International Journal of u- and e- Service, Science and Technology

Vol.8, No. 7 (2015), pp.277-284

http://dx.doi.org/10.14257/ijunesst.2015.8.7.28

A Study of Today’s A.I. through Chatbots and Rediscovery of

Machine Intelligence
Anirudh Khanna1, Bishwajeet Pandey1, Kushagra Vashishta1, Kartik Kalia1, Bhale
Pradeepkumar and Teerath Das
1
Chitkara University, Punjab, India
[email protected], [email protected],
[email protected], [email protected]
2
IIITM, Gwalior, India
[email protected]
3
Gran Sasoo Science Institute, Italy
[email protected]

Abstract
Artificial Intelligence in machines is a very challenging discussion. It involves the
creation of machines which can simulate intelligence. This paper discusses some of the
current trends and practices in AI and subsequently offers alternative theory for
improvement in some of today’s prominent and widely accepted postulates. For this, focus
on the structuring and functioning of a simple A.I. system - chatbots (or chatter bots) is
made. The paper shows how current approach towards A.I. is not adequate and offers a
new theory that discusses machine intelligence, throwing light to the future of intelligent
systems.

Keywords: artificial intelligence, machine intelligence, chatbots, theory of intelligent

machines, Turing test

1. Introduction
The creation and analysis of intelligent agents (software and machines) is called
Artificial Intelligence, or AI. It can be implemented in nearly each and every sphere of
work. Intelligent machines can do many tasks – from labor work to sophisticated
operations. Prominent trends in this field are human brain simulation, natural-language
processing and neural networking etc. One of the typical examples of an AI system is a
“chatbot”. A chatbot is a computer program which responds like an intelligent entity
when conversed with. The conversation may be through text or voice. Any chatbot
program understands one or more human languages by Natural Language Processing. Due
to this, the system interprets human language input using information fed to it. A chatbot
may also perform some productive functions like calculations, setting-up remainders or
alarms etc. A popular example of chatbot is the ALICE Bot (Artificial Linguistic Internet
Computer Entity), which utilizes AIML pattern matching techniques [1]. “Turing test” is
one of the most popular measures of intelligence of such systems. According to it, if a
panel of human beings conversing with an unknown entity (via keyboard, for example)
believes that that entity is human while the entity is actually a computer, then the
computer is said to have passed the Turing test. This test was proposed by British
mathematician Alan Turing in his 1950 paper titled “Computing Machinery and
Intelligence” published in Mind [2]. All these developments have been made, but are we
actually constantly progressing towards the higher goals of AI? These days, the evolution
in AI has been limited around some key points like the fed-in information database,
imitation of human abilities, targeting the Turing test and techniques of deception. Lesser
novel visions are being introduced in this field and further research is going on in existing

ISSN: 2005-4246 IJUNESST

Copyright ⓒ 2015 SERSC
International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

theories only. Also, today’s AI systems “pretend to act like” intelligent entities instead of
being one. For example, today’s rule-based systems just follow a large set of if-else based
logic (the database) while actually intelligent entities don’t always do so, but apply
reasoning in every stage. Thus, it is not the case that Artificial Intelligence is failing, but it
has slowed down in attaining more laurels. There is a need to view AI systems from a
new perspective and a novel theory of machine intelligence is the required tool for this.

2. Functioning of Simple AI Systems

Chatbots are one of the most elementary and popular examples of intelligent Human
Computer Interaction (HCI). For realizing the strengths and weaknesses of current AI
systems, we created two chatbot programs – FUTURE (in C++) and SARANG (in
AIML). Moreover, the SARANG Bot is hosted on the internet and can be accessed from
worldwide [3]. C++ is a typical general-purpose object-oriented programming language
for creating applications. AIML (Artificial Intelligence Markup Language) is a dialect of
XML which is designed especially for natural language processing agents, like the
chatbot programs. Choosing AIML gave us the advantage of using the open-source
ALICE AIML set directly, consisting of the 50,000 responses of ALICE chatbot in our
project and made it more efficient for research [4]. This section gives a review of their
structuring and functioning. AI designing generally requires the system to be
differentiated into simple sections of work. In chatbots, there is a separate section that
manages the transfer and understanding of messages between human user and the AI
system, another section that stores the useful key-points during the chat that may be
required later-on, another department that handles the errors and so on. The mutual
interactions of these modules are shown pictorially in Figure 1. The program is logically
divided into the following modules.

2.1. Introductory Screen

The welcome screen introduces the user to the program and may ask him some of his
details like the name for further conversation purposes. A key point like the user name is
good to be stored in a data-file with use of Data Handling.

2.2. The Chat Interface and Manager

This is the main region, where all the text-inputs are managed and the program fetches
the best-matching answers from its database. It has to search and display from tens of
thousands of rules stored in the database.

2.3. The Database

This is the brain of a chatbot. Database consists of an organized, manually created list
of suitable replies given for possible questions that may come up from user’s end. A vast
range of questions has to be covered by an efficient chatbot for standing firm in the
industry and it should also have more than one answer to the same kind of question to
avoid redundancy of replies. Each reply should be semantically connected.

2.4. Productivity Application

The chatbot becomes much better when some productivity applications like a simple
calculator or a dictionary or even a game is embedded in it. We introduced a calculator in
the FUTURE bot. If the user types in for a calculator requirement, there is a simple
calculator in the program itself. The user can work upon basic arithmetic operations,
trigonometric functions and differentiate simple expressions too. These days, AI agents
have much more to offer than a calculator. They can make memos, notes, set alarms and
open up programs on your computers or mobiles on demand.

