Below is a summary of, and a proposal inspired by, the search roundtable which closed out the Toulouse conference. The remarks are first and following them you can find my impressionistic notes of the each of the speakers’ presentations.
Open Search: A Proposal for Regulating Search the Open Way
In many parts of the world today we already have a situation in which a single search engine has a dominant market share (that is over 70%). At the same time there is, at least for the present, substantial competition — though entry costs are significant and growing. In such circumstances one must be concerned that a monopoly (or small oligopoly) will develop. Such an outcome would not only creates significant pricing/welfare issues but would raise very serious ‘cultural/social’ issues — any entity, corporate or otherwise, that operated the sole (or only significant) search engine on the planet would wield an absolutely immense, and likely unacceptable, degree of power over our political, cultural and social lives.
This concern has already begun to generate a growing discussion of if, and how, we should regulate online search. There are, however, several unattractive features of direct regulation of search: the cost of regulation itself, the inefficiencies arising from poor or inappropriate regulation (arising from the inevitable information asymmetries between the regulator and the regulated) etc.
Open Search: A Proposal
As a result most economists (and technologists) would prefer a situation where competition did the job of regulation. This brings me the key proposal of this article: there should be a concerted effort (government assisted or otherwise) to develop an open-source enterprise search system along with an associated not-for-profit open web search service built on top of this system.
The second part of this proposal is what differentiates it from previous discussions. It is also crucial for several reasons:
- Learning-by-doing (‘kaizen’ effect) is the single most important route to improving search quality (it is assumed that the service would also be run in an ‘open’ manner with all data and research diffused openly — at least as far as that is consistent with privacy etc). Hence, to develop good software you need to be actively used by a significant (and diverse) user population.
- Because of the ad-funded model it is perfectly possible for the search service to earn significant income (less, probably, than a closed service would do but the first mover and open-source ‘brand’ advantages should be sizable — look at Firefox). As a result running a complementary service could be a substantial source of development funds.
- From a societal standpoint the service is what you really care about. While it is true that other people could run a web service with the same code (exactly the point!) it would be nice to guarantee the existence of a service, particularly one closed associated (name/reputation) with the underlying software.
These are all important points from an implementation point of view but they say little about why this proposal is good from a viewpoint of social welfare. The answer though is obvious:
- We get de facto competition from a service which prices at ‘marginal cost’ (no rent extraction)
- the algorithms, software and accumulated data (by no means the least important of the 3!) are all open and therefore there are general efficiency gains
- As everything is open we have a guarantee of/commitment to future competition
- Extra competition from entry of other service providers using the same codebase (NB: license would want to be of a GPL+/Affero type so that service providers are obliged to contribute back code improvements)
- Greater transparency (we know why some sites come top, why some don’t).
Some remarks on funding: while in the long-term this ‘open-search’ system could well be self-supporting financially it seems likely that it would need a decent injection of up-front funding to get it going. Where would this money come from? Three possibilities: a) venture capital b) government or private donor c) the general web community. (a) seems improbable given the not-for-profit nature of the endeavour and the complete openness of both code and service (might be possible to modify not-for-profit to allow pay-back to backers but this is more like a straight loan and less like a VC setup). (c) is possible but also seems difficult — one would be looking for significant community support for development, testing, use and promotion but harder to see it as a reliable source of funds, at least initially. That leaves us with (b) up front funding from government or private charity. Given the large technological spillovers, the clear social benefits in promoting competition and transparency, and the ‘utility-like’ nature of search this would seem to be a promising route to take.
The Current Search Model: Existing Problems
Even without the development of a monopoly (or oligopoly) the current search situtation gives cause for concern, as Kamal Jain (Microsoft) detailed in his presentation on the ‘ugly side’ of the ad-funded search model at the Toulouse Roundtable (see below for more info on the Roundtable). In particular:
- Search engines are currently getting to extract all the rents from selling my attention (equivalently: we are selling our attention, and click-stream, for free). This situation does not appear to be sustainable. It is already being addressed in some ways — Firefox, for example, is getting money from Google for clicks — and there are plenty of other methods, for example, one could start a not-for-profit web proxy that auctions queries to search engines and then routes results to users passing money back to those users. Storing and selling your clickstream would be another method (this has been around a while but never really taken off — perhaps because privacy stuff becomes too obvious and not really clear what benefits will be).
