Paper Reading #32: Taking Advice from Intelligent Systems, The Double Edged Sword of Explanations

Kate Ehrlich, Susanna Kirk, John Patterson, Jamie Rasmussen, Steven Ross, Daniel Gruen represent IBM Research in this paper.

This paper was presented at IUI 2011.

Summary


Hypothesis
The hypothesis of this research paper was that when advice from an intelligent system is available to a user, the user tends to make different decisions.

Methods
In order to verify their hypothesis, the researchers set out to create an intelligent system that could classify various types of cybersecurity incidents. This system was known as NIMBLE (Network Intrusion Management Benefitting from Learned Expertise). Based on previous information, the NIMBLE system would attempt to classify new cybersecurity events. NIMBLE was also able to offer explanations as to why it chose the even that it did.

To complete the study, researchers found participants who were highly trained in the field of cybersecurity. All of these participants had a minimum of three years in the cybersecurity and network management field.

The researchers had these professionals complete 24 timed trials. The participants had 2 minutes to guess what type of event occurred and categorize it. NIMBLE was available to assist the participants.

Results
Researchers found that in the cases where NIMBLE suggested an answer, if the correct answer was available, correct response accuracy high. It was even higher when NIMBLE offered a justification to it's suggestion.

However, the interesting part of this study was when there were no correct choices available listed by NIMBLE. If NIMBLE offerred a justification (even though it was incorrect), researchers went with NIMBLE's suggestion most of the time.

Discussion
The results of this study is very interesting. It shows, in my opinion, that if an intelligent system presents information like it knows what it's talking about, we humans will follow what it says.
This can be very dangerous. For example, if we have an intelligent system diagnose humans, much of the time, the intelligent system will be correct. But this study shows we still need doctors to think critically. If a machine pretends to know what it's talking about, when it is incorrect, the doctor still needs to be able to form his own opinion about this.

Paper Reading #31: Identifying Emotional States using Keystroke Dynamics

Clayton Epp, Michael Lippold, and Regan L. Mandryk all represent the Department of Computer Science of the University of Saskatchewan.

This paper was presented at CHI 2011.

Summary


Hypothesis
The researcher's hypothesis in this paper was that they could identify a user's emotional state by simply using keystroke dynamics

Methods
To test their hypothesis, researchers recruited participants for the study and had them install a piece of software that would run in the background. This piece of software would measure key duration and key latency throughout the user's day-to-day activities on their PC.

The software would prompt the user through the day and ask the user how they were feeling at a certain time by asking them various questions in an emotional questionnaire. The user was then asked to type a randomly select portion from Alice in Wonderland.

The results of this entire study were put together and analyzed in an attempt to discern the user's mood through their typing habits.

Results
Researchers selected multiple features of typing sets to focus on when predicting mood. This includes keystroke duration and keystroke delay for various different keys in each key event chain.

They evaluated different models based on the user emotional state and the features of the key-entry during that area of time.

They discovered that when using these models, they were able to classify many of the emotions that were selected about 80% (and over) of the time.

Discussion
This is definitely an interesting article that could allow for several different interesting applications. One for example: when the system finds out that you're stressed or focused based on your key entry then it won't display any system notifications or it'll complete other actions to help the user stay focused.

An example of a fun application might be where a user is asked to type a given paragraph and based on how they type the paragraph, music that matches the user's mood will be played.

Paper Reading #30: Life "modes" in social media

Fatih Kursat Ozenc represented the School of Design from Carnegie Mellon University.
Shelly D. Farnham represented Yahoo!.

This paper was presented at CHI 2011.

Summary


Hypothesis
The researcher's hypothesis was that people organize their life into different groups based on social interaction.

Methods
To prove their hypothesis, researchers set out to conduct a design research study. In this study, they investigated how users interact with social media as well as how they interact with different social groups.

With each participant, they had a two hour in-depth interview to attempt to ascertain how the participant interacted and viewed each social circle. One of the exercises in this study was to draw out pictures or graphs of how the participant viewed certain aspects of their life. The goal was to have users visually map out important areas of their life. By looking at these maps, the researchers were able to find similarities between participants.

Another exercise they performed was through participant evaluation of certain scenarios and storyboards.  These scenarios had to do with boundaries and limitations of interactions. Participants were asked what they liked and disliked about each scenario. They were also asked what they might change.

