In Conversation with Mariana Mazzucato, Tim O’Reilly and Ilan Strauss on Algorithmic Attention Rents

going ahead thank you hello and welcome to this webinar I'm Zab enin chair and director of the data for policy community of interest and also Editor in Chief of data and Policy Journal I will be hosting this session together with my co-founder of the data prop policy Community John crof professor of communications systems in the computer laboratory at University of Cambridge and also researcher at large at the AL touring Institute we are organizing this session in collaboration with Cambridge University press and you and ucal Institute of innovation and public per purpose to Mark the publication of an important paper in daytime policy out outlining new theory of algorithmic attention rent how platforms control user attention and Sh shape public shape markets we are delighted to have the three authors of this paper with us in this conversation Mariana matato uh founding director and professor of uh professor in the economics of innovation and public value at the UCL Institute of innovation and public purpose Timor Riley CEO and Co and a founder of or media and visiting professor of practice at ucal Institute for Innovation and public purpose and Elan stros um senior research associate at ucl's iip again Institute for Innovation and public purpose before we start just a few keeping housekeeping notes we expect the session to to last for about an hour and we have ear marked around 10 to 15 minutes towards the end of the session to cover audience questions uh we please use the link my colleagues will share in in the zoom to pose your questions through throughout the session um anyone can also vote on these questions to prioritize so we know which ones to ask first uh at the end um please also note that the event is being recorded and it will be made available on YouTube via data for polic uh platforms uh if you have any exper If you experience any technical problems add them to the chat please Andre HDE and other colleagues from Cambridge University press will help you U if anyone is experiencing any difficulty uh now actually onto the paper we want to talk about algorithmic attention rents a theory of digital platform Market power we have published this paper recently in the day Policy Journal it might be the fastest publication to through our peer review process we received great reviews on that so thanks to all three authors for choosing to publish their work with us with our uh on our platforms uh if you have not yet read the paper yourself I would urge you to do so quickly immediately the link has been provided to you uh with the registration correspondence and also I believe my my colleagues uh can share it in chat without talking too much myself I would like to just hand it over to Marana team and Anan in that order to tell us a bit more about themselves and also share their opening remarks um few minutes each please thank you so Mariana over to you please sure so thank you so much for organizing this and it's great to be here with my co-authors maybe just some words on how this project began uh Tim and I met I think it was on Lake KO in Bel I can't remember if we had actually met before but anyway there we decided that we needed to team up on a project of this sort bringing together really what iip my Institute at UCL was founded to do which um is to kind of bring forth much more explicitly than is often done the direction of growth the direction of technological change and how contested it is there's nothing deterministic in how technology develops there's nothing deterministic in how growth develops with Tim's deep understanding of Technology around let's just call it the Silicon Valley it Revolution specifically today we're talking about algorithms but he's really the the guy who I think founded even the word Web 2.0 uh so bringing together these two kind of expertise as mine more from political economy um Tim's more from the knowledge of the technology but really to ask the big question how can we make sure that the evolution of technology today and not just artificial intelligence but we'll be focusing on that today is um talked about is governed is contested in a way that actually brings brings forth the the Deep issues and doesn't assume that there's only one way of thinking of corporate governance or policymaking and I I myself have written a lot about the the bottlenecks that we have around policy when we just think about it as fixing market failures because then by definition and by Design EUR always too little too late reactive not proactive and the bottl neck and problems with the corporate governance structure which thinks that the only thing it can do is maximize shareholder value and even though there's Concepts like stakehold value it's it's really remained I think a pretty flimsy concept so in some ways we can actually see the work that we're doing today in this project in this paper but also related work as really asking what would stakeholder value actually look like in terms of how we produce uh not just expost and how we pick up the mess afterwards um and it starts really with this concept of rent why is rent important in this paper and in economics because anytime we're looking at profit making and capitalism right we're talking about capitalism not feudalism um we have to actually differentiate how are profits made there's a whole um literature on kind of sharian profits versus ricardian rents and the idea that actually rents are not actually reflecting investment Innovation entrepreneurship but are really just reflecting control over some sort of scarse factor of production uh rent seeking refers to the attempt to generate income not by producing anything new but by either overcharging the competitive price undercutting competition or doing something that for the you know sake of speed here we'll just call bad so what is bad what is good uh that actually depends ultimately on what one's theory is a value creation um and this distinction between value creation value extraction and also value destruction is unfortunately no longer actually at the Forefront of economic thinking you know um Adam Smith Carl Marx David Ardo later Schumer that was a huge distinction even the physiocrats the first economists ever in the 1700s they really paid attention to that they looked at how value was produced in farming then got very worried about getting extracted out of the economy by the landlords um who they called the sterile class because they wanted the system to reproduce itself so using these biological