PSYU2236 BIOPSYCHOLOGY AND LEARNING

1
Practical One: Research Report
Can we learn from watching robots?
PSYU2236 BIOPSYCHOLOGY AND LEARNING
2
Today’s practical will involve:
– Providing a background to physical and observational
learning.
– Outlining some theories of how or why observational
learning works.
– Describing data from a recent experiment that
compared how well people learn from watching
different models – human versus robot.
– Explaining where you can find the data, as these data
will be used for your research report.
Practical 1 Overview
3
For a background to motor learning and observational
learning of motor skills, see Chapter 11 in Mazur.
We can learn motor skills by physically practicing and we
can learn by watching other people. But to what extent
can we learn by watching robots?
Practical 1 Overview
4
Background
We can learn new skills by physical practice and by watching
others perform (observational learning)
5
Background
In certain contexts, observational learning is highly
advantageous over physical practice
6
The ‘like-me’ hypothesis
Meltzoff, 2007
“The ‘like-me’ nature of others is the starting point for social
cognition”
7
The mirror neuron system
Gallese et al., 1996; Fogassi et al., 2005
Human fMRI
Grèzes & Decety, 2001; Gazzola & Keysers, 2009
Although these findings are noteworthy,
this was not a new idea:
James (1890)
Jeannerod (1994)
Prinz (1990; 1997)
IFG
IPL
Monkey recordings
8
Mirroring and learning
Blandin et al., 1999; Gog et al., 2009
9
What about the model’s identity?
Does it matter who we are learning from?
Experts vs Novices
A model that
demonstrates mastery
vs a model that develops
skill over time
Both types of model
are effective
And maybe a
combination is best
(e.g., Rohbanfard &
Proteau, 2011)
10
Does it matter who we are learning from?
What about learning from non-humans, such as robots?
What about the model’s identity?
11
Robots
Does it matter who we are learning from?
The ‘like-me’ hypothesis would make a clear prediction …
Human > Robot
The aim of this
practical and
Research Report
is to test this
prediction
12
Experimental details
• The data that will be used in your research report
have already been collected.
• The data will be made available to you via iLearn and
at the end of this presentation, where you can copy
or download the full dataset and then analyse it as
part of your research report.
• For the remainder of the practical, you will:
– have a go completing the task, so that it is familiar to you.
– Find out more of the experimental and procedural details.
13
Basic experimental design
Human model vs Robot model
14
First, have a go at the task
• Get a laptop and work in pairs or small groups.
• One person physically practices, the others watch.
• Place four fingers from your non-dominant hand
(typically left), on the number keys 1-4.
1
2 3 4 1 2 3
4
15
First, have a go at the task
• One person physically practices. Using one finger per
key, hit the number keys in the following order, as
fast and as accurately as possible (Repeat x 10).
• Everyone else watch carefully.
16
Now, switch positions and test
• Switch over: those who were watching have a go….
• Repeat x 10
17
Now, try a difference sequence…
• Now try a different, untrained sequence….
• Repeat x 10
18
Repeat the process yourselves…
• Now, write down your own sequences on a piece of
paper or laptop and repeat the training procedure.
• One person practices, the others watch.
• Then one person who watched gets to test
themselves on trained and untrained sequences.
19
Task summary
• Ok, so now you should have a good idea about the
basic task.
• Physically practicing and watching others perform
sequences makes you faster and more accurate than
performing untrained sequences.
• But does learning differ if we manipulate the identity
of the model and try to learn from a robot? The
current practical addresses this question.
20
Independent Variable
Independent Variable
Type of training. Three levels:
– Human, robot, no training.
Repeated measures or within-participant design (i.e.,
every participant is in every condition). For each
participant, a set of sequences are assigned to each
condition. E.g.,
Sequence 1 2 3 4 5 6 7 8 9 10 11 12
Condition Human Robot No training
21
Dependent Variable
Dependent Variable
Sequence completion time (s). The time taken to
complete 4 key-presses in a specified order.
22
Procedure
The procedure involves two phases (training and test).
