Computer Science and Psychology < Yale University
Cognitive Psychology and Computer Science? cognitivequora. For details, see my previous Quora answers: Can a computer that is designed. So what do psychologists involved in computational social science look like? They observed relationships on Twitter, and found that rhetoric. Computer science is poised to have a tremendous impact on psychology. Besides . In this vision, there is no causal relationship between data collection and its.
Sample projects include an open-ended educational game for physics that tracks student emotion in computer-enabled classrooms, behavioral measures of non-intellective traits like academic diligence and frustration tolerance, and methods to fully automate the collection and coding of classroom discourse with an eye for providing formative feedback to teachers.
Focus on the Process Work on such projects begins in the lab with precise experiments that are recorded in detail. Once the models can accurately replicate human outputs, he and his team go back to the human side of the equation by asking more questions—a true blend of psychology and computer science. Then it all repeats. He has virtually created a whole new field of study with his research on affective computing.
His interest in robotics and artificial intelligence connects our work with computer science and engineering—which just goes to show that cutting-edge, field-expanding research will be profoundly interdisciplinary, and Sidney is leading the way.
Your system is in a state of shock that makes you look at things carefully and to problem-solve. The perception that negative emotions are always bad is absolutely is not the case. Imagine technology that tracks affect and attention and knows when to intervene for an air traffic controller, pilot, or ship captain. Imagine 1-on-1 online, adaptive tutoring for college students that supplements every class experience. Besides experiments and questionnaires, it establishes a third fundamental research technique: It allows psychologists to analyze variables such as personality traits e.
Tracking hundreds of thousands of users, the resulting Big Data requires substantial modeling and cleaning. However, its sheer size in combination with machine learning techniques leverages statistical power we refer to problems with false positives later on. Most importantly, it avoids most sources of bias, because the behavior of interest is directly recorded.
Many biases are inherent to standard psychological measures, for example, the tendency to answer self-report measures in a socially desirable manner e. Yet, the approach pioneered by Psychoinformatics also poses significant challenges to the two sciences involved. Most importantly, the two must learn to cooperate and ultimately shape an entirely new discipline: Psychoinformatics [ 3 ].
Traditionally, the psychological sciences rely on two fundamental methods of data collection: The former investigate one very particular aspect in a small and entirely controlled setting. The latter assess the broader behavior of a person by means of self-report questionnaire or potentially structured interviews [ 6 ]. These methods suffer inherent shortcomings.
Experiments are usually limited to a single data point i. Clearly longitudinal experiments also exist, though these are conducted less frequently due to the high cost and effort involved.
Self-report questionnaires and interviews also encounter problems, since humans find it hard to reliably recollect past events, and they are additionally subject to various sources of bias e. In contrast, modern computer science introduces entirely new methods of assessing participants' behavior longitudinally, on large scale, and in comparison to self-reports, in a rather objective manner.
Computer science as a discipline is largely concerned with implementing algorithms using computers or similar devices. Over the past twenty years, the computer industry has produced a large range of powerful technologies, which have become ubiquitous in everyday life. Smartphones and other mobile devices provide constant connectivity and in doing so have changed our daily lives [ 7 — 9 ]. Together with online platforms such as Facebook, they have become a central venue to communicate, shop, play, or study.
As a consequence, digital technologies are pervasive in everyday life and data from such devices could be recorded on a large scale. Finally, cheap hardware allows us to store and analyze large amounts of data at little cost.
These new technical innovations provide support for classic psychological methods, such as experiments and questionnaires [ 10 ]. First, they enable psychological experiments to be implemented through mobile phones [ 11 ]. In the latter study by Dufau et al.
As discussed below, this new way of conducting experiments and gathering data needs to be compared with data acquired through classic experimental setups to ensure that data of equal quality can be achieved through Psychoinformatic methods.
Is it feasible that neuropsychological tests and other classic test batteries may be implemented on smartphones and be studied not only in patients but also in the broad population? Psychoinformatic experiments can be conducted several times per day over an extended period of time, thus generating a larger number of data points per user. Second, they allow for questionnaires to be administered over mobile phones, potentially asking the participant to contribute data on a daily level, again collecting more data points per user [ 12 ].