metaphors then they used the word sterility when there was too much extraction and really the theory of rent seeking and rent has kind of moved along from a theory of landlords and rents and and farming and in real estate which by the way still exists a lot of rents are captured in the real estate sector something like 80% of Finance in the UK actually goes to finance insurance and real estate um so from landlords and land uh patents intellectual property rights are another big area where people look at rents um I talked a lot about that in my book value of everything the financial sector as I just mentioned has been looked at in terms of rents and really what we do in this paper which I think is quite new is to extend that idea of rents at the level of production right how we produce how we govern production um to AI in the modern digital economy um and on that note I actually had some notes here where I was going to introduce how we do that with algorithmic rents but I think it's it's neater if I now hand over to Tim to say something about algorithmic rent so kind of um by the way there is a paper that I wrote with Josh Ryan Collins and and goros gilis about called mapping modern economic rents that kind of brings you through that history of rents from the landlords Finance intellectual property rights and to the modern digital economy and really what this project then did was to say let's hone in on rents in the digital economy also because this is one of the biggest problems that we face today I'll just say lastly that because rents are so large in this sector the talent that is going into the sector is being paid with very high salaries uh so it's the first time that in the history of capitalism I would argue that the technology and the knowledge is almost all inside not only the private sector but within few companies and that issue of skills is hugely important because it's almost impossible to govern this sector without the skills and knowledge kind of at the Forefront of the area um but where is the money coming from to pay for these excessive uh uh uh uh well the excessive of rents in the sector this is really what um the project also does so I'll hand over to Tim to talk about how we actually look at algorithmic rents yeah so uh let me start really uh I'll come back around to rents but I want to start with the tech industry and my in some ways my love affair with Google and Amazon um you know when they first came out they were doing something new and marvelous and particularly in the case of Google I thought was a tremendous advance in the structure of markets uh you know for for so many years uh you know we've sort of thought of of prices almost as as the mechanism the of the Invisible Hand they're the thing that coordinates and sums up the collective knowledge um you know that that allows the market to work but suddenly you had a matching Marketplace in the case of Google in particular that was matching hundreds of millions eventually billions of us users with hundreds of millions of suppliers of content maybe billions trillions of there are trillions of web pages using these algorithmic signals and this kind of what I called in 2005 when I wrote my what is web20 paper collective intelligence and Amazon too kind of harnessed collective intelligence to figure out even though they were taking price into account they were taking all kinds of other signals as well uh to to say this is really the best match for what you're looking for and so there was this huge advance in markets and so it started to concern me as I saw that change and at first it changed uh because of attacks from people in the ecosystem so for example uh you know spamers uh working on Google and there was a major update to to Google uh Google's algorithms uh in 2011 called the panda algorithm update I I I I was later told by someone at Google that was partly inspired by my feedback to them that they were losing the war against spammers and you know so it was kind of like they were still really trying to be virtuous and give value to their customers but in the second decade of the you know of their existence the or really the maybe the third 95 was the starting point for Amazon 98 for Google but certainly after the 2010s you started to see something different happen and you saw particularly at Amazon with their ad business where instead of trying to improve the experience of the customer and make the best match for the customer they were starting to make the best match for themselves and I was really concerned about this and then when I read Mariana's book the value of everything and its account of rents I suddenly had a language for thinking about that and Mariana and I actually met when we when that book came out and we actually did did an event together at Bloomberg where I interviewed her about it and uh but it really was was a a landmark point in my thinking and so we were at a AI convening in Bellagio I think it was on AI uh in 2019 and we were started talking about rents and algorithms and and Marian said well we have this new project on called mapping modern economic rents you know why don't we do something together so that was really the origin of how all this came together and I've kind of summed up this narrative in a in a u another paper uh which is called Rising tide rents and Robber Baron rents which contrasts the kind of virtuous rents that come when companies are truly innovating with the extractive rents that you see later uh but anyway as we we started talking about this and thinking about it we really started thinking about a couple of things one this this core notion of uh sort of a uh an act ual good that you can actually identify you know in in in the case of these algorithmic matching systems and this applies not just to Google and Amazon but you know dating sites to whatever it might be you know the the good is actually the best match and in the in the lore of Technology they call this the best organic results you know or or they talk about earned attention for example in social media you you get you you attract you attract attention because people like your stuff so that that's earned and in some sense the essence of rents is that they are unearned maybe they're bought uh and in the case of of of what what start we started to see at Amazon in particular uh they were replacing the best match with paid results that were actually better for Amazon and so we started really thinking about how do we actually do that what is the market power that Mark that Amazon has that allows them uh to make this bad match and get away with it