2017 the 2nd IEEE International Conference on Cloud Computing and Big Data Analysis

An Exploration on Artificial Intelligence Application: From Security, Privacy and

Ethic Perspective

Xiuquan Li Tao Zhang

Institute of Science and Technology Foresight and College of Information Science and Engineering
Evaluation Ocean University of China
Chinese Academy of Science and Technology for Qingdao, China
Development e-mail: [email protected]
Beijing, China
e-mail: [email protected]

Abstract-Artificial intelligence is believed as a disruptive the security, privacy and ethic problems in artificial
technology, which will change our economy and society intelligence application. We give out countermeasures for
significantly in the near future. It can be employed to replace these issues in section four. Finally, we draw conclusions in
human labors in completing many dangerous and tedious tasks, section five.
providing us with more convenient and efficient life. We can
benefit a lot from the wide application of this emerging
technology. However, there are also potential risks and threats II. ARTIFICIAL INTELLIGENCE AND ApPLICATIONS
in application of artificial intelligence, which need to be
Artificial intelligence is the technology that imitates
handled in a proper way before extensive usage. In the paper,
we make discussions on the security, privacy and ethnic issues
human intelligence processes by machines, especially
in artificial intelligence applications and point out the potential
computer systems. These intelligent processes include
risks and threats. Countermeasures in research, regulation and perception, learning and reasoning. Artificial intelligence has
supervision are suggested and our expectation for artificial been introduced to relieve human labors from dangerous
intelligence development is given out. tasks and situation. Currently, it is being broadly used in
different fields [2].
Keywords-artificial intelligence; security; privacy; ethics Artificial intelligence can be classified into weak AI and
strong AI. In the category of strong AI, AI system is
I. INTRODUCTION considered to have human-like high level cognition ability,
There has been 60 years development of artificial such as common sense, self-awareness and creativity, while
intelligence (AI) technology after the conference held at weak AI simulates human intelligent processes passively
Dartmouth College in 1956, at which the term of artificial without real understanding. From task resolving ability
intelligence was fust proposed [1]. Over the past 20 years, perspective, weak AI is designed to finish a particular task,
remarkable progress has been achieved in this area. In 2011, while strong AI is usually believed a general AI system,
Watson, a cognitive system designed by IBM, beat the which has the ability to fulfill multiple kinds of intelligent
Jeopardy's reigning champions Brad Rutter and Ken tasks.
Jennings, it's the fust time that computers or non-human Current AI systems are all at the stage of weak AI and
intelligence won human beings in quiz show. In 2016, as one strong AI does not yet exist so far. It is supposed that it
of Google's artificial intelligence systems, AlphaGo achieved would take decades for human to realize strong AI [3].
a big success in a match challenge with Lee Se-dol, one of Typical artificial intelligence techniques include machine
the world's top go chess players. learning, speech Recognition, natural language processing,
With further maturation of AI techniques, there will machine vision, robot, etc.. At the moment, artificial
likely be more robots or intelligent programs that can help us intelligence has been used everywhere in our life, from
as human's assistants and do a lot for us, such as reading speech text input and personalized network shopping, to
email, cleaning room, or even driving cars for us. However, various intelligent answering systems. We have benefited
every coin has its two sides, so does artificial intelligence. It and will benefit in long term from AI applications in these
will also bring us impact on privacy, security and ethic. fields such as science, engineering, education, business,
Many uncontrolled consequences might arise from AI medical care and manufacturing etc. [4].
applications if we fail to identify and prevent related threats 1) A1 application in healthcare. With the help of AI
beforehand. In this study, we discuss the potential risks and techniques, your smart home system can monitor your daily
threats of artificial intelligence, raise warnings and give our life, including your sleep time, exercise, diet preferences, and
suggestions. capture signs of their changes. Future toilet perhaps can
The paper is organized as follows: in section two, we detect through your excreta to determine if your body is
introduce the concept of artificial intelligence, related healthy, and provide relevant information to your doctor.
techniques and their applications. In section three, we discuss IBM Watson is a good example of cognitive technology

