Lecture 2 - Similarity and Analogy

Question 1

What is one of the reasons we are so impressed with chat-GPT?

Answer 1

The answers generated by Chat-GPT are very human-like. This similarity is very impressive and novel to us.

Question 2

What is the Turning Test and how do we need to re-think this test given the responses from Chat-GPT?

Answer 2

The Turning Test, developed by Alan Turing, was a thought experiment about how we could determine whether a computer was intelligent (in the way we perceive intelligence). He proposed that if a computer could come off as human, or respond in a way that you could not detect that it was a computer then we could classify that computer as being intelligent. With the rise of sophistication in AI technologies, such as Chat-GPT, we may need to re-think such a test for intelligence.

Question 3

Define categorisation.

Answer 3

Sorting objects, ideas, beliefs, scenarios etc into groups, based on common features that we are attending to or basing the categorisation on.

Question 4

What role does similarity play in learning?

Answer 4

Applying learnt knowledge involves recognising whether the novel situation is similar enough to the context in which the knowledge was acquired.

Question 5

What is Duncker’s Radiation Problem?

Answer 5

Duncker’s Radiation Problem asks that if the only way to destroy a tumour is with powerful radiation, but a direct stream of this radiation will also kill healthy tissue, then how can we destroy the tumour without destroying the healthy tissue?
The answer is that we fire lots of weaker streams of radiation that converge at the point of the tumour and be at the strength required to kill the tumour. This is how the tumour can be destroyed, without harming the healthy tissue.

Question 6

When participants heard The Fortress story before hearing Duncker’s Radiation Problem what was the percentage of participants that were able to figure out Duncker’s Radiation Problem compared to when they only heard Duncker’s Radiation problem?

Answer 6

When only told the radiation problem only 10% of participants figured out the solution.
When told the Fortress story then the radiation problem 30% of participants figured out the solution. And when told the Fortress Story and then the radiation problem and where told to “think about the fortress story” 70% of participants figured it out.
This is an example of using similarity when applying acquired knowledge/learning.

Question 7

What is Gestalt perception?

Answer 7

Gestalt perception is the low perception that is constantly occurring that we do not actively participate in, but which informs the way we perceive our environment and categorise our environment. So for example, we will see a cube of red dotes as just that, but if there was a cube of rows of red and blue dots then we would perceive the dots as forming lines through the cube.

Question 8

What role does similarity play in memory?

Answer 8

Memory retrieval is dependent on the similarity of the retrieval context to the original encoding context.

Question 9

What role does similarity play in Generalisation of Beliefs?

Answer 9

When we have a belief about something and we are present with a novel, but similar thing then we apply the knowledge of one thing to the other. Such as, if we know that chickens lay eggs and we know that chickens are a bird, we may then assume that owls, which are birds, also lay eggs.

Question 10

What role does similarity play in Eyewitness Identification?

Answer 10

Suspect identification is much more likely to happen if the foils are less similar to the suspect, and less likely to happen if there is a high degree of similarity between the foils.

Question 11

Many cognitive abilities from a psychological point of view are dependent on the perception of similarity.

Answer 11

True.

Question 12

Similarity is based on perception and attention.

Answer 12

True.

Question 13

Carnal Love and Meat Love - what happened here?

Answer 13

This was a United Nations case, where two companies had similar names and there one was suing the other for copyright reasons - not sure why the UN was involved.

Question 14

Define Similarity

Answer 14

A sense of sameness.

Question 15

What are some methods to measure similarity?

Answer 15

Rating Scales (and then using Multidimensional Scaling to view the stimulus space based on scaling visuallly).
Response time.
Confusability tasks.
AXB Forced choice - is X more like A or more like B.
Stimulus arrangement.
Pair-wise rating.

Question 16

What is Multidimensional Scaling (MDS)?

Answer 16

MDS is a statistical method used to present in 2-dimensions, i.e. a graph, data points that 3 or more variables that need to be taken into account. MDS was used in Hout, Goldinger & Ferguson (2013) when they wanted to visualise the responses to the Rating Similarity task.

Question 17

What did Hout, Goldinger & Ferguson (2013) find in their experiment on looking at similarity measures.

Question 18

When is stimulus arrangement too hard to use a measure of similarity?

Question 19

What are pairwise ratings? And what experiment done in 2017 discussed in lecture used this measure of similarity?

Answer 19

Experiment done by Nosofsky, Sanders, Meagher & Douglas (2017) looked at participants’ ratings of pairs of 360 images of different types of rocks - pairwise rating.

Question 20

What are some theories of Similarity? Define these theories briefly,

Answer 20

Geometric Models.
Tversky’s Contrast Model.
Structural Alignment.

Question 21

Looking at measures of similarity, such as Stimulus Arrangement, gives an insight into the cognitive processes underlying our ability to perceive similarity/categorisation.

Question 22

What are some of the limitations of The Geometric Models of Similarity?
Hint: Symmetry Contraint violation.

Question 23

What is the Features Model of Similarity?

Answer 23

Feature Models of Similarity state that similarity is based on similar/dissimilar features.

Question 24

What is Taversky’s Contrast Model of Similarity?

Answer 24

Takes into account the attention to the common elements AND unique elements when we are engaging in comparing whether objects are similar are not. People’s perception of similarity will vary on how much they attend to the common and/or unique elements.