and and uh so that's really what became this the subject of of the research but there's another component I want to get to before I I I let Elon dig down a little deeper into into the mechanisms that we explored and that is there the notion of an ecosystem because part of the match and this is reflected in the European Union's definition of these platforms as Gatekeepers right they are matching in users with suppliers of some kind in the case of of websites it's suppliers of information in the case of Amazon it's suppliers of products uh and in the case of social media maybe maybe other users uh who are are suppliers or maybe firms that are suppliers of content and the question is is there a harm to the ecosystem uh and and in some sense uh it's that understanding the mechanism G of gatekeeping through the company's algorithms that we we set out to explore and uh and and and we we we came to call them algorithmic rents generally but more specifically we dug in on what we called algorithmic attention rents there are other classes of algorithmic rents you know you can there for example with Google there a lot of rents that are being extracted in their advertising ecosystem through their control of uh o over all parts of the of the advertising uh pipeline we didn't go into the detail on that we were really much more focused on this uh set of interactions between the user and the ecosystem of suppliers and the self-serving distortions of the good that these companies are are capable of delivering as one other point that that Mariana has always when we talk about the shows reminds me to to bring up and that is you know there there's a lot of narrative that says you know like the surveillance uh capitalism narrative that sort of says that this sort of approach to Big Tech of algorithmic matching was doomed from the beginning it was evil from the beginning and I don't buy that I think it was a real advance and a real public good and there is an imperative that that makes it go bad and it is that shareholder value imperative because because part of the dynamic and this is the point of my Rising tide rents and Rober Baron rents paper is there's a virtuous period of Market expansion when you get users by doing really good work and then once you asserted your dominance and the market has stabilized you don't get the easy growth but you the market says You must keep growing your profits anyway so at that point you shift from these Shum in rents which are the rents from Innovation which are virtuous in some sense they make us all richer even though they are captured largely by the party uh that that is creating them but when they would be competed away the company instead turns to other kinds of rents uh to keep their high level of profit and that's what we saw happen as the internet growth stalled it happened earlier with the personal computer uh we won't go into that uh grow stalls but you have to keep the profits growing so you become extractive and that's a choice that companies make and and I think understanding that it's a choice I think is Central to this uh this discussion at any rate so over to Elon to talk about some of the mechanisms that we dug in on uh about how this is actually done thanks Mariana too um actually two things the one is how are algorithmic rents actually operationalized by these platforms and uh what theory did we think uh we needed to ground our arguments in and then secondly um why did we choose to kind of focus by way of application on Amazon and we wrote two companion papers dealing with the application of this to Amazon's Marketplace uh whereby Amazon instead of showing users the best search results uh you know the best uh baseball from millions of different baseballs and now shows you on the first screen mostly advertising results so based on who paid the most to appear at the top of the search results um and so we uh also decided to focus on Amazon's algorithmic R and we wrote two companion papers which are being published um one in the UC uh um Law Journal uh on technology um but firstly um on algorithmic attention RS the idea being that these platforms have power because their own algorithms and algorithms are allocating users attention to results which might not benefit the user but they're benefiting the platform so we presented this idea that platforms have power because they own algorithms and then instead of showing users the best search results Google or Amazon are going to show inferior results and these inferior results are going to benefit the platform for example advertising those results are more directly profitable to Google if you click on them um and we presented these results um about I can't remember 18 months ago to various academics to experts and they were entirely unconvinced uh by them and I went away and I was a bit disillusioned and I thought Wow we must be totally on the wrong track here clearly it's a waste of time to look at algorithms but then I looked at the media especially covering what was going on uh in antitrust uh also known as industrial organization or competition law in Europe in America and increasing all the cases were about algorithms Amazon is preferencing sellers who um use its shipping Services uh or how hotels are colluding through algorithms or Amazon is exploiting its ecosystem of suppliers its Merchants by forcing them to pay increasing advertising fees so why were academics so against the idea of L algorithms and why were Economist so against the idea and I realized it comes back to the idea that economists and competition Scholars assume that markets are basically perfect unless someone is fixing some kind of price somewhere and then they're a monopolist and the monopolist is exploiting consumers through higher prices but this totally ignores uh the fact that many of these markets there aren't often prices and the way consumers are being exploited is by being shown inferior information U so the research results uh the recommended results aren't as good but they all assume that consumers will not just click on the first results shown to them they're going to click to results page number 79 they're going to scour the internet for hours looking for the best baseball to buy looking for the best deal because consumers are unconstrained by time unconstrained by limits in their cognition they can spend all day on the internet looking for the best product to buy because consu because economists assume that homo economists the consumer is all knowing all rational and that is why leading antitrust Scholars say that there is more competition online not less