978-1-5090-4499-3/17/$31.00 ©2017 IEEE 416

application in healthcare. It can help professionals make these AI systems may do something harmful to people in
medical diagnoses, increasing patient compliance and control of criminals. One day in the future, we may be
minimizing the impact of iatrogenic disease and medical attacked just because our artificial intelligence assistants are
errors. Such system can assist doctors in making decision attacked or hijacked if this issue is not considered seriously.
without consulting specialists which are scarce resources in 2) Security problems induced by technical defects. As a
rural areas and in many developing countries [5]. developing technology, current artificial intelligence system
2) Al application in finance. Industry and government is far from perfect. Sometimes AI system may be not as
can use artificial intelligence to realize early detection of secure as it seems due to some technical defects. Thus there
abnormal financial risks and to reduce malicious behaviors are security threats in AI systems. It was reported in 2005
such as manipulating markets, fraud and unusual transactions. that a worker was killed when setting up an industrial robots
There are many AI techniques, like artificial neural network in Germany and a similar killing event happened in India in
and support vector machine, used by commercial bankers August. There was also a robot accident that happened in an
and business consultants to make bankruptcy prediction and auto parts factory in Michigan State of America in June 2015,
company financial distress prediction [6]. AI techniques are where a failed robot on assembly line attacked a worker and
also frequently employed to develop intelligent financial led to death. In that security event, 5 companies were
products, facilitating people's investment by providing high accused to be responsible for the accident.
quality fmancial assistant service. Financial automation There are some reasons that account for this kind of
transaction is another AI based financial technology, which security problem. One of them is technical imperfection as
has shown its advantages in improving efficiency and robots, tools, controllers and related components are not
reducing transaction cost. properly designed, manufactured or tested. Another reason is
3) AI application in education. AI can grade and assess improper management, causing many robotic accidents occur
students in an intelligent way and help them learn at their under unconventional operating conditions such as
own pace. Based on AI technique, we can build a new
programming, maintenance, testing, installation or
education system that includes intelligent evaluation and
adjustment. In many of these operations, when workers stay
interactive learning. With the help of intelligent education
in a robot working environment, they may encounter
assistant, teacher can be provided with additional support by
unexpected damage.
knowing the status and learning progress of each student at
3) Security problems evoked by possible self-aware
any time. In addition, AI techniques could be used to provide
intelligence. The worst AI security concern is mainly about
education push services for on-the-job people, changing
the possible that artificial intelligence develops into strong
significantly where and how people learn, or even replacing
AI stage, when robots or other artificial intelligence products
some teachers [7].
can self-evolve and may develop self-awareness, thus
III. SECURITY, PRIVACY AND ETHICS THREATS IN bringing about serious threat to human existence. Although
ARTIFICIAL INTELLIGENCE ApPLICATIONS strong AI is still far from realization, security issues should
be considered beforehand [9].
As discussed above, artificial intelligence have been With development of modern brain science, the study of
applied in different areas, and millions of people would
unsupervised learning theory is making fast progress, and
benefit from that. However, Artificial intelligence may also
cognitive intelligence may make a real breakthrough one day
be used to steal our private information and launch large
in the near future. Some research teams in the world are
scale network attacks by attackers. There are potential risks
studying high level cognition intelligence like machine
and threats in application of artificial intelligence. What's
emotion and machine awareness. Also, some scientists are
worse, it may cause ethical problems. It is important to make
exploring machine creation force and machine evolution. It
it clear the potential threats in different areas of artificial
is not known whether and when a strong AI will come, but
intelligence applications before they are put forward.
the possibility still exists. It would be a security threat if
A. Security Problems super intelligence with self-aware ability could not be fully
There are different types of security problems of AI controlled by human beings.
including security threats of technology abuse, security B. Privacy Problems
problem induced by technical defects and self-aware
In recent years, big data driven paradigm has dominated
intelligence evoked security problems.
artificial intelligence research and has led to a new bloom of
1) Security threats of technology abuse. It is believed
AI development. In current machine learning, the number
that artificial intelligence is a neutral technique. What it will
and quality of data sets will affect in high degree the training
bring us depends much on how to use and manage it. If it is
results, so most of successful AI applications rely heavily on
abused by malicious people, the technique may bring us
big data. As privacy problem is a main threat of data
problems of security, privacy and ethic. Experiment shows
exploration, inevitably, there will also be privacy problems
that attackers can launch a large-scale attack with only a little
in the applications of artificial intelligence.
resource when using intelligent methods [8]. Attackers may
1) Privacy in data acquirement. Because of wide use of
also use AI technology to access private information illegally.
smart home devices, variety of information about you and
If unmanned vehicles, robots and other artificial
your family can be kept for years or even decades. These
intelligence products are controlled through a illegal way,
data, if used properly, will make your family life better. But

