Sunday 8 June 2014


Foraging in the World, Mind and Online




PETER TODD
Indiana University
Department of Psychological and Brain Sciences

VIDEO



OVERVIEW: How do we decide when to search for something better and when to stick with what we've got?  People, like other organisms, must adaptively trade off between exploring and exploiting their environment to obtain the resources they need.  This applies to whatever space they are searching: whether the external spatial world, looking for patches of food; the social environment, looking for mates or friends; the internal mental environment, looking for concepts in memory; or the online environment, looking for information on the Web. Common underlying mechanisms may be used to address the explore/exploit tradeoff in each of these domains.  People use similar heuristic strategies to decide when to keep looking and when to give up searching for resources in patches in space (e.g., for fish in a pond), in memory (e.g., for words in a category), and online (e.g., for useful Web pages), as predicted by optimal foraging theory.  Moreover, the connections between search in these domains may have deep evolutionary roots, built on the same underlying mechanisms, as indicated by studies showing that search in an external domain can prime subsequent search strategies in an internal domain.  In this talk, I will describe how new studies are uncovering these connections between spatial search and information search (as described in Cognitive Search: Evolution, Algorithms, and the Brain, Todd, Hills, and Robbins, eds.; MIT Press, 2012).

READINGS:
    Hills, T. T., Jones, M. N., & Todd, P. M. (2012). Optimal foraging in semantic memoryPsychological review119(2), 431.
    Hills, T.T.,     Todd, P.M., and Goldstone, R.L. (2008).  Search in external and internal spaces: Evidence for generalized cognitive search processes.  Psychological Science, 19(8), 802-808.
    Wilke, A., Todd, P.M., and Hutchinson, J.M.C. (2009).  Fishing for the right words: Decision rules for human foraging behavior in external and internal search tasks.  Cognitive Science, 33, 497-529.



31 comments:

  1. Dear Todd, Thank you very for this very interesting presentation ! I have a little question : How is it possible to study conscious/unconscious or implicit/explicit modalities of search? Frequently, when I wake up, or when I take some jogging, I find answers or solutions to complex/deep problems where I or my brain, or my body is searching. Descartes liked to think in his bed...

    ReplyDelete
    Replies
    1. Good question--so far we've only developed methods for looking at conscious search processes, though it could be possible adapt methods from studies on unconscious problem solving to look at them from a search perspective...(but the problem is that you'd only see the endpoint of the search, not the search path to get to that goal)

      Delete
  2. What would be the cost associated to exploration in the patch-switching experiment? For both experiments (fish and words), the exploration is represented by a click on a button that guarantee the subject to find another patch. In this context, the optimal solution would be to switch patches as often as possible because there is no possibility of failing to find one, or at least a patch as good as the previous one.

    ReplyDelete
    Replies
    1. I agree. In the case of naming animals, we probably take longer and longer to find new categories of animals because we are exhausting our options. This is quite different than the fishing experiment where all category switching takes the same amount of time.

      I wonder how long it would take for people to name only categories of animals. Would the response time compare to when people are naming animals and switch between animal categories?

      Delete
    2. The cost of exploration is the time it takes to get to a new patch--this is called "travel time" for animals actually traveling in space between patches, but in our experiments it is just a waiting time before the next patch is presented (15 or 25 seconds usually).

      Delete
  3. The semantic space structured hierarchically. Can we find in the time of switching between patches some hints to this hierarchical structure? I.e. if search requires to go one or more steps up the hierarchy and down again, will it take longer?

    ReplyDelete
    Replies
    1. That's a good idea. Right now we assume that when people leave one patch and search for another, they hop all the way up to the top level, e.g. they look for another common animal. But perhaps the time between different types of items could be used to reconstruct a semantic tree structure....

      Delete
  4. Why do we always assume that the user has *one* goal? Users will often follow information scents that match their interests rather than a specific goal.

    ReplyDelete
    Replies
    1. In this case I would argue that the searcher still has one goal. That goal being to expand their knowledge of the subject of interest. Although it is not a concrete goal, such as finding when Vin Deisel died, it is nonetheless a goal. Upon acquiring data throughout their search, their goal will change.

      If my goal is to have no goals. I still have a goal.

      Delete
    2. I would agree with you. We all have this experience when we go on the web to get information for work/school, and follow "relevant" links until we realize we're reading something wildly interesting, but completely irrelevant to the task we're supposed to tackle. I think there might be more that indulgence to it – it might be that our search heuristic exploits something like weighted interest.

      But of course, the goal then has to be formalized differently. So we might have a new model.

      Delete
    3. The typical approach in studying web search has been that people have a particular goal (query) in mind, and they search until they find a suitable answer to that. But you're right that people can search with other types of looser goals (like, entertainment for a half hour), or the goal can change as search progresses. This is a good area for further research.

      Delete
  5. What exactly are those decision mechanisms? Are decisions very specific to people or are they systematic? Have we found some regularity in animal and humans decision mechanisms?

    ReplyDelete
    Replies
    1. There are many types of decision mechanisms that have been proposed for search and other types of tasks--see our books! Different mechanisms are appropriate in different kinds of environments, and different people can also use different mechanisms at different times (often more than one mechanism could do a good job).

