What is it like to compete against Google?
Day 7 and 8: DuckDuckGo CEO takes the stand as controversy over trial's confidentiality continues
A common refrain in American antitrust law is that antitrust laws are concerned with protecting competition but not with protecting competitors. That means the case against Google isn’t really about proving harm to Bing or DuckDuckGo — the government has to prove that Google prevented its competitors from having a chance to fairly compete in the first place.
Making that case still requires showing what it’s like for another search engine to try to go up against Google. Today we got our first significant insight into that struggle as DOJ called to the stand DuckDuckGo founder and CEO Gabriel Weinberg.
DOJ’s direct examination of Weinberg began with a basic introduction of what DuckDuckGo is and how it distinguishes itself from other search engines. Weinberg explained that DuckDuckGo is a privacy-focused search engine and internet browser company he started in 2008. Today, it employs 250 employees and receives about 100 million searches per day across the globe. That might sound like a lot, but it’s not much in the world of search — according to Weinberg, DuckDuckGo’s market share of search in the US is 2.5% (and even smaller throughout most of Europe).
DuckDuckGo’s “far and away” biggest competitor? Google, of course.
Weinberg testified about a number of ways in which DuckDuckGo has struggled to compete against Google and its default agreements with browsers.
First, Weinberg described the difficulty users face in switching their default to DuckDuckGo away from Google, which he said is “way harder than it needs to be.” He added: “If you switch some of these defaults eventually you’re just going to be switched back to Google if you do nothing.”
Next, Weinberg testified about the attempts DuckDuckGo made over the course of several years to pitch browsers on making DuckDuckGo the default search engine for their private browsing mode. “We thought it was a pretty great pitch to browsers,” Weinberg said.
But Weinberg explained that DuckDuckGo couldn’t penetrate the default agreements Google already had in place with browsers that prevented the browsers from using a different default for private browsing. DuckDuckGo’s pitch to browsers turned out to be a “quixotic exercise.”
Much of Weinberg’s testimony was sealed, but when court re-opened, he faced an aggressive cross-examination from counsel for Google who pointed out that DuckDuckGo’s market share is lower in Europe than it is in the United States, even in countries that have implemented a default search engine choice screen.
Weinberg testified that he believed choice screens in Europe were designed in ways that did not allow users to make a great choice, but that he thought it would be possible to implement a more effective choice screen. (If you’re interested in reading more about choice screens, Matt has previously written about the implementation of choice screens in Russia as compared to the rest of Europe.)
What happened yesterday
The biggest story from yesterday was probably what didn’t happen: we didn’t get any update on whether the admitted exhibits DOJ had been publicly posting on its website would go back up.1
If you missed what happened at the end of court on Tuesday, the very short of it is that Bloomberg reporter Leah Nylen stood up from the gallery to object to DOJ taking admitted exhibits off its website after Google complained about them being “picked up far and wide.”
Nylen returned to the courtroom yesterday morning — with a First Amendment lawyer by her side — but the court ended up starting the day in a closed session. This happened after Judge Mehta reportedly held a private meeting with the party’s lead attorneys in his chambers. When court re-opened to the public after lunch, Judge Mehta never addressed the issue of DOJ publicly posting admitted exhibits, which Google was reportedly doing too.
As for what did happen: we got a little more than an hour of open-court testimony from former Google engineer Eric Lehman and an argument over the admissibility of an exhibit that Judge Mehta said may be embarrassing to Google.2 If you want more details, I summarized these developments on this Twitter thread.
Google scores some points
Before Weinberg took the stand today, we heard more testimony from Lehman — mostly through cross-examination from a Google attorney. This questioning sought to highlight the ways in which Google has innovated two large language models (LLMs) — BERT in 2018 and MUM in 2021 — which according to Lehman have diminished the importance of user data in generating quality search results since the models are able to learn from text alone.
This testimony likely goes to the key question of what makes Google great: innovation or user data. Earlier in the trial, DOJ had put forward compelling evidence that Google publicly downplayed the importance of user data in its success. The testimony from Lehman today doesn’t necessarily contradict the DOJ’s theory that user data has been critical to Google search’s dominance, but it shows how innovation has also been part of the story — and why user data may be becoming less important to search results over time.
DOJ’s attempt to impeach Lehman’s testimony also seemed to backfire. In response to a DOJ question about whether Google had an advantage in using BERT over competition because of its user data, Lehman testified that Google’s “biggest advantage in using BERT” over its competitors was that Google invented BERT. DOJ then put up an exhibit titled “Bullet points for presentation to Sundar.” One of the bullets on this exhibit said the following (according to my notes): “Any competitor can use BERT or similar technologies. Fortunately, our training data gives us a head-start. We have the opportunity to maintain and extend our lead by fully using the training data with BERT and serving it to our users…”
This likely would have been an effective impeachment of Lehman if “training data” meant some kind of user data. But after DOJ concluded its re-direct examination, Judge Mehta asked Lehman what “training data” referred to. Lehman explained it was different from user search data. This exchange felt like a victory for Google after DOJ tried to throw a punch that missed the mark.
Apple executive John Giannandrea takes the stand
At the very end of the day, DOJ called Apple’s artificial intelligence chief John Giannandrea, who reports directly to Tim Cook. We only heard a few minutes of testimony from Giannandrea before he was dismissed for the day, but he is expected to continue testifying for most of tomorrow.
Counsel for Apple appeared in court to argue with DOJ about how much of Giannandrea’s testimony should be sealed. Apple and DOJ didn’t reach a resolution in court and said they would discuss it overnight — so it is still unclear the extent to which Giannandrea’s testimony tomorrow will be in open court.
I’ll be there tomorrow to report on whatever I can.
We do have a little more clarity on that now from the end of today: DOJ is finalizing a proposal that it will send to Google tonight for how to go about publicly posting admitted exhibits going forward.
This apparently embarrassing document for Google triggered another argument at the end of court today. Judge Mehta ruled to admit the document yesterday and said it didn’t seem to contain confidential information — but the context for the document was provided through testimony that occurred in a closed session. Today, the DOJ moved to unseal that redacted testimony about the document. Google vigorously opposed. Judge Mehta said he would consider it overnight and rule tomorrow.
It is not easy to distinguish the relative importance of user data over innovation in Google's success. As Hal Varian, Google's chief economist once commented, ‘every action a user performs is considered a signal to be analyzed and fed back into the system.’ It is the innovative use of such user signals that have improved search results. Saying that Google gained advantage from a larger sample size of user data is questionable. It is quite possible that similar innovations could be gained from smaller collections of user data, such as those available to other Search engines such as DuckDuckGo.
LLMs such as Bert are trained with scraped data, but refined later with user data (e.g. analyzing feedback on results produced).
What is more pertinent is the line of questioning examining the default access to Google search within user devices. Why would any average user change the default search engine (Google) if it does the job. Even if the ability to change search engines is easily there, why bother?
Let's not forget little Neeva! Who are they? If you know, congratulations, I'll bet you miss them too. If you don't know who they were I have only one comment. "Exactly" and that's the point.