You are on page 1of 7

Physical Cyber Social Computing for Human Experience

Amit Sheth Pramod Anantharam


Ohio Center of Excellence in Ohio Center of Excellence in
Knowledge-enabled Computing (Kno.e.sis) Knowledge-enabled Computing (Kno.e.sis)
Wright State University Wright State University
Dayton, OH, USA Dayton, OH, USA
amit@knoesis.org pramod@knoesis.org

ABSTRACT development of algorithms and techniques that can act on


Computing has a critical role in solving some of the grand a specific representation of real-world problems [6]. A vast
challenges spanning health and wellbeing, sustainability, and majority of current approaches address a sliver of the real
prevention of crime. Traditionally, computing has focused problem, primarily focusing on limited forms of data and
on a single narrow view of the problem to provide solutions its analysis. An often used approach is to use a limited
ignoring the essential human experience component. The form of data as input to the algorithms in a reactive manner
availability of low-cost sensors and mobile devices leading to (e.g., search query). However, they miss out on the richness
simultaneous and continuous access to events spanning phys- of human-level information processing which can be repre-
ical, cyber, and social worlds demands rethinking of the tra- sented as abstractions over multimodal and multi-sensory
ditional computational approaches. We propose Physical- data. With over 40 billion devices and sensors expected
Cyber-Social (PCS) computing, that takes a human centric to be connected to the Internet by 2015, and an extensive
and holistic view of computing by analyzing observations, amount of collectively created and curated information on
knowledge, and experiences from physical, cyber, and social the Web (e.g., Wikipedia and Linked Open Data), we now
worlds. We exemplify real-world problems that demands have a deeper and continuous view of the events that sur-
the approach of PCS computing and outline the research round us. This trend is accelerated by the use of smart-
challenges in building algorithms and techniques for PCS phones which record fine grained observations such as phys-
computing. ical movements/activity along with low-cost devices report-
ing observations to the phone or directly via the Internet
(aka, Internet of Things - IoT). We focus on the rapidly
Categories and Subject Descriptors evolving ecosystem of computing encompassing physical (sen-
H.4 [Information Systems Applications]: Miscellaneous sors/devices/IoT captured observations), cyber (Web-based
information/knowledge and tools, etc.), and social (social or
human-in-the-loop sensing, mobile interactions, etc.) com-
General Terms ponents, what we call Physical-Cyber-Social (PCS) comput-
Theory, Algorithm ing, to develop advanced algorithms and applications that
act to enhance human experience. PCS computing incor-
Keywords porates physical, cyber, and social components to assist or
interact with humans in a palatable way using higher level
Physical-Cyber-Social Systems, Machine Perception, Social abstractions. The technology itself is expected to disap-
Computing, Semantic Computing, Computing for Human pear into the background [17]. We have defined this human
Experience, Healthcare, Traffic Analytics, Intelligence centric vision of computing as Computing for Human Expe-
rience (CHE) [13]. CHE emphasizes the unobtrusive, sup-
1. INTRODUCTION portive, and assistive role of technology in improving human
The societal challenges such as healthcare and wellbeing experience, so that technology takes into account the hu-
(e.g., hospital readmission, assisted living), sustainability man world and allows computers themselves to disappear in
(e.g., traffic, water, energy), crime prevention and mitiga- the background [17].
tion (e.g., intelligence applications) needs a computational
core for observation, comprehension, and action. Most of 2. FROM SEARCH TO SOLUTION, FROM
the existing computational techniques have focused on the
DATA TO MEANING
Using a massive amount of multimodal and multi-sensory
observations collected in a PCS system1 , the computational
Permission to make digital or hard copies of all or part of this work for approaches should move toward proactive creation of ques-
personal or classroom use is granted without fee provided that copies are tion representations, for example, early detection and man-
not made or distributed for profit or commercial advantage and that copies 1
bear this notice and the full citation on the first page. To copy otherwise, to In conjunction with PCS, we use system to refer to em-
republish, to post on servers or to redistribute to lists, requires prior specific bodiment in software and its components, and use com-
permission and/or a fee. puting to refer to algorithms and techniques including the
WIMS13, June 12-14, 2013 Madrid, Spain operators discussed later, that act upon observations, data,
Copyright c 2013 ACM 978-1-4503-1850-1/13/06... $10.00. and knowledge.
Figure 1: Comparison of a search engine such as Google vs. an answer engine such as WolframAlpha used
to find information about a symptom of heartburn

