1. CS260: Lecture 2
John Canny
Fall 2006
9/19/2018

2. Human Learning Why study human learning in HCI? 9/19/2018
A: People need to *learn* new applications – they use existing mental models to generalize to new applications. As designers we should understand the process of constructing new mental models from existing ones.
A: Human knowledge is heavily layered. It can be very hard to understand users’ mental models by just observing them and trying to figure out what is going on.
A: People typically go through stages of understanding as they learn a new concept (concrete to abstract, novice to expert). This progression is very important for design.
A: Learning is an important part of knowledge work – when people work creatively, they are using existing knowledge to create new knowledge. They use mental models to anticipate what will happen, then act, and then analyze the results. This is a learning process.
A: In the learning sciences, there is a good understanding of how to measure human performance. We can adapt some of those ideas to other settings.
9/19/2018

3. Why Study Human Learning?
A: People learn to use new systems periodically
A: As people gain familiarity with systems, they evolve their use of them.
A: Learning science is one of the most-studied areas in social science, with effective success metrics.
A: Learning is not an isolated mental process, its part of most everyday “knowledge work” practices (micro-genesis).
A: To better understand adult, child and unschooled users.
A: Because formal education strongly shapes the way people think.
A: People need to *learn* new applications – they use existing mental models to generalize to new applications. As designers we should understand the process of constructing new mental models from existing ones.
A: Human knowledge is heavily layered. It can be very hard to understand users’ mental models by just observing them and trying to figure out what is going on.
A: People typically go through stages of understanding as they learn a new concept (concrete to abstract, novice to expert). This progression is very important for design.
A: Learning is an important part of knowledge work – when people work creatively, they are using existing knowledge to create new knowledge. They use mental models to anticipate what will happen, then act, and then analyze the results. This is a learning process.
A: In the learning sciences, there is a good understanding of how to measure human performance. We can adapt some of those ideas to other settings.
9/19/2018

4. Lev Vygotsky
Vygotsky was an extraordinary scholar who studied Law, and taught Literature, History of Art and Psychology by age 22.
Vygotsky pursued a social perspective and took it very far, developing theories of knowledge, development, and education that were profoundly influential.
9/19/2018

5. Vygotsky in Education
Vygotsky is (with Piaget) the leading education theorist of the early 20th century.
Vygotsky’s social theory of learning –
Like Piaget he argued that children learn by constructing their own understanding of the world they experience.
In contrast to Piaget, he insisted that “the world” experienced by children is a social, rather than a natural one. i.e. games, toys, and books are social constructions that embody social norms and expectations for the child.
9/19/2018

6. Vygotsky – Genetic method
Another of Vygotsky’s ideas is his “genetic” domains:
Onto-genesis: Development by an individual
Socio-historical: Development of the society
Phylo-genesis: Development of the (human) species
Micro-genesis: Creation of ideas & concept learning
His social theory involves the interplay between 1. and 2.
Thus Vygotsky’s approach interleaves methods that would be regarded as both scientific and humanistic.
9/19/2018

7. Vygotsky’s Genetic Principle
Vygotsky’s genetic principle has extraordinary implications.
Among other things, it implies a socio-historical approach to understanding human behavior.
This approach is still popular among researchers in Management and Cooperative Work.
And much more…
9/19/2018

8. Power laws
One of the puzzling findings in human behavior is the presence of ubiquitous “power law” processes. A power law probability distribution has the form:
p(x) = 1/rank(x)a
Where a is typically between 0 and 3 and is often very close to 1.
Power laws are not “natural” from a statistical point of view. Something very special must be happening to cause them.
9/19/2018

9. Power laws Processes that exhibit power-law statistics:
The probabilities of words in the English language
Word probabilities in any reasonably large corpus
Number of links into a web site, click-thru of web sites
Citations in academic journals
Paths that users take walking through a house
Products purchased from a vendor
Size of cities, sizes of companies
Stock market returns, trades, volumes
Incomes of people… many others
“One of the strongest [nontrivial] facts in social sciences”
9/19/2018

10. Genetic processes and Power laws
Power laws arise (and were first studied) in genetics. If there are k “types” we assume each type gives birth, mutates or dies with some probability.
With a variety of choices of parameters, this process gives rise to power law distributions.
For many choices of parameters, the power is 1.
9/19/2018

11. Genetic processes and Power laws
While genetic processes may explain many power law phenomena, they can do more than that.
Just as in classical genetics, a genetic process implies a “phylogenic” tree exists for the objects in a domain.
We can guess the ancestry of an item from its current popularity, the popularity of its parent, and the separation time.
This can help us understand the adoption and learning of activities.
9/19/2018

12. Learning and existing knowledge
Learning is a process of (genetically) building new knowledge using existing knowledge.
Knowledge is not acquired but constructed out of existing “materials”.
The process of applying existing knowledge in new settings is called Transfer.
9/19/2018

13. ZPD Learning is layered and incremental.
In real societies, learners are helped by others.
In fact learners have a “zone” of concepts they can acquire with help.
This is the Zone of Proximal Development (ZPD).
9/19/2018

