SCN: Foraging

[Steve]

x-no-archive: yes  

=======================  

Why Web designers could soon be asking anthropologists for 
advice   

(Rachel Chalmers, New Scientist)---Which of these activities 
occupies more of your time: foraging for food or surfing the Web? 
Probably the latter. We're all informavores now, hunting down and 
consuming data as our ancestors once sought woolly mammoths 
and witchetty grubs. You may even buy your groceries online.   

But in an odd sort of way, Internet shopping has brought us full 
circle. According to researchers in the US, the strategies you use 
when you surf the Web are exactly the same as the ones hunter-
gatherers used to find food. You may be plugged into the 
information superhighway, but deep down you're still a caveman.   

At least that's the opinion of two researchers at Xerox's Palo Alto 
Research Center in California. Peter Pirolli and Stuart Card are using 
foraging theories from ecology and anthropology to understand how 
people find information in data-rich environments such as the 
Internet. They believe Web surfers rely on prehistoric instincts to 
maximise their yield when they hunt and gather morsels of 
information. If they're right, their results could help others design 
websites and search tools that are as alluring to informavores as 
flowers are to bees.   

Biologists came up with foraging theory in the 1970s as a way of 
explaining some puzzling aspects of animal behaviour. A hungry 
fox, for example, might have the choice between chasing a big, juicy 
rabbit or a tiny vole. Which should it choose? Foraging theory can 
decide. It states that as far as possible, animals make choices that 
maximise their "benefit per unit cost". In other words, they'll expend 
food-gathering energy in ways that yield the best energy returns. 
The rabbit might have a high energy value, but it costs a lot to catch. 
The vole is much easier prey.   

This cost-benefit analysis is complicated by the fact that food 
resources aren't evenly distributed around the world. They're 
patchy. The longer a forager exploits one patch, the lower the 
returns will be, until the patch is overgrazed and worthless. But time 
spent searching for a new patch is unprofitable--there's nothing to 
gain from a sterile space. So when is the best time to start looking 
for a new patch?   

It turns out that the optimal strategy is to move on when the rate of 
return from a particular patch falls below the average rate over the 
whole region. This is the marginal value theorem, a cornerstone of 
foraging theory formulated by the University of New Mexico 
biologist Eric Charnov in 1976. And it doesn't just apply to animals. 
The theorem has been widely used in anthropology to explain all 
sorts of human behaviours, from food preferences to patterns of 
land tenure.   

Pirolli and Card now believe the same idea can be used to 
understand information foraging. Imagine you're a financial analyst 
looking for data about an investment company. You've found a 
useful site on the Web, but it's starting to feel a bit stale. You'd like 
to move on, but you know that a search will take time and there's no 
guarantee that other sites will be any more useful. When should you 
abandon the dwindling supply? This, Pirolli and Card argue, is 
analogous to the problem faced by hunter-gatherers. And it can be 
solved in the same way.   

The first inkling that this was the case came in 1992. Pirolli and Card 
were studying the relationship between humans and information, 
looking for a theory that explained how people performing data- 
intensive tasks decide where and how to look for data. They had 
already conducted what they call "quick and dirty" field studies of 
information-gathering behaviour, one on a group of MBA students 
and another on the author of a business newsletter.   

Pirolli knew something of foraging theory, and he quickly noticed a 
correlation between the studies' findings and the behaviour you'd 
expect from animals searching for food. Like hungry foxes, 
information foragers try to maximise their benefit per unit cost--in 
this case, "benefit" meaning the relevance of the information and 
"cost" the time it takes to find it. They are also likely to move on 
from an information resource when it no longer yields a better-than-
average return.   

It was a satisfying analogy, but they needed empirical findings to 
back it up. So they designed a computer model that obeyed the rules 
of optimal foraging theory and set it to work looking for information.   

The latest incarnation of Pirolli and Card's artificial forager is based 
on ACT, a theory of cognition developed by Carnegie Mellon 
computer scientist John Anderson. ACT stands for both Adaptive 
Character of Thought and Atomic Components of Thought, and is well 
suited to the research because it possesses human-like conceptual 
and problem-solving skills--things an information forager needs in 
abundance. On top of these, Pirolli and Card programmed in the 
rules of optimal foraging theory.   

To test whether the theory produces useful results, Pirolli and Card 
set their model to work looking for information on a database. The 
database they chose was the IR Test Collection, one of the ultimate 
challenges in information science. It's a huge reservoir of texts from 
The Wall Street Journal, the Financial Times, the San Jose Mercury-
News, the Associated Press newswire, the Department of Energy, 
the Federal Register, the US Patent Office, computer publisher Ziff-
Davis and a handful of sources in Japanese, Spanish and Chinese. 
It contains more than a million documents.   