Results
The result of the first study showed that participants preferred two different types of map. The first being a "social meme" type map. This map looked like some sort of graph. The center node of the graph contained the person and had different social circles and activities drawn off the connected nodes. The next time of map was the timeline map. Users drew this map with daily activities in order of occurrence.

As far as communication practices went, researchers found that the participants often used different types of communication media for different social circles. A user might use a certain set of communication channels (facebook, texting, etc) to interact with one of their social circles, while using a completely different communication channel set for another group.

Based on all these results, the researchers also enumerated various design suggestions for creating new social media and allowing for better social interaction with technology.

Discussion
Aspects of this paper area already being used frequently in social media. One large example was Google's been launch of Google+. Google+ allows users to place their "friends" into social circles according to where they fit. Then, users can choose what circle they want to share information with. This is very much in line with what the researchers of this paper were suggesting.

Paper Reading #29: Usable Gestures for Blind People, Understanding Preference and Performance

Shaun K. Kane and Jacob O. Wobbrock represented the Information School of the University of Washington.
Richard E. Ladner represented the Computer Science & Engineering school of the University of Washington.

This paper was presented at CHI 2011.

Summary


Hypothesis
In this papers, researchers attempted to show how blind people interact with gestures on a touch screen device as well as which gestures they prefer.

Methods


This research was broken into two separate studies.

For the first study, the blind participants were asked to invent their own gestures that could be performed on a tablet PC. 20 Participants were recruited for this experiment: 10 people were blind and 10 had sight.  The sight group was used as a baseline to compare against.
The participants, once oriented with the screen, were asked to perform to gestures to complete a list of tasks. Some of these tasks: undo, move up, delete, etc etc.

Each participant created two gestures per command.

The next study was focused on gesture performance. The blind participants and sighted participants were asked to perform the same gestures as a way to ascertain which gestures were easier or more effective for the blind to perform.

They were also asked to rate the easiness of the following gestures: tap, flick, multi-touch, shape, symbol.

Results
For the first study, they found that most of the gestures created by the blind group were more metaphorical and abstract than the sighted group. The blind group also used landmarks or the assumption of a given physical space. For example, they made the "control-v" finger tap movement for the past command.

They also found that the blind group preferred landmarks or performing gestures close to the screen.

In the second study, the blind group was found to prefer multi-touch gestures much more than the sighted group. They also slightly preferred the flick.

Discussion
This is definitely related to the previous reading that we did this week. However, the important distinction is that a blind group was asked to create and rate gestures. The previous research indicated that the preferred and most efficient gesture was the bezel mark. The Blind group also preferred bezel gestures so perhaps maybe in future mobile input, bezel gestures will be used more for both blind and sighted users.

Paper Reading #28: Experimental Analysis of Touch-Screen Gesture Designs in Mobile Environments

Andrew Bragdon is a PhD computer science student at Brown University.
Eugene Nelson also represents Brown University.
Yang Li is a senior researcher at Google Research.
Ken Hinckley is a principal researcher at Microsoft Research.

This paper was presented at CHI 2011.

Summary


Hypothesis
The researchers presented the hypothesis that there are certain ways to control mobile devices through gesture techniques and soft buttons that allow for easier interaction in different ways.

Methods
Researchers tested various different types of input methods to control the phone.

First, they looked at hard button initiated gestures. This means that a physical button mounted on the hardware would signal when a gesture could be drawn on the screen. So the user would have to first push the hard button before they could start the gesture.

Bezel gestures happen when the user starts the gesture off the screen and bring their finger over to the main touch area. This allows the phone to know when to start allowing gesture input.

To allow for a "performance baseline", the researchers also used soft buttons for testing purposes.

They also defined several different types of gestures: paths versus marks.

For testing which gesture types worked better in different environments, they looked at the accuracy and time of the gesture during various motor activity (sitting and walking) and distraction level (no distraction, moderate situational awareness, attention-saturating task).

15 participants completed this study.

Results
Researchers found that soft buttons and gestures worked about the same when the user was looking at the phone. However, when the user wasn't looking at the phone, bezel gestures were the easiest and most accurate.