compared to ordinary everyday Market because online because online there is more information online there are more websites to choose between so actually we should be devoting our uh uh resources when exploring rents and Monopoly uh not to online markets so we realized we actually had to reconstruct a theory which didn't rely on assumptions of perfect information perfect consumer rationality and Tim came up you know sent this great quote to me from Herbert Sim basically saying that uh when uh information is basically abundant then uh attention becomes really scarce and any Technologies which help allocate our scarce attention are going to become incredibly valuable and researching Herbert Simon and drawing on the tradition of institutional economics uh we reconstru constructed fully this theory of algorithmic attention rents which basically says that it's not just markets which make allocations based on prices but it's all kinds of Institutions which help markets do their job and online what is the main institution which has arisen to make market like allocations and these allocations is are increasingly based on user attention it's algorithms because algorithms are allocating resources algorithms are uh making sales by allocating attention to certain Goods to certain websites um and it's algorithms which basically through the allocation of resources uh are also allocating power and profits um and driving the production process so by kind of breaking with that neoclassical tradition uh we were able to see why algorithms are so powerful online because users rely on them to make decisions because they're not perfect and that is why we have to interrogate algorithms and in our and just to to kind of wrap up by focusing on Amazon that is why we say in our two companion papers on Amazon that if you want to interrogate the best Market allocation you know Economist will say well look at the price is it competitive and then we'll know uh because prices when they're competitive user preferences are satisfied but online when you look at Amazon search results there's a million indicators it's not just price so we said you actually have to look at the algorithmic allocation as a whole the order of the search results and the algorithmic allocation will tell you how fair or how unfair anti-competitive and rent extractive the market is so if you want to take the competitive temperature of a market whether it's Google search whether it's recommendation videos on Instagram or whether it's the search results on Amazon look to the algorithmic allocations look to the algorithmic search results which come up and the more that those uh search results are flattered with self serving results whether that's addictive content which keeps you on the platform for longer even if it doesn't fully satisfy the US of preference preferences or if it's advertising results which benefits the platform because in the case of Amazon those advertising results is just another fee which sellers have to pay in order to gain user attention so we said if you want to take the competitive pressure uh competitive temperature look at the search results and then the last thing I conclude this is that we the final ingredient why these online markets are so unique and why advertising online is so unique it's not just that it's instantly clickable but it's that it's an instantly clickable substitute because consumers are making decisions through a screen and that means that more of one search result means less of another because we hardly ever click to the next screen so that if you see 10 search results on Amazon the first five search results really matter and they're going to crowd out uh competing search results which they uh are in place of so because of the limited aperture of the screen and especially on mobile as Tim emphasizes more of one kind of content means less of another and that is why advertising so pentious online because it crowds out the best results and um in the presence of continuous clicks from users at the top of the screen it is used as the kind of rent extraction device and yeah I just urge the audience to kind of read this paper further as a fantastic outline of our Theory and then also consider not just that important paper which Mariana mentioned on mapping rents but also the companion papers on Amazon which try and apply this Theory to Amazon's Marketplace look at the evidence of how it is a rent extraction device and why advertising in particular is this algorithmic uh rent extraction device thanks thank you so much um now John I would like to give it over to you to reflect and also have go through your yeah um really great introduction to the background why everyone should read this work and the companion papers I'm I'm delighted with this uh drawing attention to attention rent um and we are all concentrating on this and not looking at the sidebar ads that Cambridge University press has been pushing out behind us and the virtual screens um so actually there's an interesting phrase I'm sure Tim's familiar with that that we spend a lot of time optimizing how quickly we can get things in front of you there's actually a phrase that we redesigned all the Internet Protocol stack for happy eyeballs this is literally to deliver the maximum amount of advertising real estate on the screen in the shortest amount of time and we you know the entire Google stack was reoptimized over a new protocol called quick to replace tcpip in the internet just for that reason um and it you know to increase their clickthrough revenue so I I buy the story and and the cleverness that went into fixing things like that is really alarming so I could ask a couple of questions just I think I'm going to roll them together because they're kind of related but there are there have been proposals for fixing uh Market problems and you know one of the areas people have said is well could we not just move an awful lot of these platforms into subscription rather than advertising and certainly when you look at the pernicious nature of content in traditional media like PR newspapers in the 19th century and then radio and then TV in the 1950s the commercial sector were driven into dominating the the content with with advertising and and and and disappearing the the good content but but there are Corrections that have been proposed so I'm going to mention one which is a bit older it's not that old which is Thomas py suggested uh taxing capital for people that were just extractors of rent and the piy tax doesn't have to be very large to shift the equations towards thinking about different Market