Attention-enhancing information retrieval

Last week I was at SWIRL, the occasional talkshop on the future of information retrieval. To me the most important of the presentations was Diane Kelly's "Rage against the Machine Learning", in which she observed the way information retrieval currently works has changed the way people think. In particular, she proposed that the combination of short query with snippet response has reworked peoples' plastic brains to focus on working memory, and forgo the processing of information required for it to lay its tracks down in our long term memory. In short, it makes us transactionally adept, but stops us from learning.

Unfortunately, Diane's observation did not make it into one of the half-dozen "big tasks" in the future of IR. (There was a session on information literacy and "smart IR" that I thought was heading in that direction but ended up being about e-learning.) In fact, while there was the obligatory talk of the importance of "understanding the user", there was scarcely any talk (aside from Diane) about trying to understand the user's---our---true predicament in the contemporary information age---harassed, distracted, incessantly processing disparate information with little of it sinking in to us and little of us permeating into it. (How often do you end a day of work, or an evening of web surfing, with little recollection of what you actually did, saw, or thought?)

Information retrieval scientists think of the user like lab scientists think of rats (Diane's example), observing their reactions to various stimuli; or at best as horse trainers think of horses, figuring out how to optimize their performance. In other words, they are viewed from the outside, through their behaviour. It is rare to think about matters from the user's---the sentient, autonomous human's---perspective. We only feel we are being serious if we approach the user through observational, repeatable, and measurable, studies. That, I think, is misguided: understanding the user should involve understanding ourselves. We need less information science, and more information humanities. That is why my proposed catch-cry for the next forty years of information retrieval talkfests is "fewer user studies!".

As if to underline the inability to think of the user as an autonomous, sentient being, rather than as a repository for stimuli and behaviours---the failure, in Kantian terms, to think of people as ends, rather than mean---one of the grand goals for the future of IR that did get up at SWIRL was zero-query and less-than-zero-query search---the notion that it is not you that should be querying the retrieval system, but the retrieval system that should proactively be telling you what to do. Added to all the distractions that we're currently subjected to from other human-initiated sources---emails, phone calls, instant messages---would be carefully personalized machine-generated prompts. Various fanciful scenarios were given, but the ultimate end-point of such a research direction is that you walk into the shopping mall, and then your mobile phone leads you round telling you what to buy.

When I suggested this at the meeting, various people said "this is just an information tool, it is up to you what you do with it" and "you underestimate people, they are not stupid"; but the reality of it is that the technology we have today has shaped our way of thinking and behaving, without our choosing it. If a convincing, well-tuned, sympathetic machine learning algorithm starts directing people what to do, and directs people to do things that they get (short-term) satisfaction from, then people will initially choose, and then gradually learn, to let their mobile phone tell them what to do.

Edit, 2012-02-27: fixed Dianne to Diane

10 Responses to “Attention-enhancing information retrieval”

  1. [...] Attention-enhancing information retrieval [...]

  2. Neelan Nair says:

    We are already seeing how humans are bending their own behaviour for search engines. The query terms used are far from what the actual query may be - in fact people think in terms of keywords.

    I guess the key point here is that Machine Learning is normally used on data which is produced irrespective of the result of the ML algorithm used. In web search we are now seeing humans modifying their habits to better use a system which is modifying itself according to human habits.

    Interesting post.

  3. Michael B. says:

    I think that this book: The Filter Bubble promotes a similar hypothesis: search engine/social network algorithms significantly limit and shape the information we consume and have a yet under-appreciated impact on our society.

  4. william says:

    Michael B.: thanks for the book recommendation, will have a look.

  5. Stefano says:

    Interesting post and comments. If I can I add some "history" bits:
    - There's also a TED presentation by Eli Parisier on filter bubbles:
    http://www.ted.com/talks/eli_pariser_beware_online_filter_bubbles.html
    - And Nicholas Negroponte wrote already in the 90s, in "Being Digital" about the "Daily me", a hypothetical (?) newspaper that would give me only the news that I want to see.
    - However, what's different in search engines and "less-than-zero-query" from, for example, writing, books, abacus, clock, and so on? I mean, the human history is full of these "cognitive prostheses" that help memory, reasoning, etc. Don't take me wrong, I completely agree that these are important issues and that we as IR community should better understand them, I'm just trying... to better understand 🙂

    BTW, it's Diane not Dianne.

  6. Vanessa Murdock says:

    Actually, there was one "big six" task related to people whose report was crafted by Diane, Nick, and others. In addition there was a "little 30" task about user engagement which seeks to understand people's patterns of attention on web pages, and in apps. Granted there can never be too much focus on people in information retrieval, but Diane's very excellent talk was indeed represented in the final "product" of the workshop. Information Retrieval is so broad, and there is so much still to do, that no one area can be given its due in a workshop of this nature. I don't think that is because people fail to see the importance of one area or another - it is rather that we can only do so much with the time we have.

  7. william says:

    Stefano,

    Hi! Thanks for the link to the TED talk. The filter bubble questions relates to another interestng topic coming out of SWIRL, which I will write up shortly, which is understanding and controlling our self-representation on the web.

    I do think there's a qualitative difference between passive extensions to memory, and active prompters that initiate our actions. The former modify capacity, while the latter modify agency.

  8. william says:

    Vanessa,

    Hi! Thanks for pointing out Diane and Nick's big-six task; I'm guessing this is "understanding people to improve IR". I'll read it with interest when it's released.

  9. [...] with forty or so of my most valued and esteemed colleagues and friends, I attended SWIRL 2012, “the occasional talkshop on the future of information retrieval”, hosted by RMIT in Lorne, near Melbourne, earlier this month. Swirlers at [...]

  10. jeremy says:

    How did I miss this blog post the first time around. Just came across it today.

    Well put. I agree with your main premise: "Understanding the user should involve understanding ourselves. We need less information science, and more information humanities."

    Personally, I am of the opinion that the purpose of technology is to get out of the way.. to become more humanities-oriented because we now no longer have to spend time... engaging with technology. Technology does its job best when we don't actually have to use it that often.

    But many in our research community take the opposite view. Sure, systems are optimized so that any one information need is satisfied with the least amount of effort (and the quickest), which means working on things like query speed and zero-query "push" search. But the overall "KPI" is measured not by a single query, but by engagement over time. The more the user comes back and runs another query, the more successful the system is assumed to be. Why? Because the user liked their experience so much the first time, that they kept coming back for more.

    On the surface that sounds great. But think a little deeper, and it's Diane's rat+stimulus situation. The more the rat is stimulated, the better things are assumed to be. But what I as a user want out of a technology is not that I get more pleasure in using it more and more, but that the more that I do use it, the less I actually need to use it. Not because it is bad. But because it is so good, because it satisfies my information need with something that is so filling, that I don't have to come back to it for days, to get another quick fix. It frees up my time to engage in other, more humanities-oriented pursuits, which is supposed to be its purpose in the first place.

    Unfortunately, we can't seem to tell the difference between users not using a system because the system is really bad, vs. users not using the system because it is really good, and they no longer need to. Compound that with the issue that much of what we build is advertising-driven, and there is a clear economic disincentive to build systems so good that users actually need to use them less, because they are left feeling satisfied for longer.

    And the whole dream of "information humanities" goes up in a puff of smoke.

Leave a Reply