Prospective Memory in Aviation, Everyday Tasks, and the Laboratory Key Dismukes Psychonomic Society

Download Report

Transcript Prospective Memory in Aviation, Everyday Tasks, and the Laboratory Key Dismukes Psychonomic Society

Prospective Memory in Aviation,
Everyday Tasks, and the Laboratory
Key Dismukes
Human Systems Integration Division
NASA Ames Research Center
Moffett Field, CA
Psychonomic Society
18 November 2006
Major U.S. Airline Accidents Involving
Inadvertent Omission of a Normal
Procedural Step (1987-2001)
•
•
Five of 27 major accidents in which NTSB found crew error to
be a causal factor
–
1988, Detroit: DC-9-82 Flaps/slats not set to takeoff position,154
killed; 1 seriously injured
–
1989, Dallas-Ft Worth: B727 Flaps/slats not set to takeoff
position, 14 killed; 5 seriously injured
–
1995, LaGuardia: MD-82 Pitot-static heat not turned on, aircraft
destroyed, no fatalities/serious injuries
–
1997, Houston: DC-9 Hydraulic boost pumps not set to high
before landing, aircraft badly damaged, no fatalities/serious
injuries
–
2001, Little Rock: MD-82 Spoilers not armed before landing, 11
killed; 110 injured
Multiple factors involved, but a central aspect was experienced
pilots forget to perform a normal, highly-practiced procedural
step
Theoretical Accounts of
Prospective Memory Still in Infancy
Our view of evidence:
• After intention for deferred action is formed, attention
turns to other tasks
• Deferred intention fades from WM but remains in
LTM
• Retrieval requires noticing some cue associated in
memory with intention
– Can be environmental cue or stream-of-thought cue
– Stored intention is retrieved if sufficient activation spreads to
it from cue
Linking Real-World PM Phenomena to
Underlying Cognitive Processes
• Three complementary approaches:
– Ethnographic
– Analyses of accident/incident reports
– Laboratory studies
• Structure of airline operations well suited to study of
skilled performance & error
– Operations are highly standardized
– Can observe deviations from formal SOPs
– Fair consensus about appropriate/inappropriate actions
An Ethnographic Study
(Loukopoulos, Dismukes & Barshi, in preparation)
•
Focused on B737—widely used in air transport
•
Reviewed written operating procedures, participated
in classroom and flight simulation training at two
major airlines
•
Observed large number of flights from cockpit
jumpseat in normal revenue flights
Some Findings from
Ethnographic Study
• Pilots are:
– Frequently interrupted
– Forced to postpone planned tasks
– (Each) forced to perform multiple tasks in parallel
•
Timing and nature of task demands is not entirely predictable and is
only partly under control of the crew
• These results confirmed by a study analyzing NTSB reports of
the 19 major U.S. airline accidents attributed primarily to crew
error 1990-2001 (Dismukes, Berman, & Loukopoulos, in press)
Memory Errors Reported to the
Aviation Safety Reporting System
(Nowinski, Holbrook, & Dismukes, 2003)
•
Sampled 20% of air carrier reports over 12-month
period
–
•
Selected reports clearly involving some type of memory
error
74 of 75 reports involved prospective rather than
retrospective memory
–
Does not necessarily indicate PM more common than RM
failures, but suggests PM failures may be:
•
•
•
more consequential
more likely to be reported
and/or more common
Varieties of Prospective Memory
in the Cockpit
•
In these three studies analyzed:
–
–
–
–
•
What tasks were being performed concurrently
Whether forgotten task was habitual
What cues would normally be present to trigger retrieval
Whether those cues were actually present
Concluded PM demands emerge in five types of
situation:
–
–
–
–
–
Episodic tasks
Habitual tasks
Atypical tasks substituted for habitual actions
Interrupted tasks
Interleaving tasks, including monitoring
Varieties of PM
(1) Episodic Tasks
• Tasks not performed habitually
– Type of PM task most often studied experimentally
– Example: ATC instructs crew to report passing through
10,000' during descent
• In real world, ongoing tasks often divert attention
from cues that might trigger retrieval (Holbrook, Dismukes
& Nowinski, 2005)
– May be a major source of variance, but not studied
extensively
• This is the type of task studied most often in
laboratory paradigms
Varieties of PM
(2) Habitual Tasks
