] questioned. Here is after the [04:58:11] Spanish-American war when it was evident [04:58:13] to everyone that this technology was [04:58:15] a very, very crucial for military [04:58:17] success. [04:58:19] And so in the [04:58:21] early 19th century [04:58:23] when the British [04:58:25] had to build British cables, [04:58:27] you see here on the left [04:58:29] the United States, France, Germany [04:58:31] building most of the new cables [04:58:33] in that period. [04:58:35] And you see on the right where they [04:58:37] built these cables, mostly in corridors [04:58:39] that were dominated by British [04:58:41] investments and British presence. [04:58:43] So we see this as evidence [04:58:45] of, well, these two pieces of [04:58:47] technology that are [04:58:49] showing that this is also sort of [04:58:51] happening already today, where [04:58:53] there is a drive to build the cables [04:58:55] that try to avoid the Malacca [04:58:57] trade and things like that [04:58:59] that is recently ongoing and that's [04:59:01] something very exciting that we [04:59:03] might be adding to a future [04:59:05] iteration of this project. [04:59:07] Okay, so again, thank you very [04:59:09] much for having us [04:59:11] in the program. ## Discussion (04:59:13 – 05:15:10) [04:59:13] And yeah, looking forward to [04:59:18] having you here. [04:59:20] Thank you so much. [04:59:28] Thank you so much for inviting me to [04:59:30] discuss this paper. [04:59:32] I entered my back so I'm just [04:59:34] going to lean here. [04:59:36] So this is [04:59:38] a paper where I learned [04:59:40] a lot. [04:59:42] So I learned that the data [04:59:44] that we all depend on [04:59:46] for all of our work [04:59:48] basically travels through little [04:59:50] garden hoses that go [04:59:52] to the sea and across the sea bed [04:59:54] and that these little garden [04:59:56] hoses, while they form an [04:59:58] incredibly effective and [05:00:00] efficient network, when things are [05:00:02] good, also have characteristics [05:00:04] of being [05:00:06] having very highly concentrated [05:00:08] ownership of going through [05:00:10] some geographically [05:00:12] concentrated areas [05:00:14] and in some [05:00:16] senses being, [05:00:18] sorry, featuring limited [05:00:20] observability that actually [05:00:22] the data that we [05:00:24] depend on is very vulnerable [05:00:26] to what we here would call [05:00:28] a choke point. [05:00:30] So the reason we know that [05:00:32] now is because the paper has this [05:00:34] really incredible new data set that [05:00:36] has literally all the ways [05:00:38] in which continents have been [05:00:40] connected [05:00:42] to transmit data from [05:00:44] 1850. So the telegraph network, [05:00:46] the telephone network, the modern [05:00:48] fiber [05:00:50] the fiber optic cables, as well [05:00:52] as satellites. [05:00:54] And a [05:00:56] fact that comes out of [05:00:58] all of this historical data [05:01:00] is that [05:01:02] specific countries dominate [05:01:04] that global network, both [05:01:06] historically as well as today. [05:01:08] That dominance [05:01:10] has a lot of implications [05:01:12] for countries around the world [05:01:14] as Michael showed. [05:01:16] And the way that they quantify [05:01:18] these implications [05:01:20] are through a very [05:01:22] straightforward model [05:01:24] where you have cables carrying valuable [05:01:26] information, but of course there's [05:01:28] congestion and there's the possibility [05:01:30] of needing to be diverted across different [05:01:32] lines. [05:01:34] And so because [05:01:36] of the way the network is set [05:01:38] up, an individual country [05:01:40] who is not the hegemon [05:01:42] is going to be exposed to [05:01:44] different, essentially sources [05:01:46] of pressure from the hegemon because [05:01:48] of that control over [05:01:50] this network. So one of these [05:01:52] is this notion of coercion [05:01:54] where you could be bullied [05:01:56] because you depend so much on this cable [05:01:58] network. And so you might [05:02:00] be forced off of this network and what [05:02:02] would you do if that were the case? [05:02:04] Another is this notion of surveillance [05:02:06] which becomes incredibly important during [05:02:08] the historical episodes that were discussed, but also [05:02:10] you know you can imagine how that would [05:02:12] also matter today where if you're [05:02:14] transmitting your data over someone else's cables [05:02:16] then that data can be intercepted [05:02:18] and used against you. [05:02:20] So both of those elements [05:02:22] of pressure from the [05:02:24] hegemon who actually owns the cable network [05:02:26] in the presence [05:02:28] of geopolitical tension will [05:02:30] induce [05:02:32] countries to invest in their [05:02:34] own cables and invest [05:02:36] in ways to avoid these [05:02:38] pressure points. However [05:02:40] that also comes with its own [05:02:42] vulnerability which is that [05:02:44] you build new cables but those new cables can be cut [05:02:46] and so the multiple [05:02:48] historical episodes [05:02:50] of sabotage and [05:02:52] cutting or today [05:02:54] I guess investment in sabotage is [05:02:56] so what I'm going to do [05:02:58] in this discussion actually really compliments [05:03:00] what Michael presented because [05:03:02] he focused a lot on helping us to [05:03:04] understand what's going on in this global network [05:03:06] I'm going to focus on the [05:03:08] extent to which decoupling [05:03:10] happens from [05:03:12] given each of these different sources of pressure [05:03:14] and then how in reality [05:03:16] countries would be able to do it and I'm [05:03:18] going to also lean on the historical narrative [05:03:20] to do that. [05:03:22] So these are the pictures [05:03:24] that I had from the paper where we are [05:03:26] actually able to see how [05:03:28] in the British dominant era [05:03:30] you had these sub-C cables [05:03:32] that were all primarily controlled [05:03:34] by the British and today that network [05:03:36] is primarily controlled by the [05:03:38] US. [05:03:40] And throughout [05:03:42] everything that I talk about [05:03:44] but also that Michael talked about [05:03:46] in the background you should keep in mind that [05:03:48] data is traveling around [05:03:50] these different paths where when you divert [05:03:52] a path it both is costly [05:03:54] directly because you have to go through a less [05:03:56] efficient path but also it [05:03:58] imposes this externality that's [05:04:00] the congestion on the rest of the network [05:04:02] and so in the welfare that they showed [05:04:04] you that's all [05:04:06] incorporated. [05:04:08] Okay. [05:04:10] So now I'm going to talk about some measures that were in the paper [05:04:12] but not in the presentation [05:04:14] which is the ways in which [05:04:16] the hegemonic domination [05:04:18] of this cable network makes [05:04:20] countries vulnerable [05:04:22] to the hegemon. [05:04:24] One measure is [05:04:26] this surveillance exposure. [05:04:28] So this is the extent to which [05:04:30] you are using cables [05:04:32] and your information is passing through [05:04:34] those cables and that information can be [05:04:36] intercepted and so it's this [05:04:38] dependence that is about [05:04:40] how much of your data is actually [05:04:42] being transmitted on another country's cables. [05:04:44] Another [05:04:46] is this notion of [05:04:48] coercion where you might [05:04:50] be forced off of the network [05:04:52] completely and therefore you [05:04:54] have to find a different way [05:04:56] to go around it and so it's [05:04:58] essentially a measure of how [05:05:00] equitable are a given country's [05:05:02] cables from the alternative. [05:05:04] What's interesting [05:05:06] here is that even though the two [05:05:08] of these are in many ways very [05:05:10] correlated and especially if you think about it [05:05:12] they feel like they should be almost the same [05:05:14] thing, they're actually very [05:05:16] different because they're all about [05:05:18] one is about [05:05:20] the least cost path and the other is [05:05:22] about substitutability and the degree of [05:05:24] substitutability. So just in a [05:05:26] very simple [05:05:28] to node where country N [05:05:30] wants data from country J [05:05:32] well you can see one element [05:05:34] where there's a substitute it's more costly [05:05:36] but you're still okay because [05:05:38] it's not that hard to divert [05:05:40] whereas in the other version [05:05:42] you would be incredibly [05:05:44] poorly off because [05:05:46] it's actually you know in [05:05:48] for example the Dahai Wan case [05:05:50] what you have as an alternative is something like a satellite that can do very little. [05:05:54] We can see [05:05:56] this globally in terms [05:05:58] of these measures that actually [05:06:00] Michael Christofen [05:06:02] co-author have in the paper itself. [05:06:04] The first on the left here [05:06:06] is basically how much of a [05:06:08] country's data is plausibly [05:06:10] surveyed by the US [05:06:12] which is basically what share of your [05:06:14] data is going through [05:06:16] US cables and [05:06:18] it's a very dark picture because [05:06:20] the US has a lot of cables. [05:06:22] The second picture though [05:06:24] this point about coercion and basically [05:06:26] the degree to which you can leave the US [05:06:28] network that looks pretty different [05:06:30] and I think