417
some of your private information would also be used Some ethicists claim that artificial intelligent robots have
illegally by technology companies for commercial purposes. their own rights as we human beings, because robots might
Other data produced by electrical activities, such as have emotion, sentience, cognition and intelligence in the
pharmacy notes, mobile phone geographical coordinates and future. So if robots have self-improving intelligence and
travel routes, also involve personal sensitive information. sentience, what's the role of robots in our life?
Some of these privacy information is collected by different In some key areas such as criminal justice systems and
companies for intelligent applications, which may cause health care, many researchers have begun to explore the
privacy invasion. decision-making capabilities of AI on parole and diagnoses.
2) Privacy in cloud computing. With the introduction of But after giving decision-making power to the machine, we
cloud computing, many companies and government would be faced with not only the risk of out-of-control of AI,
organizations are migrating their data to cloud, because it is but also ethical problem itself, i.e. Does the robot have the
cheap, easy to use, and convenient in getting on-demand qualification to do that?
network access to a shared pool. When our private 3) Destroy of robots. There is a thought-provoking
information is stored at cloud, our privacy need to be ensured. question about the ethic of AI system: can we kill an
Amid higher computation requirement of artificial intelligent robot if we realize that robots are dangerous to us?
intelligence, cloud computing has been configured as main The first problem in this issue is that if we can kill robots as
infrastructure of many AI applications, so privacy problems we want. The killing of robots in the past may be an accident.
are what we need to reconsider when using such intelligence However, it may be intended someday in the future. If these
techniques. robots are not friendly or don't obey human beings, we must
3) Privacy in knowledge extracting. As knowledge switch off them or kill them. The more serious problem in
extracting tools are becoming increasingly powerful, this issue is that if artificial intelligent robots have
numerous seemingly unrelated data fragments may be intelligence and sentience, how can we kill them successfully
integrated to identify individual behavioral characteristics, against their intelligent self protection behaviors and actions?
thereby exposing personal privacy. All these problems should be considered beforehand in
For example, when internet sites visit traces, shopping design process of such AI systems.
processes and other different types of record data are
combined together, you can outline a person's behavior map, IV. COUNTERMEASURES AND DISCUSSIONS
and may analyze their personal preferences and behavioral Artificial intelligence is able to make human being more
habits, thus further predicting the potential needs of powerful, yet with security, privacy, ethic and other risks at
consumers. In this way, business shops are able to provide in the same time. We must pay attention to the problems arising
advance them with the necessary information, possible therefrom and handle the threats in advance. Therefore, in
products or services. Personalized customization has become the design, test, and use of robots and artificial intelligence
a major feature and highlight of current intelligent products, a series of countermeasures are needed to ensure
applications. But these personalized customization processes that AI systems could operate in control and in harmony with
are accompanied by discovery and exposure of personal human.
privacy, and how to avoid privacy invasion is a problem
worthy of investigation. A. Emphasizing Safety, Privacy and Ethic Research
Due to potential capabilities and complexity of AI, and
C. Ethical Problems
its close interactions with human users, research on the
As to the most special problem that rapidly changing AI safety of artificial intelligence technology is particularly
technology may bring to us, almost all scientific and important. Scholars and researchers need to emphasize more
technological personnel will believe ethic is the focus of on security protection and try their best to make artificial
attention, due to the human like intelligence ability of this intelligence more secure when applying artificial intelligence
powerful technology. Ethical problems might be induced by in more areas and fields.
following issues. 1) Embedding ethic rule in AI design. When the artificial
1) Behavior rule. Artificial intelligence robots must learn intelligence systems take actions by their own, we hope that
rules before making decisions. For example, if unmanned their behavior will be able to comply with formal and
vehicle comes in the face of pedestrians and can't brake in informal rules that we humans need to follow, including
time, artificial intelligence should choose whether crashing ethical, legal rules. These rules should be considered and
into the five people on road or turning to one pedestrian on embedded in the AI system during development stage.
sidewalk. If the design of intelligent agent is not integrated 2) Making AI system more transparent and explainable.
with the human constraints, it is likely to follow a different Current AI technology, especially deep learning, still works
logic with human behavior and lead to terrible consequences. in a paradigm of black box. For such algorithms, few
We must make these new systems benefit the whole mechanisms is available to explain the results of them. So
community, not just the controller of the system, through sometimes it's not easy to explore the cause if an unexpected
constraining behaviors of AI systems to comply with action is taken by AI system. More transparent, explainable
predefined social ethic rules. models of AI should be one of the future endeavors of AI
2) Role of robots. Another ethical problem we need to basic science research. We need to establish interpretable
consider is that what the robots or machines actually are.