Training Test
– Sequences are observed being
performed by two different
models
– All sequences are performed
– 4 human sequences
– 4 robot sequences
– 4 human sequences
– 4 robot sequences
– 4 untrained sequences
23
Data
• Each participant completed a series of trials per
condition in the test phase.
• The mean execution time in these trials was
calculated to represent the average speed per
condition. See the data below for the first 4
participants (pID = participant ID).
pID Human Robot No
training
1 1.93 1.26 2.47
2 1.25 1.68 1.03
3 1.02 1.67 2
4 0.67 0.89 1.65
24
Hypotheses
• A hypothesis combines the IVs and the DVs into a
prediction statement
• A good hypothesis statement is an “if” “then”
statement
• The IV is the “if” and the DV is the “then” part
• What are we manipulating (IV)?
• What effect do we expect from that manipulation on
the DV?
• Are we expecting a particular direction in that
relationship?
25
Analyses
• Your statistical training from Year 1 is sufficient
to analyse these data.
• Include:
– Descriptive statistics (e.g., mean, SD per condition)
– A plot of the main findings that illustrates the data
– An inferential statistical test, which addresses your key
hypothesis
Hot tip: Keep it simple!!
26
Possible discussion topics
• What were the effects of watching a human compared to a robot?
• Was there an improved performance for the human compared to the
robot condition? If so, can it be unequivocally attributed to theorising
from the “like-me” hypothesis?
• How do the results potentially relate to the function of mirror neurons?
• To what extent can we generalise from the findings with this task to
other tasks? Can you think of tasks where a robot model might be
superior to a human model?
• We used one type of robot, but how would the results relate to other
types of robotic agent?
• In the light of the data you have gathered, suggest ways in which other
practical tasks might be organized for optimal learning.
27
The dataset
• You can copy and paste the data
directly into statistical analysis
software, excel or whatever you
will find useful.
• You can also download the data
from iLearn.
• You can download Stata using
the following link:
https://students.mq.edu.au/support/t
echnology/software/stata
pID human robot no training
1 1.93 1.26 2.47
2 1.25 1.68 1.03
3 1.02 1.67 2
4 0.67 0.89 1.65
5 1.23 1.63 1.97
6 1.04 2.51 2.36
7 0.87 1.92 0.68
8 0.99 1.85 1.41
9 1.09 1.45 1.17
10 0.99 1.57 1.79
11 1.42 1.35 2.2
12 1.05 1.63 2.38
13 1.49 1.92 2.41
14 1.88 1.09 1.94
15 1.94 2.46 1.48
16 0.93 0.97 1.78
17 1.05 1.3 1.97
18 0.71 1.28 1.24
19 1.69 1.24 1.01
20 0.37 1.68 1.5
21 1.3 1.6 1.73
22 0.76 1.07 1.85
23 0.67 1.41 1.25
24 2.08 1.23 1.53
25 1.21 1.49 1.46
26 1.84 1.28 1.25
27 1.03 1.46 2.08
28 0.48 1.97 0.79
29 0.42 2.3 2.13
30 0.47 1.33 2.23
31 0.95 0.62 1.08
32 2.23 1.73 1.07
33 1.63 1.84 1.21
34 1.18 0.94 2.02
35 1.05 1.72 1.71
36 1.76 1.25 2.34
37 2.85 0.8 1.32
38 1.51 1.54 1.6
39 1.37 1.64 1.56
40 1.16 1.41 1.69
41 0.62 0.99 2.84
42 1.41 1.54 1.82
43 1.77 0.8 1.52
44 1.19 1.07 1.75
45 1.3 2.01 1.77
46 1.25 1.09 2.13
47 0.73 1.43 1.63
48 1.47 2.37 1.13
49 0.64 1.13 1.23
50 1.06 2.05 2.91
28
Any questions?
More information can be found on iLearn, including:
• The slides from today’s practical.
• A Research Report Assignment Sheet, which
summarises the assignment and the experimental
data.
• Background reading articles.
• An assessment rubric.