here you can really [05:06:32] focus on Australia. So Australia [05:06:34] right now has a lot of [05:06:36] its data going through [05:06:38] US cables but were it [05:06:40] forced off of US cables there's actually [05:06:42] quite a lot of richness of alternative [05:06:44] cables in the South Sea area that [05:06:46] it could use and therefore it's [05:06:48] less prone to coercion [05:06:50] than say South and Central America [05:06:52] where geographically they just have [05:06:54] fewer alternatives. [05:06:56] Okay, so then [05:06:58] what does this mean if you are [05:07:00] susceptible to these different sorts of pressure [05:07:02] points? So this is where my presentation [05:07:04] really focused on decoupling which [05:07:06] now really nicely [05:07:08] compliments what they showed. [05:07:10] One of the ways you would [05:07:12] decouple is to [05:07:14] build a new cable route [05:07:16] and that's going to lower [05:07:18] your exposure to coercion because [05:07:20] you're literally on your own cable as [05:07:22] opposed to on someone else's cable [05:07:24] and it also lowers the amount of data [05:07:26] that's going through somebody else's [05:07:28] surveillance. [05:07:30] But once you build it [05:07:32] it's also a physical object [05:07:34] that's vulnerable to being cut [05:07:36] and depending on where you've [05:07:38] built this physical object [05:07:40] it might have very strong [05:07:42] vulnerabilities because you might be [05:07:44] in physical locations [05:07:46] where [05:07:48] if [05:07:50] someone were to sabotage [05:07:52] that physical location they can do [05:07:54] a lot of damage. [05:07:56] So this slide is where I'm [05:07:58] emphasizing [05:08:00] the three ways [05:08:02] in which actually not having [05:08:04] control over the [05:08:06] international cable network is costly [05:08:08] and [05:08:10] this slide then is to summarize [05:08:12] what you would actually have to do [05:08:14] if you wanted to decouple from it, right? [05:08:16] So the first channel [05:08:18] of being [05:08:20] subject to coercion means [05:08:22] you just need to find one [05:08:24] good enough alternative path [05:08:26] but if you're worried [05:08:28] about sabotage then you might [05:08:30] want to actually care about [05:08:32] where these cables [05:08:34] are geographically [05:08:36] and that they're diverse enough [05:08:38] from the current [05:08:40] sensitive spots [05:08:42] and if you really care about [05:08:44] data protection which I think one can [05:08:46] argue today, China, the US [05:08:48] does care a lot about that [05:08:50] then you would want to have [05:08:52] the Hedgeman off of every part [05:08:54] of the route that you need. [05:08:56] Why does this matter? Well, actually [05:08:58] I think the example [05:09:00] of Germany [05:09:02] in the 19th century [05:09:04] captures this dynamic pretty well. [05:09:06] So in the late [05:09:08] 1800s Britain [05:09:10] decides to kind of weaponize [05:09:12] the fact that it has [05:09:14] built the most comprehensive [05:09:16] telegraph network [05:09:18] and first it uses that to tell [05:09:20] France it can't use it [05:09:22] and then it blockades some of usage [05:09:24] during the war war [05:09:26] and as a result of these actions [05:09:28] Germany says [05:09:30] we need to build our own [05:09:32] network because we cannot [05:09:34] as the Reich be dependent on Britain [05:09:36] but what [05:09:38] Germany does is it builds [05:09:40] exactly where [05:09:42] Britain actually already has [05:09:44] the cables and that makes a lot of [05:09:46] sense from an efficiency point of view [05:09:48] like these are the paths that [05:09:50] are the least cost but [05:09:52] most of the building is in these brown [05:09:54] areas where the British already have cables [05:09:56] right. So then what happens [05:09:58] is that [05:10:00] the Germans expand along these red cables [05:10:02] now they have all this new ownership [05:10:04] of these new cables [05:10:06] but during the war they get cut [05:10:08] they rebuilt them [05:10:10] in the interwar years and then during the [05:10:12] second world war they got cut again [05:10:14] every time they were cut [05:10:16] they had to divert [05:10:18] their data traffic to other cable [05:10:20] networks where they were then surveyed [05:10:22] and so [05:10:24] one could argue that a turning point [05:10:26] of world war one is when the US [05:10:28] joined [05:10:30] the conflict and one reason the [05:10:32] US joined the conflict was because it [05:10:34] learned of something called the Zimmermann [05:10:36] telegraph which was when the [05:10:38] Germans messaged the [05:10:40] Mexicans and said hey [05:10:42] if you [05:10:44] if the US joins the war [05:10:46] will