Cognitive psychology is about knowledge and thinking. In this lecture Iordanis Kavathatzopoulos covers the differences between human beings and computers, i.e. how we think, make decisions and solve problems and how computers "behave". The lecture also contains some illustrative tests.
In human-computer interaction, there is one very important matter that requires attention. Do human beings and computers behave in the same way when making decisions and solving problems? This matter is crucial for the interaction between the user and the computer. Computers are completely rational - they process all information systematically, and correctly, in accordance with their program instructions. Our thinking and reasoning, on the other hand, are not perfectly systematic and consistent. This difference is of great importance for the design and use of IT systems.
|Premises||All teachers are human beings||All cars are green|
|Some human beings have cars||The sky is a car|
|Conclusions||Some teachers have cars||The sky is green|
Deductive reasoning means that you go from the general to the specific. You draw conclusions about the specific based on knowledge about the general, as illustrated in the two examples above.
Most people would agree that some teachers have cars, given the fact that teachers are human beings. Few would, however, agree with the second conclusion, that the sky is green, knowing that the premises are all wrong. A computer, on the other hand, would come to contrary conclusions. The premises that all teachers are human beings and that some human beings have cars, do not provide sufficient ground to assume that some teachers have cars. The sky, on the other hand, is indeed green, given that the premises are true. These differences in thinking and reasoning give rise to a number of communication problems between the computer and the user. The computer complies with the format of the information, whereas human beings use the meaning of the information when reasoning about it.
Test 1: Confirmation bias
Human beings use stored knowledge when solving problems. Without that knowledge, we would not be able to reason about and solve problems. Sometimes our previous knowledge becomes an obstacle to problem solving. For instance, we tend to use our stored knowledge, even when we know it is inappropriate or inadequate for solving the present problem (Automation). The Stroop effect illustrates this limitation in our problem solving ability, see test below.
We also tend to see only the well-known characteristics of the situation or object, for instance, how we would typically use a particular object (Fixation). See the links on functional fixedness listed below in the Links section.
Moreover, our motivation may stop us from thinking and reasoning clearly about a problem. Experiments have shown that a reward makes people perform worse not better. Subjects who are promised a reward for solving the functional fixedness problem within a specified period of time, actually do worse than subjects who are just asked to solve the problem, without any rewards.
All these obstacles may create communication problems between a user and a computer if, for instance, the information on the computer
Test 2: Stroop effect
Human decision making is not based on the calculation of probabilities. We use empirical knowledge, practical experiences and guessing.
Computers, on the other side, use algorithms to systematically calculate probabilities.
When human beings make decisions weighing uncertain options against one another, we do not calculate probabilities related to the different options. Instead we base the decision on factors that help us decide quickly, but with the drawback that they may lead us in the wrong direction. These factors are called heuristics, and human decision making is heuristic to its nature. Computers use algorithms to calculate probabilities.
Flipping a coin - which series of head or tails is the most likely to occur?
|A||H T H T H T H T H T|
|B||H H H H H H H H H H|
|C||H T T T H T H H T H|
Would you bet on head or tails, knowing the previous results? How certain can you be?
One such heuristic that we use in decision making is the representativeness heuristic. The above example shows three series of head and tails. If you want to make a bet, in any one of the three cases above, what would be the safest bet? Would any of the bets have an exact probability of 50%? Would there be any case where the probability would be greater than 50%? Most people would favour one of the options (head) in the A series and the B series (unless you think that it is high time for tails). In the C series, however, most people would assign equal probabilities to head and tails.
The computer would assign 50% probability to head and tails respectively in all three series. This is also the correct answer according to statistical calculations, given that there is nothing wrong with the coin. Each flip of the coin is independent of any previous and future flips, and the probability of head is equal to that of tails, i.e. 50%. Human beings tend, however, to look for patterns that may help us come to a decision. We believe that series such as the ones in A and B are unlikely to occur. Therefore, we believe that our bet is "safer" in series A and B above, than the bet we make in the C series.
In addition to the representativeness heuristic we use when we make decision, there are also the Availability heuristic and the Conjunction fallacy.
The availability heuristic means that we use the information which is uppermost in our mind when judging probabilities. If you, for instance, ask people to compare the number of words starting with an "r" and those that have an "r" in the third position, most people would claim that there are more words starting with an "r". The truth is that there are more words that have an "r" in the third position. A computer would never make that kind of mistake. The reason that we make a wrong estimate is that we store and retrieve the words based on the first letter. It is easier for us to come up with words starting with an "r" than words with an "r" in the third position.
The conjunction fallacy describes our tendency to assign greater probability to the simultaneous occurrence of two events, than the occurrence of a single event. For example, is it more probable that an area will be flooded or is it more probable that an earthquake will take place in the same area and it will be flooded? Linking two events like that provides an explanation and a coherence that we need in order to make the events meaningful. The probability for the single event is, however, greater than the probability for the two events occurring at the same time, statistically speaking. A computer would come to the correct conclusion since it would multiply the probabilities with one another, and not add the numbers, like we do.
From the above discussion, you may think that human reasoning is completely irrational. This, is, however, not the case. On the contrary, our behaviour is perfectly rational seen from a practical point of view. Our reasoning ability is limited by nature. We have neither the time nor the capacity to make calculations in the same way as a computer does. We want to come to a quick decision that we may act upon. In order to reach such quick decisions, we need order, coherence and confirmation. We need an answer to our problem, not a perfect solution.
As described above, we are not completely irrational - we can in fact reason in perfectly rational ways at times. We can also identify and correct errors, either afterwards so that we can do something about them at a later time or straight away so that we can correct them immediately. Moreover, we can think "in groups" - if somebody makes a mistake, the other people in the group can discover the error so that it can be corrected. In this way we can act in accordance with our intentions and reach our goals.
Knowledge, as cognitive models, can be categorised into structural models and functional models. Within HCI, we discuss the differences between these two types of models. A structural model describes, for instance, how a computer works, representing a "technical" view. Functional models describe what to do and how to use the computer. These two different types of models may be contradictory. Very often, information on the computer or training is organised in accordance with the structural model. We tell the users what's "behind" the screen. This kind of information is not particularly useful to the user although it is a lot more useful to the computer scientist or programmer creating the computer or system. The user needs to learn how to use the computer, i.e. he/she needs information that helps create a functional model of the computer. Thus, manuals, training, online help, etc, should be organised in accordance with the functional model of the computer or system.
The above provides a brief introduction to cognitive psychology. In order to complete the picture you should read the literature and visit the links listed below.