• Most cockpit tasks are habitual
– Many tasks must be performed, many with multiple steps
– Normally performed in set sequence by SOP
– Execution becomes largely automatic—deliberate search of
memory not required
– Explicit (episodic) intention not required—intention is
implicit in action schema, stored as procedural memory
• Performance of habitual tasks normally quite reliable
– Performance undermined if normally present cues are
removed
– Example: Setting flaps to takeoff position is deferred
because of freezing slush on taxiway
• Action out of sequence removes normal cues & normal context
• Habitual task becomes episodic task, but pilots may not realize
need to encode explicit intention
Varieties of PM
(2) Habitual Tasks
• Most cockpit tasks are habitual
– Many tasks must be performed, many with multiple steps
– Normally performed in set sequence by SOP
– Execution becomes largely automatic—deliberate search of
memory not required
– Explicit (episodic) intention not required—intention is
implicit in action schema, stored as procedural memory
• Performance of habitual tasks normally quite reliable
– Performance undermined if normally present cues are
removed
– Example: Setting flaps to takeoff position is deferred
because of freezing slush on taxiway
• Action out of sequence removes normal cues & normal context
• Habitual task becomes episodic task, but pilots may not realize
need to encode explicit intention
Varieties of PM
(3) Atypical Actions Substituted for
Habitual Actions
• Example: Crew often departs from an airport via a
Standard Instrument Departure requiring left turn to
300 degrees at 2000 feet
– Execution of procedure becomes largely automatic
– If ATC on one occasion amends instruction to turn left to 330
degrees at 2000 feet:
• Crew must form episodic intention to turn to 330 and must
inhibit normal response to level out at 300 degrees
• When busy with other tasks, crews vulnerable to habit intrusion
(Reason, 1984) and level out prematurely at 300
Varieties of PM
(4) Interrupted Tasks
• Interruptions very common in cockpit
– Duration lasts from seconds to minutes
• Interruptions are often so abrupt and salient that
pilots may do little to encode explicit intention to
resume interrupted task
• Common error is to deal with interruption then
proceed to next task, forgetting to complete
interrupted task
– Cockpit does not always provide salient cues for status of
interrupted tasks
– Perceptually rich cockpit environment associated with many
tasks remaining to be performed
Varieties of PM
(5) Interleaving Tasks, Including Monitoring
• While performing ongoing tasks pilots must
periodically monitor status of other tasks, e.g.:
– While running checklists, communicating, programming, etc.,
First Officer must monitor progress of taxi by Captain
• Monitoring is important, but difficult to maintain
monitoring goal in WM as dual task if event rate low
– Similar to difficulty with vigilance (Parasuraman, 1986) but
different:
• Must interrupt ongoing task and shift attention
– Attention allocated preferentially to areas with high info
content over monitoring for low probability events, albeit with
high consequences (see Wickens et al., 2003 for attention
allocation model)
– When attention shift delayed, monitoring task may slip from
WM & must be retrieved, as in other forms of PM
Varieties of PM
(5) Interleaving Tasks, Including Monitoring
• While performing ongoing tasks pilots must
periodically monitor status of other tasks, e.g.:
– While running checklists, communicating, etc., First Officer
must monitor progress of taxi by Captain
• Monitoring is important, but difficult to maintain
monitoring goal in WM as dual task if event rate low
– Similar to difficulty with vigilance (Parasuraman, 1986) but
different:
• Must interrupt ongoing task and shift attention
– Attention allocated preferentially to areas with high info
content over monitoring for low probability events, albeit with
high consequences (see Wickens et al., 2003 for attention
allocation model)
– When attention shift delayed, monitoring task may slip from
WM & must be retrieved, as in other forms of PM
An Experimental Study of Interruptions
(Dodhia & Dismukes, submitted)
From aviation studies, hypothesized individuals forget to
resume interrupted tasks largely because:
1.