help you [05:10:48] take over the southwest of America [05:10:50] and so this [05:10:52] angered America and [05:10:54] was part of the reason they then [05:10:56] joined the war. [05:10:58] Okay. [05:11:00] So what the paper does [05:11:05] actually is it quantifies [05:11:07] the response [05:11:09] the decoupling that [05:11:11] a country can do in response [05:11:13] to exposure to these different [05:11:15] sorts of pressure points. [05:11:17] In particular, if a [05:11:19] country is worried about being coerced [05:11:21] off of [05:11:23] a network then it can build [05:11:25] its own lines [05:11:27] and so the paper actually [05:11:29] estimates and this is really cool [05:11:31] the amount that [05:11:33] a country will build so the [05:11:35] parameters of new cables that it [05:11:37] will build [05:11:39] in response to rises in global [05:11:41] threats. [05:11:43] And so this gamma [05:11:45] parameter that's in the paper [05:11:47] is capturing basically how much [05:11:49] countries are already thinking about [05:11:51] decoupling and how much they're [05:11:53] responding to those pressures. [05:11:55] What I think is really cool though is [05:11:57] that the global network that we [05:11:59] have and the data and the methodology [05:12:01] here allows you also to think [05:12:03] about how much the global network [05:12:05] is being built and how completely [05:12:07] the new cables reroute [05:12:09] away from the potential of being [05:12:11] surveyed. [05:12:13] And so it is actually possible to [05:12:15] see where are the new cables [05:12:17] or where they would be going. [05:12:19] Are they in vulnerable corridors? [05:12:21] How complete is a new network? [05:12:23] Can the Hagemon survey any part [05:12:25] of the network and it's again [05:12:27] with the methodology that they have [05:12:29] in place it's possible to price [05:12:31] and resolve the model and the welfare [05:12:33] implications along each of the new routes. [05:12:35] So actually not only do you [05:12:37] have the sense of what are [05:12:39] the welfare costs of the current [05:12:41] status quo, but also if you [05:12:43] were particularly worried about surveillance [05:12:45] as a particular country [05:12:47] what would it take to get out of that? [05:12:49] Okay. [05:12:51] And I think this is actually [05:12:53] what the world is kind of looking [05:12:55] like it's going to today. [05:12:57] So in 1900 we see from [05:12:59] the data that there's a lot of duplicate [05:13:01] ownership. So when [05:13:03] an area is [05:13:05] contested like the English Channel [05:13:07] countries just built in that same [05:13:09] English Channel, but then they were [05:13:11] vulnerable to attack. [05:13:13] Today countries and firms [05:13:15] are investing in cable networks that are [05:13:17] in completely different areas [05:13:19] precisely to get away [05:13:21] from the potential of [05:13:23] being caught up in for instance [05:13:25] a conflict in the South Sea [05:13:27] or a conflict in the Suez [05:13:29] or in the Red Sea. [05:13:31] And that's a list of projects [05:13:33] that are basically either in works [05:13:35] or are being [05:13:37] proposed where you can [05:13:39] see these are paths that are [05:13:41] longer, they're more expensive, [05:13:43] they're less efficient [05:13:45] and so there's really no reason to do [05:13:47] that if all you care about is coercion [05:13:49] the reason you want to do this is because [05:13:51] you also care about [05:13:53] vulnerability to sabotage, you also care [05:13:55] about [05:13:59] being surveyed. [05:14:01] And I think this is also an area where [05:14:03] because we know so much about the new [05:14:05] projects being proposed it would be cool [05:14:07] to think about combining both [05:14:09] what the model predicts [05:14:11] and what these new projects say. [05:14:13] Okay, so I'm going to [05:14:15] conclude. I think this is a really nice [05:14:17] paper where I learned a lot. [05:14:19] I mean it's an incredible data set [05:14:21] and a really transparent [05:14:23] methodology for just quantifying [05:14:25] everything that's contained in that [05:14:27] data set. And I think what's [05:14:29] really cool is that the paper [05:14:31] has the ability to [05:14:33] separate all these different [05:14:35] sources of network dependencies [05:14:37] and quantify how hard [05:14:39] or easy it would be [05:14:41] to decouple [05:14:43] from them. And you know, I [05:14:45] we always say things have policy implications [05:14:47] but clearly this is where the world [05:14:49] is actually investing a lot of money [05:14:51] and so thinking about it [05:14:53] with all this data and framework [05:14:55] of where the connections [05:14:57] currently are