      Delete
  6. Great talk! The point discussed by Professor Todd that I found most interesting is the idea that the “extended mind” extends not only into literal three-dimensional space, but also into virtual space. What strikes me most about this is that both kinds of extension seem to work according to similar (or perhaps even identical) principles. My question for Professor Todd is whether or not he thinks that cognitive extension in the virtual sense is isomorphic to cognitive extension in literal space, or if there are differences between both kinds of cognitive extension. If there are important disanalogies between these kinds of extension, what do these consist in?

    ReplyDelete
    Replies
    1. I'm also interested in this question. I believe that our cognitive abilities and mind extend to anything we interact with. I don't see too much special about the internet or AI. Cars give us the ability to move faster and fire gives us the ability to cook quickly. Both of these add to our cognitive capacity (the things that we can do). The internet gives us the ability to do many new things and t do them quickly. However, I don't think speed and number merits the internet special status in regards to an extended mind.

      Delete
    2. Yes, another good question, and our work suggests that we face the same kind of search challenges that occur in different environments (e.g. internal and external) with the same kinds of mechanisms. Specifically, we appear to search through information spaces (memory, Web) with the same mechanisms that evolved to guide search in physical space.

      Delete
    3. These are incredibly interesting results. Thank you again for the great talk.

      Delete
    4. What I find interesting the analogue between searching memory for information and searching the web. When looking for information in memory, you have a vague sense or meta-memory as to whether or not the information is there. While you may only be able to recall semantically nearby information, this feeling or vague sense guides your search until you arrive at that information. Similarly, if I am searching the web for information on a topic for which I only have a vague sense, I use semantically near search terms that provide me with links that I can mine to guide me towards information on the precise concept for which I had a vague sense.

      Delete
    5. So, there appears to be an interplay between exploiting and exploring here. I can use an exploration to exploit a resource that opens up new exploration possibilities.

      Delete
  7. This reminded me of Dr. Goldstone's talk about learning along with others. How do you think imitation and collaboration influence exploration and exploitation time?

    ReplyDelete
    Replies
    1. Social search is a rich area on foraging theory--Prof. Luc-Alain Giraldeau here at UQAM is a world expert (and wrote a book on it!). He talks about producers--individuals who figure out how to get resources, e.g. by exploring and finding them--and scroungers--individuals who copy what others are doing, and get the benefits of exploiting without having to do the exploring. But even at this population level, an appropriate adaptive balance must be struck between explorers (producers) and exploiters (scroungers)--if everyone scrounges and nobody produces, there will be nothing to scrounge!

      Delete
  8. You said that we explore or exploit, is there any other ways for heuristic? Isn’t it any animal doing something else?

    ReplyDelete
    Replies
    1. Exploration and exploitation are two major categories of behavior, but of course an organism can do other things, like resting/sleeping/grooming, which wouldn't fit either of these categories.

      Delete
  9. Wonderful talk, thank you.
    My question: if I well understand there is a purpose to keep people more time in some patches in the web by providing more resource in the same area before switching. As application, can we use this behavior to put more advertising announcement in these hot patch (by hot patch a mean a place where people stay more time before leaving) ?

    ReplyDelete
    Replies
    1. Yes, it would make sense to target messages to users (including ads) in places where they spend more time, such as the more useful patches--but for different people, the patches they spend more time in could differ. So companies would want to target ads for particular items to likely customers who spend more time in particular web patches, if they can determine those search patterns.

      Delete
  10. Maybe I missed this in the talk, but when people "switch patches" in the web search, does this mean that they switched the web page they were looking at, by going back to their original search results and choosing another, or does switching patches mean that they reformulate their search by putting in different words?

    ReplyDelete
    Replies
    1. Good question, which I didn't really say--switching patches usually meant going back to the original search-engine results page and following a link to a new site, or using the search engine to do a *new* (modified) search to try to get something closer to the original query. Staying in a patch would mean following links from the current webpage to other nearby pages (which would often mean staying in the same site).

      Delete
  11. Would you describe the decision making step to be an intentional act or instinctual? Have you considered using data from digital standardized testing to measure internal knowledge search?

    ReplyDelete
    Replies
    1. I think switching from exploiting to exploring (leaving a patch) is often intentional and conscious ("hmm, I can't think of any more pets--what other kinds of animals do I know?"--though not stated that explicitly in one's mind!), but the mechanisms that trigger when to do this switch may well be instinctual (evolved). I'm not sure that data from standardized tests would help us identify search paths and processes, but rather just the starting and end points of search--but maybe you have other ideas about how to do this?

      Delete
  12. Great talk! I wonder if you've looked at the effects of modulating risks, and modulating reward values on time to giving up. If so, do you see people calibrating their behavior at the next fishing hole in light of experienced risk and rewards? If not, do you have any idea why not?

    ReplyDelete
    Replies
    1. We haven't looked at modulating the risks/benefits associated with finding particular resources (exploiting), but we have varied the costs associated with exploring, by changing the travel time between patches (e.g. from 15 to 25 seconds)--and people respond appropriately to this change, by staying longer in patches if it's going to take longer to travel to the next one.

      Delete