When distracted, bezel gestures again performed better than soft buttons. Researchers stated that this was encouraging because it means that gestures can actually improve performance.

Discussion

As I mentioned last week, these kinds of studies are important. These kinds of tests ensure that gesture space (or a given input method) is actually truly better than what is currently used by the mainstream groups. Not only does this research prove that gestures can perform better than soft buttons, but it also helped show which gestures would work better.

Paper Reading #27: Sensing Cognitive Multitasking for a Brain-Based Adaptive User Interface

Erin Reacy Solovey, Francine Lalooses, Krysta Chauncey, Douglas Weaver, Margarita Parasi, Matthias Scheutz and Rober J.K. Jacob represent the students and faculty of Tufts University in the Computer Science department.
Angelo Sassaroli and Sergio Fantini represent the Biomedical Engineering department of Tufts University.
Paul Schermerhorn represents Indiana University and their Cognitive Science school.
Audrey Girouard represents Queen's University and their School of Computing.

This paper was presented at CHI 2011.

Summary


Hypothesis
In this paper, researchers set out to prove that it was possible to detect multitasking contexts and to react to these changes.

Methods
The researchers first labelled three different scenarios that describe the brain's ability to multitask. First is branching. Branching is a state where the primary goal still requires attention but the secondary goal must be attended to as well, thus the user must keep certain information regarding the primary goal active in their mind. The second type is dual-task. This is where the user must make quick changes between the primary task and the secondary task regularly. The final type is delay. Delay occurs where a secondary task is ignored when the user is working on the primary task.

Researchers main goal was to determine whether or not they could actually detect when a user's mind was in a certain context.

To test this, they created an experimental task which required a human and a robot to work together over a given task. The robot simulated a Mars rover on an exploration mission. The rover would give status updates about a newly found rock or a new location change. The participant would then have to respond to the robot with some command. There were three separate test runs that each tested a different multitasking mode.

They also conducted a proof-of-concept user interface test where the participant was given a task and based on their multitasking mode, the robot would have different behaviors and interact with the human differently.

Results


Through analyzing the first experiment, researchers were able to extrapolate the way hemoglobin levels worked to determine the three various multitasking states. Below shows the three graphs of the multitasking states:

Since these levels are different, that means that multi-tasking modes can indeed be discovered and the multitasking mode can be predicted in the users mind.

Results

Again, a very fascinating paper on Brain-Computer interfaces. There are many potential uses for a system like this. One system that I'd find very useful is an application that could help keep a user focus during tasks that require concentration. For example, writing a paper or coding. So, if I was writing some code and needed to focus, this BCI would detect I'm in the delay multi-tasking mode and would therefore block out certain notifications that my computer currently sends me (new emails, new facebook notifications etc).

Paper Reading #25: TwitInfo. Aggregating and Visualizing Microblogs for Event Exploration

Adam Marcus, Michael S. Bernstein, Osama Badar, David R. Karger, Samuel Madden and Rober C. Miller all represent student or faculty of MIT CSAIL.

This paper was presented at CHI 2011.

Summary
Hypothesis
Researchers proposed that through Twitter timelines, events could be identified and visualized.

Methods
To complete this study, researchers used Twitter to search through various tweets about a given subject. The TwitInfo algorithm can begin to track certain keywords and when they're used together. It then begins to build a list of tweets that have those keywords mentioned. After these tweets are gathered together, the algorithm builds a list of peaks and determines what words are being used most often in the peaks. These peaks are then classified as events.

The system is also capable of detecting positive and negative sentiment over a given topic over time.

All of this data can be then be organized and displayed in order to better visualize results.

In order to test the user interface of TwitInfo, researchers recruited 12 participants to perform certain tasks to locate an event.

Results


Tested over time, TwitInfo correctly organized most events properly. It was used frequently to analyze the events of a given soccer match. For every match, it's easy to see when the goals were scored due to the peaks in the events.

In the participant study of TwitInfo, participants were quick to reconstruct events when given information from TwitInfo, in many cases, in less than 5 minutes.

Discussion

TwitInfo struck me initially as a very useful, concrete research project. It is a classic example of using humans to complete a given task and then aggregate the data. Essentially, humans, without even knowing it are classifying events and when they occur. This makes TwitInfo's job easier. All the algorithm has to do is collect and analyze the data.