agement of an impending heart problem, detection of on-


set/progression of Parkinsons Disease, early detection of Table 1: Comparison of traditional computational
traffic delays, and early detection of crime. All these chal- paradigm with PCS for Human Experience
Application, Traditional Computational PCS Computing for Human
lenges require us to ask contextually relevant questions over Activity Paradigm Experience
of Human
multimodal and multi-sensory observations. One form of Interest
Healthcare threshold based monitoring actionable information in
deriving question representation is the use of background the form of risk score with
knowledge of the domain in asking contextually relevant Intelligence monitoring raw data from
explanation
actionable information in
questions [11, 10]. For example, observing elevated blood surveillance cameras, mobile the form of threats obtained
phone logs, open source in- from threat modeling and
pressure may result in a question such as as what are the telligence assessment
Traffic Man- monitoring speed of vehicles, actionable information to
reasons for elevated blood pressure?. These representations agement prediction of delays mobilize units based on
are actively generated by observing the state of the person event types causing delays
and worst affected locations
rather than waiting till this question is asked in a reactive
manner. The actionable information that a human needs
will be such as reduce the salt intake or try reducing your
stress level at work depending on the result of processing such as demographics, prescribed medication, co-occurring
observations from PCS system. symptoms, and connection to smoking and obesity. While
Consider the following example, when Mark experiences Mark can quickly glean the knowledge of heartburn symp-
heartburn, he translates the information need from the phys- tom and its causes, he still needs to make a connection be-
ical world (the physical experience of discomfort due to heart- tween his personal activities and observations to population
burn) into a search query heartburn. Figure 1 shows an level statistics. Mark is still not sure whether to schedule an
attempt by Mark to find information about the symptom appointment with his physician or visit the ER.
heartburn using a search engine such as Google vs. an an- We need a computational framework that can proactively
swer engine WolframAlpha. When using Google, the search sense the discomfort of Mark, ask contextually relevant ques-
for heartburn results in a ranked list of relevant documents tions, and personalize the analytics leveraging background
from authoritative sources such as WebMD and community knowledge. Mark needs some actionable information rather
knowledgebase such as Wikipedia. While the information than relevant documents describing his symptoms or a popu-
from authoritative sources are useful, it does not explain lation level statistics of his symptom. Beyond understanding
why Mark has a heartburn. Mark has to combine informa- the reason for his heartburn, Mark needs to act on it by tak-
tion about possible causes of heartburn (gleaned by read- ing appropriate measures such as decrease fatty food intake,
ing and abstracting from authoritative sources) and his past have timely food intake, schedule doctors appointment, or
activities (in the physical world) to explain his heartburn make arrangements for an ER visit (actionable information).
making a personalized interpretation. Comparison of the actionable information provided by
An answer engine such as WolframAlpha provides Mark PCS computing with the traditional computing paradigm
with lot more relevant information such as population level is presented in Table 1. Providing actionable information
statistics of heartburn cases in the US based on parameters requires the use of background knowledge, semantic inte-
gration, interpretation, and abstraction.
Figure 2: PCS operators for semantic integration and semantic abstraction to transcend from massive data
to actionable information an asthma scenario

