From the Canon: Robert Taylor’s Levels of Information Need
[This post was originally published on Medium]
Every so often, we’re taking a classic paper in information science and breaking it down into its argument, some clear examples, other perspectives from the field on the same issue, and what it can mean for an information architecture practice today. This week, we’re looking at Robert Taylor’s “Levels of Information Need.” [PDF]
I occasionally like to daydream about the very glamorous jobs I could have if information science was properly appreciated as a discipline — it usually involves something approaching Indiana Jones meets High Fidelity — but it still wouldn’t compare to Robert Taylor, who recruited Nazi and Soviet informants for US counterintelligence after WWII. I have no proof for this, but I like to imagine that this work informed his revolutionary contributions to information science (including giving it the name “information science”). My favorite work of his centers around how people form questions while seeking information, or “inquirers” as they “negotiate the question” through various “levels of information need”.
If you have no idea who Taylor is or why you should care him, know this: librarians are sitting on a repository of great research about how people seek out information, also known as theories of information behavior. These would be of great value to designers if we ever talked about them outside of library and information science programs (this is part of the reason why we started Known Item). If you are in the business of creating information experiences, familiarity with this canon can give you a huge leg up.
Negotiating the Question
“Negotiating the question” is Taylor’s name for the general process of figuring out what question to ask in order to solve your information need. He tells us that this negotiation usually takes one of two forms:
asking someone for help or
jumping right to self-help; trying to find an answer on your own by going directly to the information source that you think might have the answer (or more likely, the resource with which you are most familiar).
Prenegotiation decisions by the inquirer, from “Question-Negotiation and Information Seeking in Libraries”, Robert S. Taylor, 1967.
When we think of information needs in digital environments, we might assume that we have mostly moved away from Option A, asking someone for help, in lieu of Option B, self-help. After all, I can Google just about anything, right? So can’t we just incorporate a search tool into every information environment we create and call it good?
In many cases I do know exactly what I’m looking for (see known item searching) and for that, search tools are lovely.
But wait! How do I search for something that I don’t know the words for? How do I fill a gap in knowledge that I can feel, but can’t put into language? There will always be occasions when we fail so miserably at “negotiating the question” that we benefit asking someone or something for help. Even the best full-text search tool can’t read minds (yet). We still need tools to help people hone in on their information needs through a series of back-and-forth interactions that refine an initial thought down to the query that will retrieve a useful answer from a system.
Wait, this sounds familiar. Did Taylor know the chatbots were coming? Did he know the recommender systems were lurking? Friends, I believe he did.
I see places where Taylor’s negotiation principles are useful all the time, but I think we actually need them to design effective chatbots, conversational interfaces, and even recommendation frameworks. The number of services adopting conversational UIs as information retrieval mechanisms means we need to know how to design one that is actually helpful. In order to be helpful, we need to ask, “How do I design a conversational UI to answer queries effectively?”
This is a huge question, but Taylor’s levels of information need are here for us. The entire trend toward conversational UIs as reference tools is a hybrid of the two negotiation tactics that Taylor teased out in 1967 — ask an agent for help or search for yourself. So, let’s get real with some query negotiation.
Levels of Information Need
To understand how people successfully negotiate questions (i.e. figure out what they really want to ask), Taylor studied reference interviews — that’s right, he watched actual reference librarians help people in the library. Why? Because everybody knows that reference librarians are black-belts in maneuvering what Taylor describes as “a very subtle problem — how one person tries to find out what another person wants to know, when the latter cannot describe his need precisely.”
After observing these reference interviews, Taylor identified four levels of information need that inquirers went through in order to form questions that a system could actually answer:
(1) First you identify a visceral need, or “vague sort of dissatisfaction” that you might not be able to put into words.
For example, let’s pretend I’m the president of my HOA (kidding, I wish I was pretending, it’s actually a real choice that I made for god knows what reason). I’m leading a meeting. Cheryl often gets out of hand and starts bullying Karen. Stuff gets ugly. The meeting eventually ends. I leave in a stupor.
At this point I know I need to solve some problem, but I can’t even get a grip on what the problem is, let alone how to solve it. At this point it’s just an acute feeling of dissatisfaction.
(2) Next, you formulate a conscious need that may be expressed as “an ambiguous and rambling statement.”
My “vague dissatisfaction” starts to grow legs, especially as I try to describe it to my incredibly patient partner while I “think out loud.” If I were to do this with an agent who could pick out the elements of my ranting, they could help me identify that there are certain unpredictable aspects of my neighbor’s personality that make it hard to communicate with them, which is making me miserable.
(3) After you’re done rambling, you’re able to construct a formalized need, expressed as a “qualified and rational” statement.
With the help of an agent, at this point I can identify the problem I need to solve: “I need to learn how to communicate with my unpredictable-bully-of-a-neighbor and also keep her from blowing up the HOA into teeny tiny smithereens at every meeting.”
(4) Finally, you have a compromised need: a query you could actually put into a system and expect good results from.
At this point, a black-belt reference librarian (or heeey super-well-designed chatbot 👋 ) can help me translate my rational statement into a query that will get me some helpful results. It is worthwhile to note that “the compromised question is the information specialist’s business, the representation of the inquirer’s need within the constraints of the system and its files.”
THIS is where the power of the agent really shines: the art of translating my need into a query that will return helpful results from a set of resources. The agent should ask, “What does my system have that could help this person with this identified need?” and then show me those things.
In my case, a talented agent (librarian or otherwise) could take my rational statement of need and suss out what resources the system might have that could help me. Maybe they run a search on “engaging with high conflict personalities”, and another query like, “Robert’s Rules of Order” and hand me a pile of materials about those things. Now I can walk into the next HOA meeting with my head held high, armed with wisdom about managing high conflict personalities alongside the art of running a meeting, and for now at least, Cheryl is vanquished, Karen is protected, and I am a certified HOA Boss Lady.
Robert Taylor gave us “query negotiation” and four stages of information need that people go through in order to move from a vague feeling of discontent to a fully satisfied information need. This framework still matters today. It can inform our design of conversational UIs, or really any interaction where an agent is attempting to help a user hone in on their information need (recommendation systems could take cues here, too). An obvious application is that we can build rules into our conversational UI models that follow this framework, moving a user from a visceral need to a compromised need successfully, so that we can actually provide responses that are helpful, rather than just best guesses at what the user may need.
