Google Search Bias Case: The Difficulty is Proving or Disproving Bias

- Madhavi Singh and Ganesh Khemka

In Google’s “search bias” cases[1] involving allegations of abuse of dominant position the first question which arises (assuming Google’s dominance) is whether there can exist an objective definition of “relevance” or are ranking algorithms merely one of the many ways to rank results- a reflection of the uniqueness of various search algorithms and a legitimate and necessary mechanism of differentiating competitors.[2] If there were a single definition of “relevance”[3] or one correct way of ranking results according to their relevance then the subsequent question would be whether it is possible to assess if Google’s results have been ranked according to such relevance. This issue is analysed here and the essay looks at the difficulty in establishing either that results have not been ranked on the basis of relevance (by antitrust regulators) or vice versa, that results have been ranked on the basis of relevance (by Google).In other words, this essay attempts to analyse the difficulty of both proving and disproving bias.

The difficulty in proving bias in Google’s ranking arises due to the confidential nature of its algorithm. Google claims that it does not purposefully elevate its own vertical search engines and the reason why its own verticals frequently appear at the top of the results page is because they meet the objective criteria of “relevance” and are in fact, more relevant than the other results.[4] This is a problem of evidence where demotion of supposedly more relevant content cannot be proven[5] using any objectively verifiable evidence. The evidence often relied upon to prove bias is through comparison to other search engine results pages[6] which do not display Google’s own verticals as one of the top results. Such deviation from other search engines however, could be attributed to the difference in search algorithms employed by these engines to distinguish themselves from their competitors,[7]that is it could be a consequence of either a different understanding of “relevance”[8] by competing search engines or just a different way to assess the same relevance depending on the data they possess and the criteria they adopt.Therefore, for antitrust regulators to prove bias in Google’s ranking is extremely difficult.

Notwithstanding the arguments made above if competition authorities were to consider evidence from other search engines in ascertaining whether Google ranks results on the basis of relevance then for Google to rebut these arguments by proving the relevance of its results is an equally uphill task.[9] While the overwhelming number of times Google’s own verticals appear at the top of the results might indicate some bias,[10] such evidence is definitely not conclusive.[11] Given the confidential and complex nature of Google’s algorithm which renders it difficult for it to establish non-biased functioning, the only substantive argument made by Google to prove relevance of its search results is the argument that it would not rank its content on any basis other than relevance lest it lose consumers.[12]This is a tautological argument whose essential import is: if Google were to degrade its results then it wouldn’t be able to retain customers and remain dominant and therefore, the fact of its dominance shows that its results are relevant.[13]

This assertion is based on three assumptions: (i) users would be able to assess relevance; (ii) switching costs between search engines are zero or lesser than the harm suffered due to irrelevant results; and (iii) there are other search engines which users can switch to without appreciable depreciation in quality. These three assumptions are disproved in turn.

Assumption 1: Users of search engines would be able to assess whether the results are relevant. According to Google relevance of results in both the absolute and relative sense can be assessed. In the absolute sense consumers would become aware if they feel dissatisfied with the results and in the relative sense if there exist better search verticals, consumers would know about them through advertisements.[14] Therefore, Google has constant pressure to provide relevant results to prevent users from shifting to competitors.[15]

Such arguments seem to be ignorant of the nature of Google’s search functionality. The inability of users to assess relevance is because of two reasons: (i) difficulty in objectively measuring relevance;[16] and (ii) branding effect. If a search engine shows completely irrelevant results or in response to direct factual questions gives wrong answers then it would be possible to notice the degradation in quality.[17] However, in other cases the degradation in quality would not be noticeable.[18] The quality of the search results are dependent on: time, quantity of results available on a topic etc.[19] Therefore, it is mostly not possible to measure relevance objectively in absolute terms.[20] Even in relative terms often times it is not possible to inform consumers about the existence of alternative search engines due to high information costs[21] and status quo bias.[22] Further, Google through years of successful branding has ensured that consumers trust it[23] making it even more difficult to detect irrelevance of results shown on Google.

Assumption 2: Switching costs for consumers are zero or lesser than the cost incurred due to use of irrelevant results.Google has argued that since all search engines are freeswitching costs are zero.[24]Such a simplistic understanding of switching costs does not take into account the rarity of multi-homing[25](that is, studies show that not many consumers tend to switch from one search engine to another even when search engine services are free) and network effects[26] (a phenomenon discussed later in the article) which are both relevant to the calculation of switching costs. Assumption 3: There exist alternatives to Google which can be switched to without appreciable depreciation in quality. This assumption is fallacious because of high entry barriers in the form of network effects and positive feedback loops. Network effects exist where the utility of a service increases when more people subscribe to it.[27] Google being dominant attracts many users and collects their data. Such data collection allows it to improve its services through data localisation and customised results.[28] Thus, Google’s dominance perpetuates itself. Another manifestation of network effects is positive feedback loop which refers to the phenomenon in multi-sided markets where success on one side of the market also promotes success on the other side.[29]An increase in consumer base makes Google more attractive for advertisers who are now guaranteed both a wider audience and more targeted marketing using the large amounts of data collected from the customer base.[30] The additional revenue which such positive feedback loops generate increase funds for Google, allowing it to offer more free services, attract more consumers and collect more data.[31] Therefore, the existence of high entry barriers means that there aren’t alternatives which users can switch to.

Since, all the three assumptions above are false the tautological argument of Google’s dominance indicating its relevance cannot be accepted. Hence, both proving as well as disproving the statement that Google ranks its results on the basis of relevance is fraught with practical difficulties.