3. PCS COMPUTING source of the observations, and (2) the knowledge base is
The central idea of PCS computing [14] is to provide ac- not limited to a formal ontology - it spans from statistical
tionable information in the form of abstractions that puts knowledge to social experiences.
human at the center of computing by utilizing knowledge,
observations, and experiences from physical-cyber-social sys- Figure 2 exemplifies transcending from multimodal and
tems. multi-sensory data to actionable information for asthma at-
There have been different computing paradigms ranging tack prevention and mitigation. Mark has heartburn which
from the focus on making machines intelligent to the fo- is not an obvious indication of an asthma problem unless we
cus on human experience. Artificial Intelligence (AI) fo- have the contextual information of Marks history of asthma
cuses on building machines that can simulate human intelli- and the domain knowledge that states the relationship be-
gence. Ambient Intelligence makes the surroundings of hu- tween heartburn and asthma [8]. Given that heartburn is a
mans an intelligent space. Human Computer Interaction generic symptom (heartburn is a common symptom of many
(HCI) builds seamless interfaces between humans and ma- ailments), any search or answer engine will provide informa-
chines. Computing for Human Experience [13] is a human tion that may not be immediately relevant to Mark. The
centric vision of computing where the focus is on using com- association between Marks house windows and the level of
putation for enhancing human experience. Ambient Intel- CO2 in the house may not be immediately observable by
ligence is more machine centric compared to HCI as HCI humans. While the action of closing windows before leav-
puts human experience first and machine intelligence next. ing for work would be the actionable information that a
Table 1 provides a comparison of solutions offered by tradi- human expects to know. This actionable information is
tional computational paradigm and PCS computing. a contextual (e.g., open windows) and personalized (e.g.,
PCS Operators: The computational algorithms for PCS Marks asthma history) interpretation of observations span-
computing can be thought of as operators on data for seman- ning physical (e.g., open window), cyber (e.g., knowledge of
tic integration, annotation, interpretation, and abstraction. CO2 causing asthma aggravation), and social (e.g., report
Two such operators are described here (Figure 3). of air pollutants/asthma attacks near Marks house) worlds.
Horizontal Operators: Horizontal operators provide opera-
tions for semantic integration of data from multisensory and 4. PCS COMPUTING EXAMPLES
multimodal observations. These operators play a key role in
PCS systems as the observations may span from machine
sensors (quantitative observations) to citizen sensors (qual-
4.1 Intelligence Scenario
itative observations). One modality may complement the Attacks on physical, cyber, and social infrastructures have
other or provide corroborative evidence for the other. been on a constant rise [1, 2, 5, 4]. Intelligence agencies
Vertical Operators: Vertical operators provide operations should perform simultaneous (multi-sensory) and continu-
for transforming massive amounts of data to abstractions ous (real-time) monitoring for identifying malicious activi-
that are palatable to people. These operators are unique in ties. Forensic analysis of these observations is a reactive way
the following two ways: (1) the operator is agnostic of the of solving the problem. Proactive assessment of risks based
on observations from multisensory and multimodal sources
Figure 3: Example intelligence scenario with questions (from decision makers) and utilization of multimodal
and multi-sensory observations in answering these questions