Salient intrusion of interruptions diverts attention, discouraging
encoding explicit intention to resume or identifying specific reminder
cues
–
2.
Cues indicating opportunity to resume interrupted task may not
match form of (implicit or explicit) encoded intention
–
–
3.
Resumption will depend on happening to notice cues previously
associated with interrupted task—very unreliable
“End of the interruption” is not a discrete perceptual cue but a state of
affairs requiring interpretation
Poor match between cues and encoding provides little activation to
retrieve intention from LTM
End of interruption often followed by other task demands that divert
attention from interpreting significance of completing interruption
–
New task goals spread activation to task subgoals rather than to
retrieving implicit goal of resuming interrupted task
An Experimental Study of Interruptions
Experimental Paradigm
• Participants answer questions resembling SAT,
Category of questions changed between blocks
• Instructed that when blocks were interrupted they
should remember to resume interrupted block after
completing interruption before starting new block
• In baseline condition interruptions were abrupt—
current screen replaced with screen of different color
and with different category of questions
• After each completed block a screen appeared for
2.5 sec: “Loading next block”, followed by next block
in series
An Experimental Study of Interruptions
Experimental Paradigm (continued)
• Without explicit prompt participants had to remember
to return to interrupted block
• Proportion of successful resumptions of interrupted
task in baseline condition: 0.48
An Experimental Study of Interruptions
Encoding Manipulations
• Hypothesis 1: Intrusion of sudden interruption discourages
adequate encoding of explicit intention to resume interrupted
task
• Tested with encoding reminder manipulation:
– Interruption began with 4 sec message: “Please remember to
return to the block that was just interrupted”
– Results: Performance increased from 0.48 to 0.65 (proportion of
interrupted blocks resumed)
• Was improvement due to reminder or to 4 sec delay before
beginning interrupting task? Tested with encoding pause
manipulation:
– Interruption began with 4 sec blank screen
– Results: Performance increased from 0.48 to 0.65
• Conclusion: Pause before beginning interrupting tasks helps
remember to resume interrupted task
An Experimental Study of Interruptions
Retrieval Manipulations
• Hypothesis 2: Individuals are likely to forget to resume
interrupted tasks because of mismatch between form of
encoded intention and diverse cues that must be interpreted to
recognize interruption has ended
• Tested with retrieval reminder manipulation:
– 2.5 sec “Loading next block” screen also said: “End of interruption”
– Results: Performance increased from 0.48 to 0.90 (proportion of
interrupted blocks resumed)
• Hypothesis 3: Individuals are likely to forget to resume
interrupted tasks if interrupting task is directly followed by other
tasks that demand attention
• Tested with retrieval pause manipulation:
– Interval between end of interruption and beginning of next block
increased to 8-12 sec and countdown clock appeared on screen
– Results: Performance increased from 0.52 to 0.88 (proportion of
interrupted blocks resumed)
An Experimental Study of Interruptions
Summary of Results
An Experimental Study of Interruptions
Implications
• Theoretical:
– Supports three hypotheses but more research needed
– Consistent with model of PM in which intentions are
retrieved from LTM by activation from cues processed
attentively
• Practical:
– Pausing before dealing with interruptions to explicitly encode
intention to resume interrupted task and to identify specific
reminder cues
– Periodically pausing after completing tasks to ask what other
tasks remain to be done
Final Thoughts
• Prospective memory probably operates in similar
ways in the skilled tasks performed by other kinds of
experts, e.g., medical personnel
• To be effective, countermeasures to PM errors must
be realistic for experts to apply while performing their
task
• Understanding the real-world performance requires
integrating diverse research approaches, including
observing real-world performance and well-controlled
experimental studies
– This integrative approach enhances the power of both field
and lab studies
Thanks to my colleagues in this research:
Rahul Dodhia, Jon Holbrook, Kim Jobe,
Loukia Loukopoulos, and Jessica Nowinski
For more information:
http://human-factors.arc.nasa.gov/ihs/flightcognition