418
and understandable AI systems for users, which operates in a 1) Enhancing cybersecurity. Cybersecurity has been
user-acceptable manner. recognized as one of the most complex threats to society.
3) Improving security and robustness of Al system. In Artificial intelligence is regarded as a promising technology
many cases, artificial intelligence system is designed to for cybersecurity defense. For example, Randrianasolo
operate in a complex environment. AI system should be proposed an artificial intelligence based security system
robust to deal with unexpected conditions and should be safe which has the ability to detect intrusion and learn to improve
enough to cope with a wide range of intentional attacks. performance [10]. Sahil designed a user profiling system for
Before putting an artificial intelligence system into a wide cloud environment using AI techniques to enhance cloud
range of applications, it is necessary to ensure that the computing security [11]. The further breakthroughs in
system is safe, reliable and controllable. Lots of research knowledge understanding, representation, handling, and
endeavors are thus needed in the future to make AI systems machine learning all will greatly enhance the cyber defense
more secure and robust, avoiding security accidents caused capability using latest AI methods.
by malicious attacks and abnormal operation. 2) Securing social life. There were studies showing that
AI is a powerful supporting technology in policing, fighting
B. Strengthening Regulation on Al Development
crimes, cybercrimes, and enforcing law [12]. Your home
The management of the usage of artificial intelligence is security system can be equipped not only by facial
also important for handling the security, privacy and ethic recognition software, but also by biometric authentication
problems of AI besides the technology research itself. sensors, from fingerprints, heartbeat and other features to
Research on standardization, legislation and policy should be identifY identity. AI can be applied to detect attempted credit
carried out to make sure that the application of artificial card fraud, and to conduct predictive analysis to prevent
intelligence is in control. crime. Police can also patrol using unmanned aerial vehicle
1) Law & Policy making. Because the introduction of (UAV).
artificial intelligence may bring potential threats and risks, 3) Improving privacy protection. We are in a world filled
the governments can make laws and related policies to define with private data, the promising trend of ubiquitous
what artificial intelligence can do or what is not allowed. computing, such as location based services and radio­
Government needs to improve terms of laws and lay down frequency identification, might violate people's privacy.
rules for AI industry and products usage, such as the Individual privacy protection must be addressed in order to
responsibility for the accident of unmanned vehicles, air guarantee the healthy evolution of information society. Many
UAV's invasion of personal privacy and so on. new AI techniques such as pattern recognition and machine
2) Laying down standards. Because of the still learning could play important role in improving privacy
preliminary stage of AI industrialization, there are no protection by detecting privacy invasion behaviors [13]. The
specific safety standards for many kinds of AI products, research on AI technology application in data desensitization,
especially for robots that have the ability of moving, running disclosure limitation and privacy invasion detection should
and working frequently with humans. The rules and be further encouraged in the future.
management are all urgent need for ensuring the security of
AI applications. The future standards, recommended V. CONCLUSIONS
standards and other related documents should be updated
Artificial intelligence is developing at an exciting speed.
periodically and the interval should be as short as possible,
The widespread use of AI technology can bring us efficiency
because artificial intelligence develops more rapidly than
and convenience, but we need to avoid harm to humans.
other techniques.
Although wide range AI applications and their far-reaching
3) Management and supervision. We need to control and
impact have not appeared in our lives, it is necessary to
supervise the process of development of artificial
discuss the social and ethical issues might be raised by AI in
intelligence because of its great influence. Google and many
advance.
other well-known companies have set up an ethics
Emphasizing on security, privacy and ethic issues should
committee to monitor the development and deployment of
be paid enough attention by AI researchers. The
artificial intelligence technology. We must find ways to
establishment of AI security norms and robots legal
effectively monitor and review the AI systems in their
framework are inevitable problems for the development of
applications. Many works still need to be put forward to
artificial intelligence. In spite of the warning about the
promote the development of artificial intelligence while
possible security, privacy and ethic threats, we believe that
protecting human security and information security.
AI would help to bring us security. We can utilize the
C. Improving Security and Privacy by Utilizing Al advantage of AI technology to improve the security and
Technology privacy protection for human society and cyberspace. With
the development of AI, we strongly believe that this new
AI technology can also be used to protect our privacy,
technology is more likely to bring us benefit rather than harm.
and enhance the security of systems. Not only the security
AI is an emerging technology, and the regulation of it
level of our society and cyberspace, but also the individual
can't go too far. It is not advisable to carry out severe
privacy protection could be significantly improved with help
supervision on artificial intelligence especially at current
of AI.
application stage, so as not to build obstacles to technology
innovation. On the other hand, it's very necessary to explore