models and I wonder what you think about could we think about taxing the owners of your attention the platforms that hold a large fraction of the screen that everyone's looking at or indeed in the AI case uh the AI companies that have extracted value from someone else's content the whole of the common crw in the internet for example is the main training platform for at least 50% of what goes into open ai's platform for example and that's was going to go into the same space pretty soon you could easily imagine raising a picky tax on either of those things but in a regul framework might be very difficult so if you got any thoughts on that you are some of the world leading experts and thinkers in that area the second thing I mentioned is in in of course in I think um I think Mariano kind of hinted at this in Europe we now have the digital Market act which requires interoperation of Platforms in particular the larger platforms interoperating the smaller and is that going to help at all in this particular a part of the space so there two kind of questions is there a is there a kind of tax lever that could help and is there a regulatory lever that could help um shift away from this pattern that we seem to have fallen into yet again so that's my my question and uh any and all of you please have a go responding to that or or dismissing it if you don't think it's good question let me start by um saying that while I think about Regulators my primary audience has always been the CEOs and top management of the tech companies themselves you know I went back when I was doing my advocacy for open- Source software for example even before Web 2.0 I was saying hey this is a tech this is an approach to technology that's really useful you should learn from it yeah when I was talking about Web 2.0 I was saying here's an approach to to to how you use technology to build a better future you should learn from it and that's really what we're trying to do here with uh the first bottom level of algorithmic rents if you're self-interested and you actually understand that you are going down a path that other companies have gone before IBM went from being a value Creator in Computing to being extractive they got and part of it was maybe the government turned on them uh Microsoft went down the path of being a value Creator to being extractive uh government turned on them uh markets turned on them uh you know and you got innovators were driven out and go to create new competition precisely because you were not doing the right thing uh and now it's happening again you know so there is a self-interest to companies in actually not taking too much of the value I remember uh Walt Mossberg the famous tech journalist told me of a conversation he had with Steve Walmart when he was CEO of Microsoft and he said if you would just dial back the greed only 5% it would make a huge difference and it's true you know like there's simple things that Amazon could do to have an ad business but also you know uh say okay we're going to give the top slot uh to the top organic result which is actually the one that we think is best rather than making people scroll down to find it you know sometimes it's a couple of of Scrolls uh down before you can find the thing that they will even tell you this is our Top Choice you know and so uh you know with Google I talked to them about hey you know when you put a you know like if you do a search for something you could buy locally and you basically all give over all the screen space to National or online advertisers and people have to scroll down to find the local Merchants do you not think that that has a market shaping impact and you know what should your responsibility be so that's part of the answer uh is is just to actually understand that they're opening themselves up to competition by no longer giving the best results and and you see this you know in some sense with Google not going forward it's classic innovators dilemma they didn't want to threaten their their revenues so they didn't do the Bold thing that came from the outside with uh uh with at CPT if you Mariana or you want to talk about ideas on regulation just just without speaking too much about it I think one big issue and this is actually what schum peder I think would have really wanted and then somehow we got confused with Sher and just talked about creative destruction forgetting his biggest contribution Beyond Marx because markx already said everything that Sher said on Innovation sher's contribution was on competition and his point was we have a theory competition um that needs to talk about imperfect competition to even introduce Innovation which makes no sense in a mode of production where Innovation is how firms compete so just the fact that there's no Benchmark theory of perfect competition that takes into account a lot of the features of ways that modern companies compete is the problem and and from the point of view of I think regulation and policy it kind of requires there to be less of a division between here you can see who I usually talk to the Ministers of innovation and the Ministers of competition so in Europe for example there's a complete mismatch and lack of communication Lost in Translation between um you know vest dagger's unit or she used to be ahead of it is she still ahe of it I can't remember has she been can't remember is it still Margaret bestager anyway um head of competition and digital versus what's happening in the ministry of innovation um and that kind of comes back to my point about it's really hard to to regulate something well strategically intelligently if you don't understand it um and so it does come back also not only to the issue of skills which I think is a huge issue today because it is the first time I think in history um you know uh I wrote the entrepreneurial State 10 years ago about how so much of the Innovation that we attribute to the tech sector actually came from at least partly government Investments but today actually that's almost no longer true almost all the knowledge all the tech um is is actually happening within kind of five or six companies and that's not unlined to the rents that are being used then to pay these massive salaries to University researchers who are leaving universities and going to these kind of five or six companies um but just as a broad statement without going into maybe the details of tax I think a general thing is we need much less of a division between the policies that are coming out of ministries of innovation and those coming out of of ministries of competition and regulation because can't