Given this difficulty of proof, perhaps the best way forward to evaluate existence of bias is the approach adopted by the European Commission of referring to specific search parameters within search algorithms. Reference may be made to the EC’s evaluation of “Panda” (Google’s search algorithm). It was shown that having “original content” (and demoting websites with copied content) as a primary parameter for Panda’s functioning meant that most competitor comparison shopping services would not be considered relevant.[32] In light of this, using Product Universal for Google Comparison Shopping would in effect exempt Google’s own vertical from the operation of Panda and the requirement of “original content” thereby biasing the search.[33] Given the problems associated with leading evidence to prove or disprove bias, competition authorities across the world while determining this question should look at the specific search parameters which an algorithm adopts to see whether their adoption and the weightage given to them in itself is indicative of bias.

* Singh and Khemka are V Year B.A. LL.B (Hons.) students at National Law School of India University, Bangalore. [1]In re: Matrimony.com v. Google 2018 SCC OnLine CCI 1; Case at. 39740, Google Search (Shopping) [There are similar cases in other jurisdictions but these two are of particular relevance here].

[2] Joshua D. Wright, Defining and Measuring Search Bias: Some Preliminary Evidence, 3 (2011).

[3]Adam Raff, Search, But You May Not Find, N.Y. Times, (27 December 2009), <https://www.nytimes.com/2009/12/28/opinion/28raff.html> accessed 24 May 2019.

[4] In re: Matrimony.com v. Google 2018 SCC OnLine CCI 1, ¶ 178.

[5] Wright (n 2) 10-11.

[6] Benjamin Edelman & Benjamin Lockwood, Measuring Bias in “Organic” Web Search (19 Jan. 2011) <http://www.benedelman.org/searchbias/> accessed 24 May 2019.

[7] Lisa Mays, The Consequences of Search Bias: How Application of the Essential Facilities Doctrine Remedies Google’s Unrestricted Monopoly on Search in the United States and Europe, 83 Geo. Wash. L. Rev. 721, 738-740 (2015).

[8] Wright (n 2) 3.

[9] This assessment is notwithstanding arguments on burden of proof, that is, whether the burden of proving bias initially is on the party alleging bias or Google (being the sole party in possession of information relating to its own algorithm that other parties do not have access to and hence, being in the unique position to disprove such bias). Further, even if the burden of proving bias is initially on the party alleging bias then at what stage does the burden shift to Google to rebut such allegations by actively producing evidence to prove relevance.

[10] Benjamin Edelman (n 6).

[11] Chris Sherman,Study: Bing More Biased than Google; Google not Behaving Anti-Competitively, Search Engine Land (3 November 2011) <https://searchengineland.com/study-bing-more-biased-than-google-google-not-behaving-anti-competitively-99774> accessed 24 May 2019.

[12] Robert H. Bork & J. Gregory Sidak, What does Chicago School teach about Internet Search and the Antitrust Treatment of Google? 8(4) J. Comp. L. &Ec 663, 664 (2012).

[13] Maurice E. Stucke & Ariel Ezrachi, When Competition Fails to Optimize Quality: A Look at Search Engines, 18 Yale J.L. & Tech. 70, 98-99 (2016).

[14] James D. Ratliff & Daniel L Rubinfield, Is there a Market for Organic Search Engine Results and can their Manipulation give rise to Antitrust Liability?, 10(3) J. Comp. L. & Econ. 517, 522-524 (2014).

[15]Geoffrey A. Manne& Joshua D. Wright, Google and the Limits of Antitrust: The Case against the Case against Google, 34 Harv. J. L. & Pub. Pol'y 171, 244 (2011).

[16] Mark R. Patterson, Google and Search-Engine Market Power, Harv. J. L. & Tech. Occasional Paper Series, 12-13 (July, 2013).

[17]Stucke (n 13) 98-99 (2016).

[18] Patterson (n 16) 12-13.

[19]Stucke (n 13) 98.

[20]Stucke (n 13) 98.

[21] Patterson (n 16)  24.

[22]Stucke (n 13) 104.

[23] Amy Gesenhues, Study: Top Reason a User Would Block a Site From a Search? Too Many Ads, Search Engine Land (April 15, 2013) <https://perma.cc/6P59-GF56> accessed 24 May 2019.

[24] Aaron S. Edlin; Robert G. Harris, The Role of Switching Costs in Antitrust Analysis: A Comparison of Microsoft and Google, 15 Yale J.L. & Tech. 169, 212 (2012).

[25] Case at. 39740, Google Search (Shopping) ¶ 221.

[26] In re: Matrimony.com v. Google 2018 SCC OnLine CCI 1, ¶ 199.

[27] Stan J. Liebowitz& Stephen E. Margolis, Network Externality: An Uncommon Tragedy, 8 J. Econ. Persp. 133, 135 (1994).

[28]Andrew Langford, gMonopoly: Does Search Bias Warrant Antitrust or Regulatory Intervention, 88 Ind. L.J. 1559, 1574 (2013).

[29] Kristine Laudadio Devine, Preserving Competition in Multi-Sided Innovative Markets: How Do You Solve a Problem Like Google, 10 N.C. J.L. & Tech. 59, 63 (2008).

[30]ibid.

[31]Devine (n 29).

[32] Case at. 39740, Google Search (Shopping) ¶ 358- 359.

[33] Case at. 39740, Google Search (Shopping) ¶ 408.