actually does have [05:14:59] really first order policy implications. [05:15:01] Okay, thank you. ## Q&A (05:15:10 – 05:34:00) [05:15:11] So let me kick it off here for [05:15:13] questions. I feel like I have a hundred [05:15:15] questions. They kind of come down to [05:15:17] some version of how does the internet work. [05:15:19] But I'm going to zoom in on like a specific one. [05:15:21] Which is at [05:15:23] what level of [05:15:25] who in, you know, [05:15:27] the United States or other countries actually makes the [05:15:29] decision of where data gets [05:15:31] routed. Because you can imagine that as an individual [05:15:33] I may say, ah, I don't want [05:15:35] my data going to a Chinese cable [05:15:37] but that decision is made upstream. [05:15:39] And so one of the questions I would have is suppose [05:15:41] you didn't actually want to invest in a new [05:15:43] cable but a country or a firm [05:15:45] said, I know, I'm going to [05:15:47] internalize, I'm going to try to internalize my [05:15:49] customer's desire to, you know, move [05:15:51] the shipment, to move this data. [05:15:53] Who would actually make that decision and is it possible [05:15:55] to use your framework to say, you know, if [05:15:57] the US were to say we're going to avoid Chinese cables [05:15:59] or vice versa, where does the cost [05:16:01] actually get shared among the world. [05:16:03] Steve? [05:16:05] Super just to take a second, sorry, I was a bit late. [05:16:07] I'm going to just give a story that, sort of, will cost [05:16:20] leadership the information but the goods trade [05:16:22] would still happen, maybe slightly lower audio [05:16:24] but it would still take place. [05:16:26] And so in the cross-section when you look at the [05:16:28] relationship between data and distance, is [05:16:30] partly influenced by this underlying relationship [05:16:32] between distance and trade flows. It's like, how [05:16:34] do we separate out those other [05:16:36] determinants of other types of interactions [05:16:38] that generate data flows. [05:16:40] I guess the ideal experiment would be a cable [05:16:42] cut with nothing else changing. [05:16:44] During war, so obviously that's complicated, [05:16:46] so things are changing but maybe like Russian [05:16:48] sabotage of cables in the [05:16:50] Baltic Sea between northern [05:16:52] competent countries could be some [05:16:54] interesting experiments like that. [05:17:03] Given the specificity of the [05:17:05] application, I thought you could go deeper [05:17:07] into the outside option, [05:17:09] not just being, I read out [05:17:11] on the existing network but I [05:17:13] can repair the cable, I can lay a new [05:17:15] cable and to me [05:17:17] it was particularly interesting [05:17:19] where a lot of the activity might be [05:17:21] between two neighboring countries, so the English channel [05:17:23] but the English channel is short [05:17:25] and is relatively shallow and so it's easy [05:17:27] to re-cable and so thinking about [05:17:29] how hard it is to actually repair a cable [05:17:31] depending on where it's cut [05:17:33] and how easy it would be to, you know, [05:17:35] in a pinch lay a new [05:17:37] cable would be pretty interesting. [05:17:39] Yeah, so, yeah, very cool paper. [05:17:45] I guess in the model every [05:17:47] bit of information is [05:17:49] differentiated, productive knowledge in real life [05:17:51] is mostly entertainment, streaming videos, [05:17:53] things like that, right? [05:17:55] One idea would be like suppose [05:17:57] that my cables are cut [05:17:59] and so I need to reduce congestion, [05:18:01] maybe I just slow down streaming [05:18:03] or something like that, [05:18:05] slow down sub-parts of my [05:18:07] traffic that are not productive. [05:18:09] That's another sport. [05:18:11] Can I answer these? [05:18:13] Yes, okay, yeah, let me answer these. [05:18:15] Many, many very interesting questions. [05:18:17] So let me start with [05:18:19] the idea of routing preferences. [05:18:21] So we know for example [05:18:23] that there is a cable that connects [05:18:25] Russia to Europe, sorry, [05:18:27] Beijing to Europe via Russia [05:18:29] and that cable is not used by [05:18:31] internet service provider in Europe [05:18:33] and that suggests [05:18:35] that some of these [05:18:37] preferences or concerns are reflected [05:18:39] in routing decisions also upstream. [05:18:41] We do have in mind [05:18:43] one thing we can do [05:18:45] to talk about that specifically because [05:18:47] for a few years we observe the routes, [05:18:49] the data packets flow through [05:18:51] by country. [05:18:53] So we can in principle back out the preference [05:18:55] that you might have into avoiding [05:18:57] specific territories. So that is [05:18:59] another massive data problem [05:19:01] that we have to handle but [05:19:03] we are doing steps in that direction [05:19:05] because for sure that's at least [05:19:07] for big things [05:19:09] it seems to matter. [05:19:11] So in terms of trade costs [05:19:13] and data, so the first version [05:19:15] of the model we wrote was actually exactly that, [05:19:17] that data flows were meant [05:19:19] to reduce Heisberg costs and that was how [05:19:21] we linked them to welfare. [05:19:23] So it's a very valid point [05:19:27] so what we find when we do [05:19:29] our gravity estimation, [05:19:31] actually if you control for [05:19:33] geographic distance, [05:19:35] you find that geographic distance doesn't matter [05:19:37] for data flows but what matters is the network oriented [05:19:39] distance. So we think that [05:19:41] that tells us that at least [05:19:43] the network matters more [05:19:45] than just the co-founders that also can be [05:19:47] cultural, proxemic and things like that. [05:19:49] But yeah, of course this is not a perfect [05:19:51] response and I think there are many ways in which data [05:19:53] matters for welfare and we are capturing [05:19:55] one that is relatively simple and easy to [05:19:57] estimate to say something more concrete [05:19:59] which leads me actually [05:20:01] to the question of entertainment. [05:20:03] So it's true that the share [05:20:05] of international traffic that is entertainment [05:20:07] is growing but it's not a major part [05:20:09] of data flows internationally [05:20:11] because most of Netflix and things like that [05:20:13] they have their own data hub, [05:20:15] their own server in some country [05:20:17] and they use the domestic network rather [05:20:19] than the international one but of course part [05:20:21] of it is indeed low value [05:20:23] data flows that we're capturing [05:20:25] but it is sort of captured by the curvature [05:20:27] of the production function a little bit [05:20:29] and we have a lot of extra [05:20:31] when you're already importing one terabyte [05:20:33] the extra terabyte is much less than the first [05:20:35] so we have in the background [05:20:37] a way to think about that [05:20:39] and finally well, [05:20:41] Matias is not here but [05:20:43] repair cables [05:20:45] well what we have in mind [05:20:47] so it's true that our ships [05:20:49] are dedicated to repairing cable [05:20:51] but it's only 70 of them in the world right now [05:20:53] and thinking about the [05:20:55] episodes in the Arnold Strait [05:20:57] strange to well [05:20:59] let's say it's not obvious to think that you can [05:21:01] send a repair ship and repair a cable [05:21:03] in a straight during an active war zone [05:21:05] so we have that concern that [05:21:07] it's not obvious that repairs can be done [05:21:09] in the scenarios that we have conditional [05:21:11] on the scenarios being triggered [05:21:13] so that's why we don't give too much attention [05:21:15] to repair in principle we could think [05:21:17] about [05:21:19] characteristic of straights and their ability [05:21:21] to be conducive to repairs [05:21:23] that's for sure but [05:21:25] we just don't do it [05:21:27] at the moment [05:21:29] let me just say a word about the discussion also [05:21:31] which of course [05:21:33] I'm sorry that we focused on a completely different part [05:21:35] it must have felt like two papers [05:21:37] because indeed also to us it feels a little bit like two papers at the moment [05:21:39] there are too many [05:21:41] two very different ways to sell this [05:21:43] and we just happen to focus on different things [05:21:45] so thank you very much for complimenting [05:21:47] the presentation [05:21:49] with also the other stuff that is in the paper [05:21:51] okay [05:22:41] so if you're like that here [05:22:43] a country may actually want to kind of cable [05:22:45] with the price on country [05:22:47] connectivity [05:22:49] because just purely economic [05:22:51] the grounds [05:22:53] is better off [05:23:00] so one is [05:23:02] I really remember [05:23:04] at the war with the age [05:23:06] we used to reply way as data [05:23:08] we said [05:23:10] how all of the data at this site [05:23:12] is not really as a society's estimate [05:23:14] remember when I was in the army [05:23:16] and the mission transparent [05:23:18] we have a micro second [05:23:20] so pretty [05:23:22] interesting [05:23:24] so I'm happy we are in Taiwan [05:23:26] we said we have a problem [05:23:28] we have a problem [05:23:30] when we do [05:23:35] the satellite we are here [05:23:37] the second point [05:23:39] so that we can quickly use [05:23:41] sorry [05:23:43] sorry [05:23:45] there is another part [05:23:47] of why [05:23:49] better now [05:23:51] better now [05:23:53] okay [05:23:55] so there is another part of why the United Kingdom [05:23:57] cut the cables [05:23:59] in 1914 [05:24:01] which is it gave the United Kingdom control [05:24:03] of the narrative [05:24:05] so everything is happening in Belgium [05:24:07] the Germans invade Belgium [05:24:09] and the United Kingdom starts to spin out the story [05:24:11] about the Germans with very nasty in Belgium [05:24:13] but nice [05:24:15] but the British are really able to [05:24:17] put things in a much worse way [05:24:19] that they were probably [05:24:21] so I could imagine that one of the problems [05:24:23] that Taiwan may face if China cuts its cables [05:24:25] is that [05:24:27] China is going to be able to dominate the narrative [05:24:29] China is going to be able to send photographs [05:24:31] is going to be able to send videos [05:24:33] and the average person out there [05:24:35] is going to see a Chinese video [05:24:37] not that Taiwan is video about what is going on [05:24:46] think about the cost list during normal times [05:25:12] there is a paper [05:25:16] gigabyte transmitted [05:25:18] which would [05:25:20] reward efficiency on the data side [05:25:22] and give a tool that can use [05:25:24] these sort of situations [05:25:26] thank you [05:25:37] I think what a few questions have been [05:25:39] touching on what I have been sort of wondering [05:25:41] is what is the nature [05:25:43] of these international [05:25:45] flows of data [05:25:47] so it doesn't have to do with trade [05:25:49] it doesn't have to do with [05:25:51] finance [05:25:53] what sort of communication is it [05:25:55] and I think [05:25:57] that would help me understand [05:25:59] what the non-linearities would be [05:26:01] in a stress time [05:26:03] so imagine it's nice [05:26:05] for the US to talk to England [05:26:07] usually because I can call up a friend there [05:26:09] but if there is a war and I need to make plans [05:26:11] because we are attacking Germany [05:26:13] then it becomes really important [05:26:15] and so how do you think about that non-linearity [05:26:19] so I think some of these questions [05:26:21] are making sort of a variant [05:26:23] around the same point which is a little bit [05:26:25] the idea that [05:26:27] data can be compressed or shrunk [05:26:29] what's [05:26:31] why, what's the non-linearity [05:26:33] in the value of data [05:26:35] and also [05:26:37] yeah [05:26:39] okay so let me address [05:26:41] these points so I guess that [05:26:43] we'll [05:26:45] um [05:26:47] so we do have [05:26:49] these non-linear effects in the value [05:26:51] of data [05:26:53] because we know that [05:26:55] these data are partly redundant [05:26:57] partly not particularly [05:26:59] productive, partly not very valuable [05:27:01] so we do think [05:27:03] that [05:27:05] being able to transmit a small amount of gigabytes per second [05:27:07] is very very valuable compared to [05:27:09] the extra gigabyte per second that you can [05:27:11] send it after the millionth [05:27:13] that you send [05:27:15] but still [05:27:17] there is a fundamental [05:27:19] way in which the network is wired [05:27:21] for which it's very hard to all of a sudden [05:27:23] switch to a very compressed [05:27:25] kind of communication perhaps for the military [05:27:27] that's what the military [05:27:29] would do be very efficient but for the [05:27:31] average person in Taiwan [05:27:33] that opens their phone they will find [05:27:35] that they just cannot access the websites [05:27:37] Wikipedia for example which is also in Singapore [05:27:39] for example of what these [05:27:41] data flows are [05:27:43] so what these data flows are are many [05:27:45] things including I think trade and finance [05:27:47] AI for sure [05:27:49] is a big part of it now and [05:27:51] these are all valuable and [05:27:53] productive inputs [05:27:55] into countries economies [05:27:57] I agree that we might be [05:27:59] at the moment we are probably being a bit [05:28:01] exaggerated on the [05:28:03] normal way to put to that input [05:28:05] of production it does come from [05:28:07] an estimation exercise we do which doesn't [05:28:09] mean that it's the right number but it means that [05:28:11] we are indeed working on disciplining [05:28:13] that specific parameter [05:28:15] and I think we have a good methodology [05:28:17] we just have to [05:28:19] nail it down and then we will have a defensible [05:28:21] number 0.5 is on the upper bound [05:28:23] but the lower bound of what we started [05:28:25] doing is still one quarter [05:28:27] one fourth sorry [05:28:29] so that still suggests very high welfare [05:28:31] effects [05:28:33] similar well not similar but [05:28:35] in line in terms of magnitude to