14. Back to learning..
Example: Who knows what this is?
100k
9/19/2018

15. Back to learning..
Example:
9/19/2018

16. Learning new applications
Applications are designed to fit in ordinary users’ ZPD.
In most cases, you can’t assume that there is human available to help a user learn the new system.
A tutorial help system can provide much of this support.
9/19/2018

17. Learning new applications
People learn best by doing (constructing new knowledge).
Using a system exposes a user’s conceptual models of how it works, and allows them to diagnose mistakes.
A tutorial help system should be able to recognize and respond to common user misunderstandings.
9/19/2018

18. Learning and experience
Learning is most effective when it connects with the learner’s real-world experiences.
The knowledge that the learner already has form those experiences serves as a foundation for knew knowledge.
9/19/2018

19. Learning and transfer
Transfer is certainly enhanced by similarity between the old and new contexts.
What other factors should affect transfer?
9/19/2018

20. Transfer and understanding
Transfer depends on thorough learning in the first situation (learning with understanding*).
The more thorough the understanding in the first situation, the more easily knowledge will transfer.
9/19/2018

21. Understanding
By understanding we mean that a person has a mental model of why a thing behaves as it does.
This model allows the person to predict how the thing behaves in other situations, and to “explain” their reasons for that conclusion.
9/19/2018

22. Transfer and Generality
Generality of existing knowledge: has the learner already seen it applied in several contexts?
9/19/2018

23. Transfer and Motivation
Motivation: is the new knowledge useful or valuable?
Motivation encourages the user to visualize use of the new knowledge, and to try it out in new situations.
Students are usually motivated when the knowledge can be applied to everyday situations.
9/19/2018

24. Transfer and Abstraction
Is the existing knowledge abstract or specific?
Abstract knowledge is packaged for portability. Its built with virtual objects and rules that can model many real situations.
E.g. clipart
9/19/2018

25. Summary Factors affecting transfer: Real-world experience
Degree of Understanding (the earlier concept)
Generality of the earlier knowledge
Motivation
Abstraction
9/19/2018

26. Metacognition
Metacognition is the learner’s conscious awareness of their learning process.
Metacognition helps transfer
9/19/2018

27. Metacognition Strong learners carefully manage their learning.
For instance, strong learners reading a textbook will pause regularly, check understanding, and go back to difficult passages.
Weak learners tend to plough through the entire text, then realize they don’t understand and start again.
Have I learned this yet?
9/19/2018

28. Let me guess what’s coming next..
Metacognition
Another very good strategy is to predict the next main point in an argument before you read it:
“What would a user interview be like?”
“What techniques will improve learning”?
Then when you see the real answer, the new knowledge will tie with real experience – the experience you just had.
Let me guess what’s coming next..
9/19/2018

29. Structuring Learning
A similar strategy is very effective for teaching.
Ask students to work on a problem first, trying out their own approaches.
Then provide an explanation (a set of principles to explain the problem’s behavior).
Reading, lecture
Problem, lecture
Problem work only
9/19/2018

30. Structuring Learning
Again this gives students some rich and immediate experience with the problem.
When the explanation is given, students can relate the new information with the experience they just had.
Reading, lecture
Problem, lecture
Problem work only
9/19/2018

31. Break
9/19/2018

32. Piaget: Stages of learning
Piaget observed very systematic progression of knowledge in young children through stages:
Sensori-motor (acting, observing, remembering)
Semiotic or symbolic (naming things)
“Concrete” operations (relationships, transformations)
Propositional or formal thought
9/19/2018

33. Sensori-motor stage (< 2 years)
Conditioned behaviors, and first hand-eye coordination.
Grasping, manipulating things.
Some indirect manipulation.
Object persistence.
9/19/2018

34. Semiotic stage (>1.5 years)
Children continue to play with “missing” objects, and may use gesture to invoke them.
This soon turns to imaginary play.
Drawing.
Speech – naming first the things that are present.
Then referring to things that are not present, and to the past and future.
9/19/2018

35. Concrete thought (2-7,7-11 years)
Concrete thought: a system of (real) objects, relationships, and operations on them.
Children “understand” things by being able to relate them to similar things, and to predict the consequences of their actions.
They can plan and act to achieve a desired outcome.
9/19/2018

36. Concrete thought
But early concrete thought is still tied to direct experience – it is not “de-centered.”
E.g. children in this stage can navigate through their neighborhood, changing their route if needed.
i.e. they can mentally model and predict the results of their actions.
But they cannot indicate that route abstractly, say on a map.
9/19/2018

37. Concrete thought
Concrete thought includes rich spatial and temporal relationships.
Visual design is a “concrete” process.
9/19/2018

38. Formal thought (11+ years)
Objects and operations no longer need to relate to the world. Things don’t need to be true or consistent. Thinking is a “game”.
“Operations” are more abstract, and often complementary e.g. joining-separating.
Children learn a number of principles, like reversibility, proportion, chance.
9/19/2018

39. Formal thought caveats
Researchers have found that the transition to formal thought is not as reliable as Piaget had thought.
Many features of this stage are missing in children who do not attend school.
This stage corresponds with the transition from learning from experience (pre-school), to learning from texts (school).
9/19/2018