really I mean even regulation seen as like too big you know then potentially if you just divide up the big you just get a lot of little rent seeking animals as opposed to going to the core of how those rents are being produced through Innovation through understanding of the algorithms and the ways that we've been talking about I don't know Elon if you want to add something yeah say I want to two things thanks um I used to think that subscriptions were part of the solution moving away from advertising business models but I'm less certain because if a company has dominance they have dominance because of strong Network effects meaning I get more benefits from this platform the more other people use that platform then once the company has dominance from Network effects that dominant company can take advantage of it in many ways and increasingly they take advantage of it through using subscriptions and advertising um and they do so in such a way where they constantly are changing the price which the user is charged often that's a free product like storage and then you're locked in and then now you have a monthly fee then what's to say that on top of that monthly fee they charge you advertising I have a subscription to the financial times now there's more advertising on the financial times in Bloomberg and they can keep on retching that up so there are other effects which are partly due to dominance sometimes because there's Network effects with Network effects they can do what they want even if it is a subscription model and partly it's because there's some lock in like I have a subscription now and from a subscription I've lock in and that opens up other opportunities for exploitation even if it is not a dominant firm because there's a subscription so I think it is more complex and I think the key thing there is to keep in mind that with network effects you need a bunch of regulations because past that Tipping Point The Firm can kind of do what they want and I think interoperability is an important regulatory lever um but the point is that interoperability let's say I could move my reviews from Amazon to another platform okay and then like that can make entry in the market more at least slightly more easy for a potential competitor to Amazon that inoperability can help but it's slow and Amazon has 10 years and probably $40 billion of fixed Capital investments in delivery networks so inter interoperability matters but along the way you need to say if this platform is dominant then regulate as this dominant platform now competition experts like Herbert Hamp saying no don't regulate Amazon because it's not always dominant maybe when it sells um this dress it's dominant because it accounts for 60% of the market share in the US but when it sells cell phones you know when it sells um uh like listening uh in a bluetooth listening it's not dominant so it can only be regulated on a product by product basis but one of the implications of our framework is saying you know what Amazon's dominance it doesn't just stem from its market share right in these products you know even when Amazon isn't producing clothing even Amazon is not selling the airpods it's dominance is because so much Discovery takes place on Amazon even when I'm in a physical shop I will use Amazon for Discovery to compare products and I might not even make that search that uh that purchase on Amazon so it is true that Amazon itself doesn't account for um the dominant portion of sales for every product happening in America um but we're kind of saying that this algorithm uh governs visibility in kind of a cross cutting way throughout the economy and part of our approach to dominance is to say that Amazon isn't just a delivery platform it's not just a Marketplace but it's also a Marketplace for information and Discovery and that dominance and that approach to dominance needs to inform regulation you know I I think one of the things that we're uh trying to say and trying to in in inform Regulators is with a sense of how to think about the problem because if you think about it through the the lens of prices you don't have very many levers if you think about it through the lens of information allocation and attention allocation you notice some things you notice for example that companies have an idea which is fundamental to their business of what is the best match and you can see when they divert diverge from it you know we know that that that Google's best match is its top organic result we know that Amazon's best match is its top organic result how far how how hard is it to find that result how much has it been pushed down and this is part of why we've ended up going uh sort of a big part of our thinking there's yet another paper which we didn't link to about disclosures it's not that disclosures are a magic w but if you can start to build a language that that gets people thinking about oh the companies are disclosing what they're optimizing for you know you think about about um you know just Financial disclosures you kind of go okay we we know how much profit they're making we know how much Capital they're investing uh you know there's a lot there's a welldeveloped uh set of metrics that that the the market uses to pay attention to the things you know and there's analysts and all all the rest but there aren't analysts who are looking I mean there are actually actually I think back there are analysts who look at this but they're not Central to the thinking of regulators or of people who are uh looking to Value the stocks you know they you know like if you talk to the the you know Marketplace pulse uh folks who look at Amazon or uh you folks who do Google SEO analysis they can tell you how the the results are getting worse but it's not Central to the discussion of how these companies are evaluated you know and and we're hoping to raise it so that that some hedge funds start saying oh we're going to bet against Google because we actually now know that they're giving worse results and we think that there's going to be a uh an impact we're g there's GNA be an impact on Amazon competitively because they're no longer giving the best results and on that it's interesting that even though now there's this whole backlash against ESG because of I don't know Texas being weird but um we know that there's a lot of greenwashing around ESG but the G of ESG and the s in some ways was was almost more weak than the E on the environmental and there is an opportunity again coming back to my point initially about stakeholder value Done For Real uh which is how you create value not just creating