Google Search is facing competition that it has never before encountered. Will we soon replace ‘Googling’ by ‘Binging’, or are we better off with the status quo?
Microsoft has revealed an ace up its sleeve since it added an AI chatbot to Bing, taking the fight to Google’s flagship Search function. In response, Google execs hurriedly greenlit Project Magi in response, in the wake of adding Bard to its feature set. The latter is a clear competitor to the new version of Bing that has incorporated AI technology from OpenAI. Magi supposedly should become an AI-based search engine that, like Bing, features ChatGPT-like functionality.
To put it bluntly: internet search seems set to change fundamentally. Google has reason to be concerned for the first time in over twenty years. Indeed, it appears it very much is. The consequences of dropping too far back in the AI race could be grave, and the company has already received a shot across the bow from one of its partners. Samsung reportedly already threatened to ditch Google for Microsoft Bing as the default search engine on its smartphones. However, conflict between the Korean company and the Mountain View-based tech giant is all too common. In more concrete terms: if Samsung were to switch to Bing, it would rip up the shared revenue model it has set up with Google, with $3.5 billion at stake. Serious business then.
Google Search: dominant, seemingly with global approval
Google has led the way in the search engine market since the turn of the millennium. Its developers have been trying to discourage undesirable search results for over more than two decades. Low-quality articles and attempts to exploit the existing algorithm have been met by an ever-increasing amount of roadblocks. Google Search in its current state is a sophisticated product that is state-of-the-art in findability of online content. It has given Google the opportunity to promote its proprietary services to the top of the search results, with YouTube being a prime example. The practical monopoly has put Google in a powerful position when it comes to global information retrieval. However, we as its users appear to agree with the company’s dominance in Googling something over and over again.
There are some upsides to having one major player: the SEO (search engine optimization) market has been playing by largely undisputed house rules. Platforms like Facebook and TikTok are perpetually under scrutiny for the supposed promotion of harmful content. Google, in stark contrast, has not experienced that kind of scrutiny. The search engine is culturally ingrained to the point where ‘to google’ has become a verb.
Microsoft Bing AI (the New Bing) with GPT-4
We have extensively documented the impact that OpenAI’s GPT model has had here on Techzine. Of course, there is the ChatGPT chatbot, which you can ask questions to and which can take care of all sorts of things for you. It’s not at all flawless, and you may also wonder to what extent these chatbots can improve search results in the first place. If the output is correct, then great improvements in efficiency are possible. That, then, is immediately the main business case for what Large-Language Models (LLMs) have to offer. In theory, we can use them to perform many tasks that we don’t really want to do ourselves anymore.
Microsoft realized pretty quickly that it needed to invest in OpenAI and pumped the necessary billions into the AI startup. In return, it has since added OpenAI technology to a host of products. Office 365, to name one prominent example. Microsoft also embedded it in development (Github Copilot), security (Security Copilot) as well as Bing. In the case of the latter, Microsoft didn’t go for the Copilot addition. Instead, it introduced it as ‘The New Bing’. This immediately betrays the fact that in the foreseeable future, it will not be considered at all interesting whether OpenAI technology has been deployed or not. Because, well, if everyone switches, the AI-powered application will obviously just be known as Bing, period.
Users of Microsoft Bing have already been able to test the new version, which we will call Bing Chat here for convenience, for about a month and a half. Unsurprisingly, this caused some problems at first. For example, it initially had some noticeably unfriendly traits. Microsoft has taken the necessary measures to remedy this. You could call the AI function domesticated now. The difference compared to before is mainly in how search results are formulated. Users are shown full paragraphs of text in a conversational style rather than a more direct display of Bing’s supposed character traits.
For the time being, there are still quite a few ifs and buts built in. Microsoft has added all sorts of disclaimers, including ones about hallucinating, or making up facts, which AI chatbots tend to do. This, combined with the somewhat tighter reins and source citing, does not yet make the new Bing a substantially different search engine from the old variant.
Google Bard and Project Magi
Google’s timeline when it comes to combining search and generative AI models is a bit more ad hoc. Behind the scenes, of course, the company has been working on models such as LaMDA and PaLM for some time, yet it seems to be primarily in response to external factors that Google released and launched first Bard and now Project Magi. Project Magi is said to have been initiated in response to the consideration of Samsung we talked about above, for example. The unexpected launch of Bard a day before Microsoft was to launch the new Bing also had a high ad hoc content.
In a tech podcast from The New York Times, Pichai explains exactly what the AI chatbot Google Bard supposedly represents. While Bard doesn’t match ChatGPT’s convincing-sounding answers, the Google top executive says that’s not the intention either. In particular, he says the bot can inspire; from leading to new insights about a research topic to coming up with a nice idea for a gift. Those who take AI further, however, Pichai said, soon have a dangerous weapon on their hands. We need more international regulation, he argues.
Should we really want generative AI in search engines?
Quite apart from what Microsoft and Google are marketing, there is also a rather fundamental question to answer. Should we want an AI search engine to transcend the function of creative assistant? Something that presents information gathered from all sorts of random sources (whether neatly listed or not) as facts? We have now seen that the solutions from both Microsoft and Google can stray quite far from the truth.
What it ultimately comes down to is whether we trust an artificial intelligence for our gathering of information. This information gathering on the Internet has an inherent flaw: the nature of the Internet itself. Because it allows free communication traffic, any individual can produce any conceivable (un)truth. So a selection process must be made. Databricks, for example, wants to demonstrate with its AI model (Dolly 2.0) that the quality of the dataset is much more important than mere quantity.
For business, specific applications of AI can benefit from targeted datasets, trained by professional staff with verified information. Even a fairly small AI model can handle that. That makes the potential of generative AI huge in business. Especially when it comes to datasets that cannot be practically analysed by humans alone, it can come to the rescue. A mode like that can spot trends in biological data that lead to new scientific insights, for example. It can also enable customer service on a larger scale and predict stock shortages. Moreover, it can do the latter before a human would ever have noticed.
Such confidence in the data that search engines can handle is virtually impossible to obtain. The issue of online information delivery requires a very labor-intensive check of the data fed to a search engine AI. And who decides which information is accurate and which is not? In addition, within the world of search engines, things like SEO traditionally play a big role as well. It’s certainly not all about quality. There is plenty of junk on the first page of search results on a regular basis. Engineers work hard to prevent this, but they can’t seem to stop it from popping up. It is however dangerous to rely blindly on the moral compass of mega-corporations. The desire for more regulation expressed by Sundar Pichai is a case of “save us from ourselves.”
Finally, you can also ask a somewhat more sobering question. How different is a world in which both Google and Microsoft deploy an LLM to display search results than the current situation? Google’s explanation of its own search function is illuminating, but it remains superficial. In practice, the issue is not whether the search engine prefers the “right” information. It is about who classifies it as “right” and on what basis. YouTube’s algorithm is equally inscrutable and changeable, with an enormous impact on the behavior of content creators. The same is more or less true of LinkedIn’s algorithm (from Microsoft).
At the end of the day, (generative) AI may not impact the search results all that much anyway. It might be much ado about nothing. One might also ask the question whether AI is supposed to improve search functionality anyway. It’s probably much more about the output it provides. It enables you to not only search for the information you want, but it also presents it in a coherent form. Obviously, that information has to be correct, otherwise it’s still not of much use to us. We will all still need to do that final check. AI won’t do that for you.