is crucial for enhanced situational awareness for protection Since Mark has a history of asthma, exposed to CO2, and
of integrity and lives of humans. Conventional threat de- experiencing heartburn symptom after having his meal,
tection techniques may rely on one modality, resulting in the most likely reason for his heartburn can be asthma [8].
an incomplete picture of the threats. Often decision makers The actionable information in Marks case is to undersand
are inundated with lot of information and it is challenging that the CO2 level at home is high and the windows are left
to focus on the most important piece of information. An- open during the day. The action to be taken by Mark is to
swering some of the questions (by a decisionmaker) posed close the windows (action in the physical world) of his house
in Figure 3, requires gathering evidence from heterogeneous to reduce the ingush of CO2 along with setting up an ap-
sources (machine sensor to citizen sensor observations). The pointment with Marks doctor. Action of Mark will reduce
knowledge that may be required in answering each question his asthma attacks at home and addresses the root cause
is listed along with the question in Figure 3. of the heartburn symptom leading to an enhanced human
experience.
4.2 Healthcare Scenario
Healthcare has great inertia in being reactive in treating 4.3 Traffic Scenario
ailments/diseases leading to an expensive healthcare system Traffic jumped 236% while the population increased by
which constitutes 17.6% of GDP for 2010 in the US. This only 20% in the US from 1982 to 2001 [3]. Indian cities
type of healthcare system is not affordable to the increas- had over 285 million people by 2001, which is almost the
ing and aging world population [16]. We envision a proac- population of the entire United States [12]. This trend of
tive and personalized healthcare by analyzing contextually concentrated population in cities is only going to get worse
relevant observations. With the increasing adoption and with depleting resources. We need a deeper understanding
decreasing cost of sensors and mobile devices for monitor- of the sustainability problems such as traffic management,
ing health and wellbeing [7], we envision the transformation water, energy, and other resources. Solutions for these prob-
of healthcare from reactive to proactive and personalized lems should not be confined to analysing one modality re-
healthcare. PCS computing will take massive amounts of sulting only in a single perspective of the problem. We need
multisensory (e.g., heart rate, blood pressure, weight) and a holistic approach to this problem which can semantically
multimodal (e.g., machine sensors, textual observations from integrate observations and derive insights from the data del-
EMR, social media interactions) observations, and perform uge. Figure 5 shows traffic related observations from 511.org
contextual and personalized analytics. With proactive care, which has both machine sensor and textual observations.
Mark (from the example introduced earlier in search to solu- The complementary nature of observations are exemplified
tion section) need not search for heartburn on a computer. in the figure. Slow moving traffic is observed by the machine
PCS computing will identify the physical discomfort of Mark sensors on a particular link (link is a unit that constitutes
through a galvanic skin response sensor and proactively ask a a road). Examining the scheduled events list (provided by
contextually relevant question to Mark (Figure 4). Further, 511.org) shows that there is a baseball game near the link.
based on his answer, the PCS computing will provide action- The schedule information of the venue indicates a scheduled
able information to Mark (e.g., risk score on a scale of 5). event of Mariners playing a baseball game at the coliseum
Figure 4: PCS computing vision of providing personalized, actionable information proactively by considering
contextually relevant observations from physical, cyber, and social modalities

Figure 5: Traffic observations spanning machine sensors, authoritative reports of traffic incidents, and event
schedule information
and personal/social observations has led to a deeper view
into our physical, cyber, and social worlds. The data gen-
eration rate has surpassed our ability to store all the obser-
vations [9]. We envision PCS computing to derive insights
from these observations to provide actionable information
to humans. Providing actionable information by taking a
human centric approach is the vision of PCS computing.
We exemplified the need for research toward a new com-
putational paradigm while we embark on solving some the
greatest challenges faced by humanity.

Figure 6: Hierarchy of PCS computing challenges


depicting a broader view beyond computational
7. ACKNOWLEDGMENTS
challenges Much of the work discussed here results from the work of
our Semantic Sensor Web team, of which Cory Henson and
Prof. T.K. Prasad are an integral part. We also acknowledge
complementing all the other modalities. partial support from NSF awards IIS-1111182: SoCS: Social
Understanding traffic events would allow decision makers Media Enhanced Organizational Sensemaking in Emergency
to ask questions such as: What are the common event-types Response and IIS-1143717: EAGER - Expressive Scalable
causing delays in San Francisco (SFO)? What are locations Querying over Integrated Linked Open Data. Any opinions,
worst affected by delays in SFO? Which location in SFO findings, and conclusions or recommendations expressed in
needs attention when there is poor visibility? this material are those of the author(s) and do not necessar-
ily reflect the views of the National Science Foundation.