value badly and then saying oh we need to give back to the communities workers or consumers but actually creating it in a so-called good way from the start in a pre- Distributive way that these kinds of um metrics that we look at in the paper especially actually around the um disclosure paper could help uh transparency and accountability to hold companies that want to call themselves purpose oriented or doing the whole stake Val stakeholder value thing to be held accountable uh in terms of how they actually uh disclose the data um around uh these issues uh under the G cool back to Z any more questions we also have quite a number of questions in slide out from the audience um yes yes absolutely I've got one question well I'll just ask one question of mine and then there are several questions yes coming through we need to move on to that those ones uh I just wanted to ask about well it's theory of competi competition it's a it's a start of a theory of competition and it's a great one um but I'm interested in understanding whether you are look whether we are looking for a pure uh economic theory or are we looking for a more soci tech technical theory that might also include at some stage more of the algorithmic behavior analysis in the process especially I mean the platforms that you're talking in your you are talking in your paper uh like Amazon Google and the current attention Market is is uh is is is more in a way understandable but when things like llms come come into play when we have these foundational models coming into play and taking affecting more of the decisions how transferable is is this sort of this sort of theory or findings into into that sort of context as well I don't know if that's clear enough can I jump in really quickly I I think it's really interesting to to frame this these are questions of political economy not just economics per se and political economy is a lot about who has power and what's the nature of that power and you know in AI you still have these it's pretty clear what are the sources of power you know that's what we need to be interrogating and then what are the sources of economic value creation how do these things intersect uh my friend Bill janway likes to talk about a uh you know kind of like a three-sided Market of government uh you know uh the private sector and and um uh and and financial capital and and in some ways these things all intersect and but they are all intersecting uh and it's a big Taffy PLL to see what the shape of the of the resulting Market is going to be K pz wrote a whole book on that by the way yeah of course she taught Bill Jan oh I know this is true bill help fund her work but yeah was deeply influenced by it yes and and I think you know our work is directly relevant in a very crude way you know we're saying there's a direct correlation between the level of information in the market how good the information is and the level of competition when the level of competition is pardon my friend shitty the level of information will be shitty you know a andb for a while didn't even tell you the total price when you were booking it only told you end right didn't include all the cleaning fees and you know if you have a dominant producer of chocolates maybe they won't tell you the ingredients in the chocolate bar so but these are mature technological markets once llm markets mature and you have dominance then you'll see a closer correlation between how good the answers are and how good how much competition there is so that would be my kind of short answer I would add one thing to the power uh discussion I think it's also about narratives I mean um not so much around llm but around the kind of big Tech uh companies even you know pre uh just how they came to power basically Google Amazon and so on I mean in in many ways that Tech was not actually funded by them initially right so that's also where this notion of the ecosystem comes in how these different types of value creating organizations public private you might even today bring in some philanthropic organizations that are increasingly investing in some of these areas how they kind of work together and in some Ways by calling Tech big Tech we forgot that actually in some cases they were just big media companies and so they should be regulated as media companies not necessarily as tech companies so there is a bit about what are we even talking about what companies and in what Market are we looking at um I mean definitely Facebook I would argue is kind of more of a media company that should be regulated as such like the BBC great thanks so much we promised 10 minutes for Q&A but uh John did you have anything like urgent to ask because there are also questions coming up I will just for the time for for eke keeping uh I'll I'll ask the three questions um that are most popular currently and you can choose to answer whichever you like uh and then if we have time we will go through more so the first question is the paper advise advises for the need for algorithmic transparency how who should evaluate the quality of these algorithms the government companies or third party um the second question is can markets even function at all when the startup cost to compete with companies like Amazon is so high uh the question three is isn't it prioritize their profit compared to users the expected expected behavior from companies whose goal is making money shouldn't be change the market rules so three questions whichever you prefer to address uh I'm happy to jump in uh um first but if if anybody else wants to go first I'm I'm happy to wait you can all right um so uh the first one about the paper uh advises the need for algorithmic transparency who who should evaluate the quality of the algorithms the government companies the third party and I I think the answer is all three and and if you look uh for example at uh ancient technology and it was a technology I'm talking about Double Entry accounting invented in you know the the 12th or 13th century uh and it was an incredible way it enabled the whole Mercantile era because suddenly companies could manage their money in a different way and that became the lingua Franca of every company that is the companies actually used that as a as as a kind of internal self-regulatory mechanism and then government canonized it and said you have to make some of that public and what that enabled was actually a thirdparty Marketplace of of a kind of regulator so to speak uh you know the we think of The Regulators in in finance of being simply you know uh the the government regulatory bodies but what about all