Chat-Based Search and the Reference Interview
The new Bing is not what it could be.
I'm Tim Gorichanaz, and this is Ports, a newsletter about design and ethics. You'll find this week’s article below, followed by Ports of Call, links to things I've been reading and pondering this week.
We are entering a new era of online search. Microsoft is piloting an updated Bing search engine that incorporates ChatGPT, answering people’s queries in a conversational format. Google recently announced Bard, a similar product, and others are springing up such as You.com’s YouChat.
Chat-based search may become more powerful than any online search system we’ve seen before. We’re used to searching for articles on cat behavior or specific facts such as Timothée Chalamet’s height; but the new Bing, for example, promises to help plan menus, create vacation itineraries with specified constraints, and much more.
What do these new products mean for the future of search?
First, let’s acknowledge that it’s about time for innovation in search. Since the rise of Google and its PageRank algorithm, search has only seen incremental evolution, and every competitor has attempted to do the same sort of thing. Moreover, in recent years, the advertising model that commercial search runs on has come into question, to say the least.
So chat-based search is an exciting new possibility. It's not ready for prime time yet (the new Bing still makes dealbreaker factual errors and sometimes slips into the bizarre), but it has promise.
To me, the source of this promise is the way it resonates with people’s information seeking preferences.
One of the most robust findings in the field of information seeking is that people generally prefer to get information from other people. This is the same reason word-of-mouth advertising is so effective—and what’s given rise to influencer marketing. Other people are fallible, but we still tend to believe each other. Last year we saw how chatbots such as Google’s Lambda evoke the feeling that one is talking to a fellow human, and people are reporting similar experiences with the new Bing.
After other people, online search is the generally next most preferred way of getting information. Google has become the gateway to the world of information, for better or worse.
So, at best, this emerging paradigm of search will combine the best of the top two information sources: people and Google.
Imagine: Online search could become more like a reference interview, which is the name for the structured conversation between a librarian and patron intended to connect with patron with the best information possible. The reference interview is one of the classic jobs of a librarian, and best practices have been developed through research over the past 150 years. Generally, a reference interview involves the librarian asking for an overview of the problem, gathering additional information about the problem and helping to pinpoint the patron's exact question, and eventually providing information (perhaps multiple sources representing diverse viewpoints) or answers in another form. It would be wonderful and helpful if a search engine could help a user refine their query and ultimately provide them better information.
But by all appearances, the new Bing and similar products have not been modeled after the reference interview. It seems the goal is, rather, to provide an answer in one shot whenever possible—but every librarian knows that patrons don't always ask for what they really need or want. People sometimes need help formulating the right question. And when a user is in “Chat” mode with Bing, a mode that would seem to be conducive to a reference interview–style conversation, Bing quickly loses the thread of the conversation.
There are other pitfalls, too, of course. For one, if Bing and others include advertising (rather than, say, running by subscription), we can imagine a future of covertly sponsored conversations and a whole new world of SEO tactics.
More deeply, it’s clear that the large language models behind products like ChatGPT and the new Bing lack common sense and human context—to say nothing of a deeply layered understanding of what relevance means in a given search task. And that’s especially important if search engines are to be useful for retrieving complex answers. This is why critics such as Gary Marcus say the currently-vogue deep learning approach to AI is ultimately too limited for what we want AI to do.
We might remember here that Google's original PageRank algorithm benefited from decades of research in library and information science on citation analysis, and these research-backed insights made Google far and away better than its competitors. As we move into a new age for online search, why not benefit again from the wealth of research done on reference interviewing?
But that’s a question for the teams at Microsoft and Google, as well as search entrepreneurs. For the rest of us, the lesson for now is to proceed with caution. And maybe go to your local library if you’re looking for a trustworthy chat-based search.
Ports of Call
Article: Apropos of a previous post here on Ports, the Financial Times asks, “Whatever happened to the metaverse?”
Another Substack Newsletter: Does the new Bing pose a global threat? Erik Hoel thinks so.
Board game: Board games can be fun (I love them but your mileage may vary), but can they also teach us things? Can they unlock hope and optimism? The creators of Pandemic, which is a wonderful game but has felt too on-the-nose for me to play since March 2020, have a new game coming out that intends to help people build their creativity in response to the climate crisis. Read about it in this Guardian article.