419
the influence of AI to hwnan society in various aspects, and Conference on Modeling Decisions for Artificial Intelligence (MDAI
to start to build up regulation systems progressively. Based 2006), Tarragona, Spain, April 2006, pp. 274-282, doi:
10.1007/11681960 27.
on that, we can make best practice of AI application and
[7] M. Kandlhofer, G. Steinbauer, S. Hirschmugl-Gaisch and P. Huber,
greet a bright future with artificial intelligence.
"Artificial intelligence and computer science in education: From
kindergarten to university," 2016 IEEE Frontiers in Education
ACKNOWLEDGMENT Conference (FIE2016), Eire, PA, Oct. 2016, pp. 1-9.
This work was supported by research project "Research [8] P. Patil. "Artificial intelligence in cybersecurity," International
on China Artificial Intelligence 2.0 Development Journal of Research in Computer Applications and Robotics, VoI.4(S),
2016, pp. 1-5.
Strategy"(No. 2016GHOlO036) funded by the Ministry of
[9] F. Kile, "Artificial intelligence and society: a furtive transformation,"
Science and Technology of China.
Al & Society, vol. 28(1), 2013, pp.107-IIS, doi: 10.1007/s00146-
012-0396-0.
REFERENCES
[10] S. A. Randrianasolo, "Artificial intelligence in computer security:
[I] D. Crevier, "AI: The tumultuous history of the search for artificial Detection, temporary repair and defense," Ph.D. Dissertation, Texas
intelligence," Basic Books, London and New York, 1994. Tech University, May. 2012.
[2] M. Flasinski, "History of artificial intelligence," Introduction to the [II] Sahil, S. Sood, S. Mehmi and S. Dogra, "Artificial intelligence for
History of Computing, Springer-Verlag New York Inc, New York, designing user profiling system for cloud computing security:
2016. Experiment," 20lS International Conference on Advances in
[3] Executive office of the president of the United States, "The National Computer Engineering and Applications, Ghaziabad, July 2015, pp.
Artificial Intelligence Research and Development Strategic Plan," 51-S8, doi: 10.1l09/ICACEA.20IS.716464S.
Washington, October 2016. [12] S. Alzou, H. Alshibly, A. M. Aitah, "Artificial intelligence in law
[4] A. Pannu, M. T. Student, "Artificial intelligence and its application enforcement, a review," International Journal of Advanced
in different areas," International Journal of Engineering and Information Technology, Vol. 4(4), Aug. 2014, pp. 1-9, doi:
Innovative Technology (UEIT), Vol. 4(10), April 20lS, pp. 79-84. 10.SI2I1ijait.2014.4401.
[S] S. S. Sikchi, S. Sikchi, and M. Ali, "Artificial intelligence in medical [13] A. Solanas, A. Martinez-Balleste, "Advances in artificial intelligence
diagnosis," International Journal of Applied Engineering Research, for privacy protection and security", World Scientific, Singapore,
Vol. 7, Jan. 2012, pp. IS39-1543. Aug. 2009.
[6] X. F. Hui, J. Sun, "An application of support vector machine to
companies' Financial Distress Prediction," Third International

420