40. Aside: Design for Developing Regions
Users in developing regions often have limited formal schooling. This leads to systematic challenges:
Users are not comfortable with formal objects such as “a red square”, i.e. they do not treat objects as sets of attributes, but as concrete things.
Graphic (picture representations) are OK, but should not be too abstract.
Users favor representations of people doing actions.
Users expect consistency and realism.
9/19/2018

41. Formal thought (7+ years)
Side-effects of abstract representations:
Context disappears – things are just true or false everywhere.
Rules are very powerful, and both the rules and the reasoning must be accurate, or false conclusions will be drawn.
Detail must be discarded or the rules may conflict.
9/19/2018

42. Thought styles
Designers and other visually-oriented people usually favor concrete thought – context-dependent, rich representations.
Technologists and mathematically-oriented people favor formal thought – context independent, sparse representations, rich consequences.
9/19/2018

43. A mismatch
Many interface researchers (technologists) tried to build UI design tools using abstract interface specs (UIMSes)
the designer specifies rules about the interface and the system finds a solution satisfying them.
Real designers hated this idea. They lost control over spatial relationships and overall layout which was lost in the rules.
9/19/2018

44. Macro and micro-Piaget
Piagetian stages are often evident in learners’ acquisition of particular concepts.
i.e. the learner’s first experience is “sensori-motor” – if I do X, then Y happens.
They develop a language for naming the operations, objects, groups of objects etc.
They acquire concrete understanding of the system’s operation: I can change state X to Y using operation Z.
Finally, they may develop a formal understanding of how the system works (as explicit rules).
9/19/2018

45. Piaget’s progression
The Piagetian progression can be a good model for the progression in learning new concepts, like how to use a computer program.
Look for a Sensori-motor  Symbolic  Concrete  Abstract progression in your own learning, and in your users’.
9/19/2018

46. Inquiry cycles
Inquiry-based learning makes student’s meta-cognitive strategy explicit.
It also treats learning as a kind of scientific research.
9/19/2018

47. Inquiry cycles Question: a new problem for the learner
Hypothesis: Learner proposes a solution or a way to understand the problem better
Investigate: Learner figures a way to try out the hypothesis (often an experiment)
9/19/2018

48. Inquiry cycles Analyze: understand the results of the investigation.
Model: Construct a model or principle for what’s going on.
Evaluate: Evaluate the model, the hypothesis, everything that came before.
9/19/2018

49. Inquiry cycles See http://thinkertools.soe.berkeley.edu
Thinkertools uses software agents to personify the different stages in inquiry cycles.
The agents help scaffold the child through the cycle.
9/19/2018

50. Scaffolding
Refers to the process of shaping the learner’s experience while learning, by creating a “scaffold” to guide their actions.
Generally, the teacher begins by doing most or all of the task.
The task is repeated, with the learner doing more and more of it.
Eventually, the learner does the entire task themselves – the scaffold is removed.
9/19/2018

51. Scaffolding and ZPD
Scaffolding produces a steady progression through the learner’s ZPD (Zone of Proximal Development)
ZPD
Inaccessible tasks
Solo tasks
Scaffolded learning
9/19/2018

52. Scaffolds and Tools
First-generation learning tools were “electronic flashcards”
System flashes a new item on the screen
User has to enter the right input (typing, multiple choice etc.)
System learns user’s weaknesses, and focuses its examples on those weak cases.
Quite effective for low-level learning (e.g. Morse code).
9/19/2018

53. 2nd Generation Scaffolds
Allow exploration of a knowledge domain:
Caclulators, spreadsheets, graphing programs, probes etc.
Modeling/Simulation (e.g. Interactive Physics)
Matlab + packages
9/19/2018

54. MicroWorlds An idea promoted by Seymour Papert (creator of Logo).
A Microworld is a simplified model of the physical world, which emphasizes certain physical principles and omits other detail.
E.g. 2D geometry (turtle geometry).
9/19/2018

55. MicroWorlds
Microworlds encourage less structured exploration by learners.
The idea is that the learner’s discoveries will be driven more by their own goals, leading to better learning.
The structure of the Microworld should ensure that they make the right inferences.
9/19/2018

56. Feedback, Reflection, Revision
One of the most important principles in learner-centered design is “Early Feedback”
The learner should be given feedback as soon as possible as they form new concepts.
This can take the form of a multiple-choice question – so the answers can be given immediately.
9/19/2018

57. Feedback, Reflection, Revision
Reflection tools encourage meta-cognition.
“Thinkertools” which we mentioned earlier encourages learners to follow an inquiry cycle.
9/19/2018

58. Feedback, Reflection, Revision
Small-group discussion is another way to encourage reflection.
Discussion makes each learner reflect on their understanding to explain to others, and to interpret others’ explanations.
Systems that do this: CSILE (Vanderbilt)
9/19/2018

59. Feedback, Reflection, Revision
Peer instruction (Mazur) is a pattern that encourages all these steps:
Students are given a multi-choice question
They write down an individual answer
The class “votes” their answer
Students discuss in small groups, then answer again.
Another vote is taken
The instructor explains the right answer.
9/19/2018