5. KEY RESEARCH CHALLENGES


8. REFERENCES
Figure 6 organizes the research challenges for PCS com-
puting into computational challenges and interdisciplinary [1] Boston bombing on April 15th. http://en.
challenges. wikipedia.org/wiki/Boston_Marathon_bombings,
Computational challenges include the algorithmic chal- April, 2013.
lenges in dealing with observations from the PCS system. [2] Dramatic increase in the attacks on cyber
Horizontal operations: Horizontal operations should be ca- infrastructure.
pable of dealing with semantic heterogeneity for semantic in- http://www.homelandsecuritynewswire.com/
tegration, interpretation, and understanding of multimodal dramatic-increase-critical-infrastructure-
and multi-sensory observations: (a) multimodal: data, knowl- cyber-attacks-sabotage, April, 2013.
edge, and experiences span physical, cyber, and social modal- [3] IBM Smarter Traffic initiative:. http://www.ibm.com/
ities, (b) multi-sensory: data from a variety of sensors, in- smarterplanet/us/en/traffic_congestion/ideas/,
cluding those that capture human senses (vision, sound, April, 2013.
touch, and smell). [4] McAfee global report on the threats facing key
Vertical operations: Vertical operators should be capable of industries.
dealing with a massive amount of multimodal and multi- http://www.mcafee.com/us/resources/reports/rp-
sensory observations and their semantic interpretation to in-crossfire-critical-infrastructure-cyber-
generating human intelligible abstractions. Change is in- war.pdf, April, 2013.
evitable in the real-world (Physical) and these abstractions [5] McAfee second annual infrastructure protection
change continuously, reflecting changes in the real-world. report.
Research on Continuous Semantics [15] deals with the ability http://www.mcafee.com/us/resources/reports/rp-
to understand and incorporate new entities/concepts with critical-infrastructure-protection.pdf, April,
respect to a situation/event. 2013.
PCS computing is a broader vision and needs the inter- [6] Physical-Cyber-Social Computing Systems.
disciplinary challenges to be addressed. Human interactions http://ngs.ics.uci.edu/blog/?p=1501, April, 2013.
are an integral part of computing, where humans can guide [7] Quantified Self, self knowledge through numbers.
the computation, provide missing information that the sys- http://quantifiedself.com/, April, 2013.
tem seeks, geared toward understanding of the situation.
[8] Relationship between heartburn and asthma. http:
The AI, Intelligent Systems, Semantic Web, and Sensor Web
//www.webmd.com/asthma/guide/heartburn-asthma,
communities should embrace research in other domains such
April, 2013.
as Social, Cognitive, and Brain & Neuroscience for develop-
ing PCS computing algorithms. [9] Sensor Networks top Social Networks for Big Data.
http://gigaom.com/2010/09/13/sensor-networks-
top-social-networks-for-big-data-2/, April, 2013.
6. CONCLUSIONS [10] C. Henson, A. Sheth, and K. Thirunarayan. Semantic
Some of the grand challenges in healthcare, sustainability, perception: Converting sensory observations to
crime prevention and mitigation require a holistic approach abstractions. Internet Computing, IEEE, 16(2):2634,
to computing for providing actionable information. With 2012.
the increased digitization of the physical world culminating [11] C. Henson, K. Thirunarayan, and A. Sheth. An
in a massive data generation from sensors, mobile devices, efficient bit vector approach to semantics-based
machine perception in resource-constrained devices. In
The Semantic WebISWC 2012, pages 149164.
Springer, 2012.
[12] J. Pucher, N. Korattyswaroopam, and N. Ittyerah.
The crisis of public transport in india: overwhelming
needs but limited resources. Journal of Public
Transportation, 7:95113, 2004.
[13] A. Sheth. Computing for human experience:
Semantics-empowered sensors, services, and social
computing on the ubiquitous web. Internet
Computing, IEEE, 14(1):8891, 2010.
[14] A. Sheth, P. Anantharam, and C. Henson.
Physical-cyber-social computing: An early 21st
century approach. Intelligent Systems, IEEE,
28(1):7882, 2013.
[15] A. Sheth, C. Thomas, and P. Mehra. Continuous
semantics to analyze real-time data. Internet
Computing, IEEE, 14(6):8489, 2010.
[16] E. J. Topol. The creative destruction of medicine:
How the digital revolution will create better health
care. Basic Books (AZ), 2012.
[17] M. Weiser. The computer for the 21st century.
Scientific american, 265(3):94104, 1991.

You might also like