Here, an interesting variable could be the assessment of mood or the inclusion of experience sampling to assess flow activities in everyday life the flow concept is explained a bit later in the paper; [ 13 ]. The basic shortcomings of both methodologies will, however, remain. Only a limited number of users can be incentivized to regularly conduct an experiment, and questionnaires remain a source of bias though, of course, self-report inventories will always be of importance in psychology, e.
However, data collection has already benefited from these technologies, for example, easier data processing enabled by the switch from paper-pencil questionnaires to questionnaires administered online, which eliminate errors in recording participants' responses [ 14 ].
This point is discussed in more detail in the section on data cleaning. Electric sensors have improved significantly and pose another powerful technology for assessing the condition and behavior of humans. They can measure physical movement via accelerometers [ 15 ], galvanic skin response [ 16 ], or heart-rate variability [ 1718 ].
Toward Psychoinformatics: Computer Science Meets Psychology
Over the past ten years, they have become very cost-effective and they require little maintenance by the participant. First, sensors can send their data automatically to a server via a smartphone. Second, efficient processors and powerful batteries have dramatically reduced the need to charge sensors [ 19 ]; current fitness trackers, for example, run an entire week on a single charge.
The rapid development of technologies gives way to the Internet of Things IoTwhere everyday things such as coffee machines or the fridge are connected to the Internet see also below and can serve as data sources. As outlined above, the main methodological advantage Psychoinformatics offers over classic psychological techniques is the ability to track human-machine interaction directly on the device.
For example, one can track the interaction between a user and their smartphone [ 20 ] or smart car [ 21 ]. This approach can also be extended to online platforms, such as social networks [ 22 ] or shopping sites [ 23 ]. Data is captured and transferred to a central server for further analysis, without requiring any interaction from the user. Such tracking outperforms traditional methods in terms of both the scale and quality of the data collected.
First, it allows researchers to track a very large number of participants, up to hundreds of thousands. Second, it collects numerous data points per day, without demanding anything from the participant.
As people increasingly move their lives online, potential data sources become ever richer, ultimately providing more data points per day. Simultaneously, such data sources become ever more plentiful, as our environments become increasingly digital. Soon, we will be able to track interaction with smart cars [ 24 ] and coffee machines [ 25 ].
Meanwhile, it has become mainstream and denotes the corresponding research area in computer science [ 27 ]. In an even broader vision, the Internet of Things IoT or the Internet of Everything refers to a world, where every item is represented and every process is conducted digitally or at least documented digitally.
Necessitating a globally agreed upon set of standards, the IoT thus forms something of a semantic infrastructure. Every device in this world produces data, documenting its actions. The storage and analysis of this data is commonly referred to as Big Data. In this vision, there is no causal relationship between data collection and its analysis; that is, data is commonly analyzed to answer questions that were only vaguely known, if at all, at the time of data collection.
Of course, this approach yields the danger of false positive results, particularly in light of the many variables of interest to be gathered via recording of human-machine interaction, resulting in endless opportunities to search for significant correlations.
Therefore, independent replication of results observed from Psychoinformatics data sets and carefully designed follow-up experiments laboratory-based will be necessary.
There are numerous visions of how digitalization may shape our world. As an initial point for further reading, we refer readers to the seminal works by Rifkin [ 2728 ] and Brynjolfsson and McAfee [ 29 ].
Yale College Programs of Study 2018–2019
However, as the methodology generates so much data on so many users, the signal should separate from noise more clearly than ever. For example, take a researcher interested in the investigation of cognitive functions, who wishes to assess cognitive function by studying the changing size of the word pool of a person's language. If the researcher only considers word use across one day, the data set is unlikely to be very representative. However, by analyzing this person's word use over a longer time window, the standard error of the measure decreases, because digital interactions with a larger number of people can be included in the analysis.
Finally, ubiquitous tracking avoids most sources of bias inherent to questionnaires. Tracking user interaction directly—for example, on a smartphone—remains subject to certain forms of bias the feeling of being monitored might change the behavior of a person.