what [05:28:37] I presented today [05:28:39] just another thing controlling the narrative [05:28:41] yes that's an unmodeled [05:28:43] cost of losing [05:28:45] access and we know how important that is [05:28:47] and I think that actually [05:28:49] speaks to the [05:28:51] potential even further [05:28:53] effect of why you actually [05:28:55] want to send videos of cats [05:28:57] during wartime right because you might be able to [05:28:59] send things that [05:29:01] become viral on the internet to push the narrative [05:29:03] and rally support around [05:29:05] the specific cause [05:29:07] and cost of laying new [05:29:09] cables so that is for sure something [05:29:11] we're moving into towards [05:29:13] for the part of the paper that studies investments and the [05:29:15] coupling we do observe cost [05:29:17] so we can do a cost benefit analysis for sure [05:29:19] we of course need to [05:29:21] as was suggested by chancey to [05:29:23] simulate the model to estimate the [05:29:25] added value of the extra lines [05:29:27] that is extremely competentially costly [05:29:29] to estimate all but we'll [05:29:31] get there it's a feasible [05:29:33] and we have the machinery to study also [05:29:35] that so why not [05:29:37] okay [05:29:39] well pricing per [05:29:41] units of data the question of why [05:29:43] why don't telecom companies [05:29:45] they do you know you typically [05:29:47] pay a fixed price for [05:29:49] accessing a certain amount of gigabytes per second [05:29:51] of connectivity and this also scales up [05:29:53] to the level of companies where [05:29:55] they typically actually have to pay [05:29:57] per route so if you want to establish [05:29:59] a route between London and Rio de Janeiro [05:30:01] you pay based on how much [05:30:03] you want to send on that route so pricing [05:30:05] does indeed reflect [05:30:07] these the cost that we [05:30:09] talk about and in fact one of the validation exercises [05:30:11] is exactly to show [05:30:13] that our latency measures can predict [05:30:15] prices essentially with an elasticity of one [05:30:17] which is one of the restriction [05:30:19] that we assume and that is good [05:30:21] to see that is burned out by the data [05:30:23] one more question [05:30:25] so [05:31:56] yes [05:31:58] so [05:32:00] yes [05:32:02] yes [05:32:04] thanks for this amazing question [05:32:06] so about the routing it's [05:32:08] an automatic system of routers that are coordinated [05:32:10] around the world we are very [05:32:12] complicated it's called the BGP [05:32:14] table I think where they have [05:32:16] you know immediate routes and [05:32:18] alternatives if that route is too trafficked [05:32:20] yeah [05:32:22] so it's a very automatic system in fact [05:32:24] our assumption that we routing goes [05:32:26] immediately to the optimum is probably [05:32:28] too optimistic about how the [05:32:30] network is actually routed but there is [05:32:32] an element of automatic routing [05:32:34] and less control of [05:32:36] routers on where traffic [05:32:38] actually goes it's something [05:32:40] pre-specified let's say that encryption [05:32:42] I agree you know that limits [05:32:44] the value of espionage still [05:32:46] it means that if you interest of data [05:32:48] you can put code crackers into [05:32:50] it and get information out of it just [05:32:52] like the British did during World War [05:32:54] 1 and 2 so encryption [05:32:56] you know is valuable but it's also as [05:32:58] limitations for sabotage [05:33:00] in satellite domain [05:33:02] yes that's a very hot topic right [05:33:04] now first of all it's relatively easy [05:33:06] to jam satellites [05:33:08] you have to be in Taiwan to [05:33:10] jam satellites that Taiwan can access [05:33:12] but that could happen and [05:33:14] but the second kind of sabotage [05:33:16] that is worrying about satellites is that [05:33:18] both China and Russia are experimenting [05:33:20] with capabilities to actually directly attack [05:33:22] satellites so that kind [05:33:24] of things you know there is a lot of redundancy [05:33:26] especially in Starlink we're talking about [05:33:28] now I think 12,000 satellites in orbit [05:33:30] and each of them is relatively [05:33:32] not particularly important but [05:33:34] these capabilities are of course worrying and of course [05:33:36] reflect the importance [05:33:38] of satellites for [05:33:40] as a strategic value that we are [05:33:42] with the Taiwan case right I think the point we try to make [05:33:44] with the Taiwan case is that satellites have a lot of [05:33:46] buffer value and therefore [05:33:48] they are valuable targets for [05:33:50] adversaries alright thank you so much [05:33:52] back to 45 [05:33:58] we are we're four papers