the Auditors for example companies are required to to do audits anyway my friend Jillian Hadfield has kind of taken that idea forward uh she wrote a paper with Jack Clark of anthropic about how do we build regulatory markets for AI you know and Regulatory markets are a really interesting idea because like it might be for example oh actually I could go deep on that I won't go too deep on it but in some sense you know it would enable uh companies with specialized knowledge about the subject to be the ones doing the audits but that requires government to have some standards about well what do you audit yeah so it's really kind of a a a u kind of a well a three player game you know the companies have to do it the government has to do it and a third party Marketplace has to do it and I've just put in the chat here that you know we need something like the task force and climate related Financial disclosures and that only happened when we had finally the consensus or I should say and part of the world given the Trump stuff but anyway that climate change was even a problem so the first thing is to Define what is is the problem the whole kind of good versus bad rents versus profits and then once we have some sort of consensus on what the bad stuff is and that's what we're trying to contribute to then that needs to translate into disclosure so digital related Financial disclosures uh not just climate related ones but that particular task force came out in 2015 after kind of 50 years of um you know climate activism yeah so um so can I just jump to another question um markets can markets function when the startup cost to compete with companies like Amazon is so high I think the answer is absolutely you know we're seeing it right now with um uh you know this disruption from AI where for the first time you know potentially there's competition for Google even though the startup cost is so high now of course the startup cost of of um uh building these large AI models is is very high but once they get built maybe it's not so high you know so and this is the thing that I think you know could be uh a critical axis in the discussion of the regulation of AI is how do we think about open source models because if open source models get to be as roughly as good as the big ones then you have a very competitive Marketplace and that Marketplace can deliver all kinds of new Services um and you know it's classic Kay Christensen you know the PC wasn't as good as the as the Mainframe but it got better faster uh the web wasn't as good as the as PC applications but it got better faster and and let it made the market bigger and I think AI is going to make the market much bigger and the question really is do we have premature centralization and and rent extraction because so much Capital has gone in already most of it coming from the big existing players by the way you know and that's the thing that's important it's not even invest who are are are driving this it's it's Google and Microsoft and Amazon who are making the giant investments in and and meta who are making the giant investments in AI so we don't know how all that's going to shake shake out but it's pretty clear that uh true Innovation can always shake up the market no matter how big it is and the one thing that government could potentially do if they're serious uh is is is would be for example to say sorry as Tim woo calls it no wimpy remedies you know he calls the AT&T settlement in 1956 at& had the totally get out of any application market they could provide telefony only and they had to give away all of their 30,000 patents you know he said Silicon Valley came from that that was a badass government intervention and you know you could imagine a badass government intervention that says yeah you can make a foundation model but you can't offer any consumer facing Services you know you're you're gonna figure out what's the equivalent of of your offering telefony that would be one r no you know um no wimpy remedies type solution there but it could well be that just enabling more uh you know not buying into the thing open source is more dangerous than than a big centralized close model uh I think would be a good start and then we see whether that works right I mean I'm very conscious we have one and a half minutes left but sorry how about a government oh no it's fantastic conversation I mean the government open systems procurement was what made the internet happen in the end where us and Europe actually decided we're going to require all government you know procurement of networking will be open and they switched from it being OSI to internet because it was pragmatic so you could imagine that happening in other spaces but the government yeah is one sector but I agree it could be a big one um I guess we have literally wrap-up minute but um I guess Elan did you have any more points to make against those questions or last remarks from no just well just to say and then going be my last remark I mean yeah you can't compete with Amazon and so the competition comes from what's called outside the market so basically you know who competed with Facebook it wasn't another Facebook it was Tik Tok and Tik Tok was a video format and it was not based on the social network it was based on the algorithm the pure algorithm the recommendation and so what that means is we got to wait for the next technological cycle and that kind of sucks because they inter proceeding 5 10 15 years depending on this uh dog type of life cycle um consumers get exploited Public Health gets destroyed information censorship goes wild you know all harms happen that competition happens just happens from outside of the market which can take longer so with that uh we really need to close it's end of our time uh for everyone if you have enjoyed this webinar consider registering for data also data for policy conf that's coming at Imperial College in July the Highlight theme is going to be trustworthy decision making with AI um also day Policy Journal is our main publishing platform we w we we welcome contributions to our blog as well as well as formal submissions to the Jour Journal you can check both conference and journal websites for for more information and also to subscribe with our new with our newsetter many thanks again to First our speakers uh and everyone joined us on this session um hope to see you again thank you very much this is the end thank you thanks so much thank you byebye bye bye bye then

Channel: Cambridge University Press

Views: 92

Keywords:

Id: iGvICGkr7I8

Length: 56min 47sec (3407 seconds)

Published: Fri May 03 2024