Tag Archives: exploration vs. exploitation

Exploring at the edges of what we know

March 11, 2015 By in associative cuing, control dialing, detail stepping, dynamic environments, making and finding Tags: , , , , ,

Sometimes when we are exploring for ideas or information online, using a search engine, we have a general sense of what we’re looking for—but we can’t put it precisely into words. Yet, we would readily recognize promising outcomes or directions if we saw them.

Some of our online searching goals are more open ended and multifaceted. Here, getting an answer quickly is not our top priority. We’d rather embark on a somewhat slower search that got us closer to where we ultimately would like to be. The journey itself is part of the learning. We make and find as we go along, with each step providing us with new pathways.

How might our search tools themselves better enable us to truly explore? What if our search tools allowed us to fluidly and rapidly express our changing sense of where we really wanted to go?

One recent example that actually registers and iteratively acts upon our search intent in an interactive fashion—repeatedly inviting our feedback—is called SciNet. Imagine you have a research question about gestures. You enter the search term “gestures” and, on a radar-like circular screen, you are presented with a range of alternative topics—a number of which you might not even have thought of, say, “immersive environment” or “accelerometer.” Suppose further, that you can then move those topics about on the screen. You can pull the most relevant topics into the center of the radar screen. Suggestions that seem more peripheral for your purposes, you can move away closer to the outer edge of the circular radar-like display. The system dynamically responds in real time with new suggestions as your expressed interests change.

Such “interactive intent” search has been shown in a study, using SciNet, to provide significantly improved quality of retrieved information, allowing users to access both more relevant and more novel information in an efficient way. The search tool allows us to deeply tunnel into a meaning space that is already familiar to us (exploitation) but also offers support for experimental forays into the currently less well known (exploration). In the words of the system’s developers: “The model and its environment (the user) form an online loop, and learning involves finding a balance between exploration (showing items from uncharted information space for feedback) and exploitation (showing items most likely to be relevant, given the current user intent model).”

This interactive visualization allows the searcher to capitalize on their natural ability to rapidly and largely effortlessly recognize—rather than recall from their memory—relevant information. With this visualization we can rapidly adjust where we are on our “cognitive control dial” as we cycle through moments of automatic recognition and more deliberate evaluation and goal setting. The interactive visual display maps to both our visual and motor capabilities—allowing rapid updating of our search intent without costly sidetracking of our thinking. In this way, the boundary line between what’s “inside” and what’s “outside” in our thinking/meaning space becomes more permeable and more fully integrated with our unfolding thought processes.

Developing such cognitively friendly and fluid interfaces for structuring and guiding our exploratory idea search and experimentation are examples of what we broadly call thinking scaffoldings. As we explain in Innovating Minds, thinking scaffoldings are a way of productively guiding our perception-action cycles. They are intentional queryings and quarryings of our idea landscapes that are meant to help bootstrap (that is, “scaffold”) our idea generation processes. Thinking scaffoldings include not only databases or tools for extracting and identifying promising ideas or directions but also many other modes of scaffolding our idea generation processes such as adopting design heuristics, engaging in reflective verbalization, and drawing on tools for analogical or biomimetic search.

Thinking scaffoldings assist us to transition and keep moving across ideas, prodding us to re-categorize and shake-up or unsettle creative objects or their configurations. They help us to see things we could try or attempt—without an assurance that what we are trying will work.  They prompt us to test and revise, look and revise, and test again.

 

—> For more on exploratory online search see:

Dorota Glowacka, Tuukka Ruotsalo, Ksenia Konuyshkova, Kumaripaba Athukorala, Samuel Kaski, & Giulio Jacucci. (2013) Directing exploratory search: Reinforcement learning from user interactions with keywords. Proceedings of the 2013 International Conference on Intelligent User Interfaces, pp. 117-128.

Gary Marchionini (2006) Exploratory search: From finding to understanding. Communications of the ACM, 49(4), pp. 41-46.

Tuukka Ruotsalo, Giulio Jacucci, Petri Myllymäki, & Samuel Kaski (2015) Interactive intent modeling: Information discovery beyond search. Communications of the ACM, 58 (1), pp. 86-92.

Learning to vary: An overlooked avenue to mental flexibility and innovation

December 13, 2014 By in creative constraints, exploration, idea landscapes Tags: , , ,

It’s easy to repeat. But, we can also ask ourselves to not repeat––and reward ourselves for deliberately varying. Although little recognized, rewarding variability is a powerful shaper of creativity and innovation.

As we will see in Part 4 of our book Innovating Minds:

“Deliberately varying our actions helps to bring different sets of thoughts and procedures close together in time and space within our individual and group idea landscapes. This, in turn, allows us to combine and reconfigure aspects of ideas and ways of doing things to make novel combinations. . . . It is not always an entirely new approach that is needed. Sometimes “repeating with a difference” frees us to see new options.”

Whether shy or bold, lab animals that were rewarded for interacting in different ways with new objects later explored more widely. Trained dolphins, too, that were rewarded for varying showed newly emerging novel behaviors that had never before been seen in dolphins.

In our own creative endeavors we can also prompt ourselves to do things differently within constraints. Some questions we can ask:

How can we better learn to (appropriately) “reinforce variability” in ourselves, and in others?

How might we structure our physical, symbolic, and technological environments to better support “useful” experimentation and variation?

Do we too strongly emphasize minor variability in what we already know and do well, with mostly “known” but smaller rewards (sometimes called “exploitation”)? Do our attempts at minor variations come at the cost of more far-afield, novel, and bold exploration that is more risky and uncertain––but also potentially yields much larger rewards and creative breakthroughs?

What might be some of the cognitive processes that underlie the demonstrated benefits of reinforcing variability? That is: What’s being learned when variability is reinforced? What cognitive and perceptual processes (besides motivational ones) might contribute to the observed effects?

 

–>To further explore routes to greater creative/productive variability in behavior see:

Wilma Koutstaal (2012) The Agile Mind [Learning to vary versus learning to repeat, in chapter 5]  (New York, NY: Oxford University Press) pp. 220-233.

Patricia Stokes (2001). Variability, constraints, and creativity: Shedding light on Claude Monet. American Psychologist, 56 pp. 355-359.

Alison Weiss & Allen Neuringer (2012). Reinforced variability enhances object exploration in shy and bold rats. Physiology & Behavior, 107 pp. 451–457.