Question 25

Define Structural Alignment Model of Similarity.

Question 26

What was the experiment done by Goldstone (1994)?
Hint: Structural Alignment Model

Answer 26

Found that MIPs contributed more to how we perceive similarity than MOPs. What does this mean for the Structural Alignment Model of Similarity?

Question 27

What are some examples of when and where similarity comes into play in our everyday lives?

Answer 27

1. White supremacy - a prioritising of the feature of being white and assuming there must be something much more similar about those that are white than other skin colours, despite DNA responsible for skin colour is below 2% or our DNA.
2. Algorithms, such as content-based filtering or collaborative fiktering.

Question 28

In Hout, Goldinger & Ferguson (2013) what were they aiming to do?

Answer 28

The study looked at three different measurements of similarity and determined which, out of the three, produced results closest to the original stimulus space. The measures of similarity used were Similarity Rating, AXB Forced Choice, and Stimulus Arrangement.

Question 29

What is Multidimensional Scaling and how did Hout, Goldinger and Fergus (2013) use it in their study ?

Answer 29

MDS is a way of mapping similarity based on responses to similarity scaling of objects. They used MDS to visualise how participants had rated objects based on similarity in their study and then compared this to the original stimulus space.

Question 30

What was the main finding from the Hout, Goldinger & Ferguson (2013) study?

Answer 30

That Stimulus Arrangement was a more valid measurement of similarity than the Similarity Rating or AXB Forced Choice measures of similarity.

Question 31

When there are too many stimuli for it to be efficient to use Stimulus Arrangement, Similarity Rating or AXB Forced Choice, what is another method we looked at that could be used instead?

Answer 31

Pair-wise rating of similarity.

Question 32

What are the four cognitive theories of how we assess similarity?

Answer 32

1. Geometric Models of Similarity
2. Feature Models of Similarity
3. Traversky’s Contrast Model of Similarity
4. Structural Alignment Model of Similarity

Question 33

At its core, what is the Geometric Model of Similarity?

Answer 33

The Geometric Model of Similarity is a cognitive theory of how we assess similarity and posits that we perceive similarity or dissimilarity as closer or greater distances between objects, events, etc that we are comparing.

Question 34

In the lecture we discussed to limitations with the Geometric Model of Similarity. What were they?

Answer 34

1. The Geometric Model of Similarity sometimes disobeys SYMMETRY CONSTRAINT.
2. The Geometric Model of Similarity sometimes disobeys TRIANGLE INEQUALITY.

Question 35

What does Traversky’s Contrast Model of Similarity explain that the Geometric Model could not explain?

Answer 35

Tversky’s Contrast Model posits that the way we assess similarity is based on attending to shared and unique features of objects, events or scenarios. This model is thus able to explain the Asymmetry of Similarity that can exist between objects, whereas the Geometric Model could not do this.

Question 36

What is the Structural Alignment Model of Similarity?

Answer 36

The Structural Alignment Model of Similarity says that the way we assess similarity is based on the Matches In Place (MIP), Matches Out of Place (MOP) and Mismatches between images/objects. I.e. we draw attention or alignment between the structure or features of the images or objects.

Question 37

In the Structural Alignment Model of Similarity, what did Goldstone (1994) find out the relationship between our ability to accurately assess similarity and MIPs and MOPs?

Answer 37

Goldtone (1994) found that MIPs were more important for how we rate similarity than MOPs, such that even if there were more MOPs, participants more likely to assess something as dissimilar if there were less MIPs.

Question 38

What cognitive theory of similarity did Goldstone (1994) use in their study?

Answer 38

The Structural Alignment Theory of Similarity - used images of birds made up of shapes and colours that varied.

Question 39

What is the theory Similarity as Transformations and what does it suggest about the cognitive process behind assessing Similarity?

Answer 39

The Similarity as Transformation theory posits that we assess similarity by determining how many transformations are required between two images of objects. The idea being that the more transformations required to get from A to B, the less similar A is from B.

Question 40

Hahn, Richardson & Chater (2003) looked at similarity. What cognitive theory of similaritu were they assessing in this study and what did they find?

Answer 40

Hahn, Richardson & Chater (2003) looked at similarity through the lens of the Similarity as Transformation theory.
They found that there was a direct correlation between what we rated as similar or dissimilar and the number of transformations between the images used.

Question 41

What does the Geometric Model of Similarity view similarity as?

Answer 41

Similarity as distance.

Question 42

Symmetry constraint is one of the limitations of the Geometric Model of Similarity.
What is the Symmetry Constraint?

Answer 42

The symmetry constraint is the observation that, in regards to viewing similarity as distance, the distance between A and B is not the same as B to A - see surgeon/butcher example.
The Geometric model of symmetry cannot explain this.

Question 43

Do Geometric Models of Similarity Obey the Triangle of Inequality?

Answer 43

In theory they should, however, in practice the Triangle of Inequality does not seem to be obeyed - see example of A(happy) to B (sad) + B(sad) to C(frightened) < A(happy) to C(frightened) in reality, but this disobeys the Triangle of Inequality.

Question 44

Does Tversky’s Contrast Model of Similarity provide a solution to ASYMMETRIC SIMILARITY?

Answer 44

Yes.
Taking into account shared and unique features when assessing similarity allows us to explain how there can be astmmetric similarity using Tversky’s Contrast Model of Similarity.