Yet, these are much less than that present in experiments or questionnaires. Moreover, after a short while, participants should no longer think about the fact that they are being tracked.
This clearly needs to be tested empirically, but we can think about this using a highway analogy. After a while, however, the noise is filtered out by the human brain and some people will no longer be aware of it [ 3031 ].
Of course, there is a big difference between awareness of traffic noise compared with being tracked by another person. Nevertheless, the success story of online social networks such as Facebook demonstrates that a large number of people are not overly concerned about their digital privacy at least after a while ; otherwise, they would reconsider their open profiles, and so forth. Tracking behavior on the smartphone is likely to lend the greatest insight into human behavior.
It captures various aspects of life via a wide range of methods movement patterns via GPS and text mining to infer mood, communication patterns, and size of the social network [ 3233 ]. It is loaded with sensors.
It can communicate its data autonomously to a remote server. It serves as the central device to access the web, shop online, communicate with friends, and play games.
And, importantly for research budgets, most people already own such a device. With this enormous distribution of smartphones worldwide, they are predestined to turn into the most prominent data source for scientists [ 35 ]. The inherently different data characteristics derived from the human-machine interaction require an entirely different mentality from researchers.
Big Data, such as that generated by means of ubiquitous tracking, is commonly characterized by the three Vs: Data arrives at a very high rate, in various formats and qualities, necessitating substantial means of storage.
This data is inherently flawed and dirty. Yet, as indicated above, signal should separate from noise clearly due to the massive amount of data points collected. While researchers of course need to check up on the collected data see data cleaning a bit further down belowthey must also sacrifice the kind of control they traditionally have in a strict experimental setup.
Instead, they need to rely on the statistical power of a large number of measurements. Frequently, this form of research will rely on data that has been collected for entirely different purposes. For example, a researcher might analyze the logs of a social network.
Or they might utilize the billing information of a telecommunication provider. Any such approach, common to Big Data applications, shifts research to post hoc analysis. The scientific question at hand has no influence on the data collection.
As a matter of fact, the question might not have arisen at the time the data was collected. This raw data, obtained via diverse applications, requires extensive processing. Initially, it is often cryptic and eludes analysis. It thus necessitates significant modeling before it can be analyzed.
Thus, there may be many more processing steps, including various forms of data cleaning. This data cleaning processes will largely depend on the unique research question under investigation. Consider a study on productivity issues in digital work environments. One could hypothesize that because more interruptions are observed, less productivity should be observable, owing to disturbance of the aforementioned experience of flow in one's work.
Flow represents a state of high productive concentration, in which a person's skill is matched with the difficulty of a task. Smartphones can distract us to a point where reaching a state of flow becomes impossible. The study would thus focus on interruptions due to smartphones in everyday life. Therefore, the computer scientist might model how often a smartphone is flicked on and shut down.
This modeling process must thus take many things into consideration. Is it more interesting to assess the length between phone sessions?
Or should we calculate the general time spent on a smartphone each day? Should we count time, when the phone is used to listen to music, but not interactively?
How should ultrashort smartphone sessions be handled, for example, where the phone's screen is flicked on, but the phone is not unlocked, and there is no further haptic interaction?
The precise research question at hand will determine data cleaning and modeling.
Toward Psychoinformatics: Computer Science Meets Psychology
And any solution will require close interdisciplinary collaboration. Naturally, there have been previous collaborative efforts between the areas of psychology and computer science.
In particular, Human-Computer Interfaces HCI denote the area of computer science concerned with the interaction between users and electronic systems, for example, by means of graphic interfaces or acoustic signals. This research direction thus comprises usability engineering, e-learning, interaction, and information design, among others.
Immediately addressing the user, it touches many areas commonly studied by psychologists. In particular, the discipline of affective computing recognizes, reacts to, or mimics human affect [ 37 ].
For an introduction, see http: More narrowly focused, Human-Robot Interaction focuses on the interface between users and humanoid robots, thus also touching on aspects of psychology.