Pirolli and Card pinpointed target documents in the IR Test 
Collection and worked out the most efficient strategies for retrieving 
them. For this, they used an information retrieval system called 
Scatter/Gather designed for sifting through large databases. 
Scatter/Gather assigns each document to one of 10 groups 
according to its content, so documents that contain similar words 
end up in the same group. It presents these on screen as 10 boxes, 
each containing a collection of keywords. The user selects one or 
more of the groups. Scatter/Gather then discards the documents in 
the unselected boxes and scatters the remainder into ten more 
groups. It repeats the process until the user is satisfied that it's 
worth reading the gathered texts.   

To find the fastest retrieval routes--in other words, those using the 
smallest number of steps--Pirolli and Card worked backwards 
through Scatter/Gather, starting from the target documents. Then 
they asked their artificial forager to go find the same pieces of 
information within the IR Test Collection. It did so with little problem. 
When Pirolli and Card plotted the forager's track through the 
collection, it matched the ideal route almost perfectly.   

They then recruited eight human volunteers and asked them to 
perform the same task. Again, their routes closely matched the ideal 
one. It seems as though informavores really do employ optimal 
foraging strategies to sniff out rich information patches and avoid 
the arid plains.   

Experts in foraging theory agree. "It's likely Web users rely on 
problem-solving abilities with deep evolutionary roots," says Bruce 
Winterhalder, an anthropologist at the University of North Carolina 
at Chapel Hill who has studied human foragers in great detail. 
"Foraging on the Web presents trade-offs analogous to those of 
hunter-gatherers. Different context, but similar cost-benefit 
problems." Biologist David Stephens of the University of Minnesota, 
who co-wrote the seminal 1986 book Foraging Theory, adds: 
"Animals have been solving search problems for millennia, and 
natural selection has made them good at it. It follows that we can 
learn something from them."   

What that means in practical terms is that database and Web 
designers could use foraging theory to help them create more 
productive information resources. The theory could prove 
particularly useful at that crucial moment when a forager starts 
thinking about leaving one patch in search of another.   

In this respect, one of the most useful ideas the research has 
produced is that of "information scent". Pirolli and Card guessed that 
information foragers--whether human or artificial--have some way of 
evaluating the likelihood of finding target information in a given 
Scatter/Gather box. This led them to the idea that associated 
concepts "rub off" on one another, leaving detectable traces, just as 
a watering hole frequented by woolly mammoths will smell of woolly 
mammoths. A hunter-gatherer seeking mammoths is likely to be 
drawn to the watering hole, if only to look for spoor. Information 
foragers do the same. Imagine you're looking for texts about 
foraging theory. If Scatter/Gather throws up a box containing the 
keyword "hunter- gatherer", you're likely to select that box. It just 
smells right.   

Xerox is now trying to capture the essence of information scent and 
infuse it into Web pages, giving surfers subtle come-ons as they 
sniff around for useful sites. "We are developing technologies that 
help designers make page layouts that give off good information 
scent--cues that allow users to assess the match of information to 
their needs and identify how to get to it," says Pirolli.   

The analogy between food and information looks like being a big 
help to Web designers. But at some point, Pirolli says, it's likely to 
break down. For one thing, there's the question of evaluating costs 
and benefits. Biologists and anthropologists can always draw up an 
energy balance sheet for a foraging behaviour in joules. The value 
of information isn't so easy to measure. Another problem is that 
foraging models tend to assume environments stay the same over 
time, whereas information ecologies are nothing if not dynamic. 
Ingen-ious informavores--and those who seek to provide them with 
information--can actively manipulate their environment.   

And even if information foraging theory works, there's no guarantee 
that it will be used to benefit the forager. Think of insectivorous 
flowers that lure flies with the scent of carrion. As Card points out: 
"The vendor's interests may not correspond with the searcher's. 
They may camouflage information to hide it or mimic something that 
they think you want. Banner ads, especially ones with fake buttons 
on them, are an example." So next time you're hunting down 
information on the Web, beware. It could smell like a juicy rabbit, but 
turn out to be a vole.   

Copyright 2000 New Scientist  





* * * * * * * * * * * * * *  From the Listowner  * * * * * * * * * * * *
.	To unsubscribe from this list, send a message to:
majordomo at scn.org		In the body of the message, type:
unsubscribe scn
==== Messages posted on this list are also available on the web at: ====
* * * * * * *     http://www.scn.org/volunteers/scn-l/     * * * * * * *