- The funding model of search (in which search is funded by advertisers and free for users) may be societally inefficient. Search engines are in a monopolistic position vis-a-vis advertisers and therefore may a) overcharge advertisting (with those charges eventually passed back to users) b) over-provide search quality.
- Little transparency over search engine rankings which are of ever-increasing commercial and social importance.
Summary of Search Round-table
Hal Varian (Berkley and Google):
- Don’t need to worry about antitrust stuff as won’t get concentration since:
- No demand-size network effects
- Very low switching costs
- Differentiated users so room for several competing differentiated products
- OK there are economies of scale but not strong enough to overwhelm other effects [ed: as you can tell i'm not convinced on this score in the long run]
- Huge benefit to having an oversold vs. undersold auction in market for ads (compare to Klemperer on dutch 3g auction).
- online world: ‘kaizen’ — continuous improvement as cost is now so low (iphone takes 2 years to get to market
- => Learning by doing like never before (new advantages for incumbents)
- Marketing is the new finance (data + computers + models).
- Quantitative methods really work (game theory really work).
Michael Schwarz (Yahoo):
- advertising auctions
- massive number of products (each term, different times, term + user click history, …). Massively combinatorial auctions.
- Is attention a normal good (IMO: yes it is, but the ultimate scarce resource)
- collusive bidding by advertisers
Kamal Jain (Microsoft):
- The good, the bad and the (mainly) ugly of search
- Why are search engines able to extract such rents when there is quite a bit of competition and differences in quality (at least among top 3-5) are small
- Because they only charge one side: the advertisers. They therefore can overcharge (monopoly price) and over-provide quality.
- ed: real solution is to have some way for consumers to get paid — that way users get rents rather than search engines (e.g. Firefox with money for clicks, what about a search engine proxy — see my comments below for more on this).
- self-bidding: google take advertising slots on their own engine which pushes up bids (cf. Varian’s example above about discontinuity between over-sold and under-sold auctions). With self-bidding many of the standard results in auction theory no longer go through.
- google checkout: 20% discount for checkout users is a clever way to increase payments of non-participating advertisers (auction is second-price)
- incentives for search engines to conceal information that conflicts with their advertisers interests (example from Page and Brin 1998 about cell-phone radiation risk and cell-phone advertisers)
- decrease quality of commercial search side to force merchants to buy advertisements
- google pack ad-blocker does not block Google syndicate ads
- value of diversity: every search engine has bias/different results. As search engines become more important as gate-keepers one search engine means one single controller of our access to information. [ed: this really leads us towards regulating search as a traditional utility with a potential natural monopoly]
Steven Crossan (Google):
- Hal Varian said he saw lots of diversity
- But currently see Google with a dominant share (even more dominant in Europe)
- ‘kaizen’ affect mean that more users mean more opportunity to improve
- High fixed costs (but not that high)
- Not a lot of a barriers to entry in terms of ideas (IR is well known)
- BUT don’t do versioning just have continuous improvement and there are a lot of them so HARD to replicate on the ground (lots of engineers for a lot of time)
- So where would a challenge to a “monopolist of near-monopolist” like Google come from?
- Answer 1: radical/disruptive innovation (discontinuity in algorithm quality) — particularly since fixed costs of entry are low [ed: disputable to my mind as data centres are costly but ...]
- Answer 2: open source [ed: but this doesn't work here so much as you need to fund the service at least if you want to do 'kaizen']. “If I were microsoft I would be open-sourcing my search engine”.
- The search box as the new command line: “move 500 euros to my savings account”, “show me my wedding photos” [ed: this is google desktop and as he acknowledged pretty old hat -- natural language command line research has been around a long time but is just hard to do. Also though a single unified simplified interface might be a very good idea really don't want this interface (and all the apis to it) controlled by a single vendor -- should be open].
Francois Bourdoncle (Exalead):
- costs of setting up search engine are high: their system is 3 million lines of code (a complete operating system is ~16 million) and took 5-6 years to develop.
- not at the end of the road for entry into search but we are for advertising (just to hard to enter and exalead partner with yahoo for advertising stuff)
- pages indexed: google 18 billion, yahoo 14 billion, exalead 8, msn 7, ask 3.
- major ongoing (or coming) wars:
- telcos and isps vs. search engines (e.g. orange)
- media companies (e.g. Largardere)
- Advertisement companies (e.g. WPP)
- SEs becoming global media brands (e.g. Google News) — this will threaten traditional entities
- “Free software is a threat” [ed: i assume he means to his type of enterprise]