278 Copyright ⓒ 2015 SERSC

 International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

2.5. Ambiguity Handling

Sometimes, the end-user poses questions that a program’s database does not have
answers to. Here, ambiguity handling comes into play. It gives answers when no relevant
data is available. Its contents fit into wide range of contexts and hence, user is satisfied
even for an unexpected question.

2.6. Data Handling

The user may not like to tell his name each time he uses the chatbot. Important data
about user can be stored permanently via data file handling. We store data for future
reference, and for modifying our answers such that they seem suitable for the user
specifically. This makes the program more intelligent.

2.7. Error Handling

One example of this is seen if the user tries to divide any number by zero while using
the embedded calculator application in the chatbot. The program should be capable of
handling such general errors and try to continue operating within its environment even
after that.

Figure 1. Interaction between the Various Modules

We infer that using AIML with designated application in AI field boosted the total
speed of development of the project. It significantly increased the efficiency of the
program created as it offers a lot of AI-oriented features. With well over 60,000 responses
fed into it, the SARANG Bot was tested for about 1500 queries. It could manage to give
suitable replies to about 1200 of them, which gives its accuracy to be 80% that has been
shown in Figure 2.

Copyright ⓒ 2015 SERSC 279

International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

Figure 2. Performance of SARANG Bot

This is a little glimpse of how efficient and diverse AI systems can be constructed
today, and that too, with not much difficulty. There are a lot of great chatbots in the
world. Many have won the Loebner prize and are really very good programs. But even
after all these advanced, aren’t we just standing there only? The core problems with
chatbot programs (and AI systems in general) remain the same. They do not hold the
power to learn, hold memories of people and incidents. They do not have (or simulate)
senses. They are not creative and original, just a fixed set of instructions. They do not
actually understand what is going on (the conversation, for instance). They give responses
by searching and finding matches from a whole lot of database inputs, that sometimes
goes up to millions of patterns. [5]

3. The Turing Test and its Problems

The Turing test is not a perfect measure to tell if a machine is intelligent. There is
perhaps no fool-proof way to tell if an entity is intelligent or not. Sometimes,
unintelligent behavior is witnessed in intelligent beings. On the other hand, dumb
machines may at times perform a lot of tasks more intelligently and accurately than
intelligent human beings. But the Turing test has its own challenges discussed as
follows.

3.1. Limitedness
The test only features intelligence in terms of conversational skills of a machine.
Intelligence is a lot more than that.

3.2. Short Purview

The scope of this test is only bounded within the limits of human ability to comprehend
and respond in a human language. But it is not the only intelligent behavior seen in the
world. What about intelligence in animals? Also, ability to pass the Turing test is not the
only sign of intelligence shown by humans.

3.3. Unproductive Developments

Trying to pass the Turing Test is merely a distraction from more fruitful research.

3.4. Disillusionment of Goals

Creating lifelike simulations of human beings is a critical task in itself. Such entities
may be interesting in a game or maybe as a friendlier user interface, but they are not part
of the core science of creating intelligent machines, that is, machines that solve problems

280 Copyright ⓒ 2015 SERSC

 International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

using intelligence. The test is not compulsory to be passed for achieving higher goals of
AI research.

4. Back to the Basics: The Meaning of Intelligence

If the present postulates need to be improved, we have to provide some alternatives.
For this, we restart from the basics. There is need to rethink about the fundamental
abilities upon which intelligence around us works. These are as follows.

4.1. Arithmetic
The power to compute is the fundamental of intelligence. It includes arithmetic
operations like addition, subtraction, division and so on. Today’s machines perform well
on this part. They can carry out even complex calculations in no time.

4.2. Comparison, Logic and Reasoning

The scope of machine intelligence becomes wider when a system acquires the ability to
apply logic and make comparisons. Today’s computers can perform logical operations
well with the principles of Boolean algebra.

4.3. Learning, Heuristics and Memory

Power to recollect past incidents, learn new things and gaining experience constantly
are major features of intelligent entities. R&D is going on in this direction these days.
Heuristics is being implemented in newer software and as time passes, the ability of
machines to learn and gain experience will grow.

4.4. Senses
This is how we know about the environment around us. We humans are fortunate
enough to have really efficient and effective set of senses. Some animals are said to have
even greater abilities to sense. A machine with rightly installed equipments to sense the
surroundings will prove to be a great body for its intelligent brain. But as of now, even a
caterpillar can outperform machines in the ability to know about their surrounding
environment.

4.5. Perception
This is how we receive and process the signals our senses provide us. This is how
creativity is born to intelligence. To have a machine with a perception seems a distant
dream of AI.

4.6. Consciousness
Consciousness of intelligent beings is perhaps the most difficult property to define. It is
even more difficult to be implemented in a machine. How can an abiotic system come to
notice the presence of itself in the world? This is a question which is really difficult to
answer. Everyone would have their own opinions. It’s evident that today’s systems have
no consciousness as such.
Figure 3 shows the quantitative and qualitative nature of the above listed fundamentals
of intelligence and also a rough extent up to which they are being implemented in today’s
intelligent machines.

Copyright ⓒ 2015 SERSC 281

International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

Figure 3. Analysis of Fundamental Traits of Intelligence and Today’s AI

5. Redefining Machine Intelligence

This paper intends to offer a novel overview regarding Artificial Intelligence entities
and systems. The statements given below in regard to a newly offered theory of machine
intelligence provide the vision.

5.1. Statement 1: Partially Intelligent Systems

Any machine, system or program exhibiting some but not all of the above listed
fundamentals of intelligence is a Partially Intelligent System. For example, chatbots
exhibit some of above listed traits, namely Comparison, Logic & Reasoning and
Learning, Heuristics & Memory. If there is an additional module, they may exhibit some
more traits also, like ability to perform arithmetic operations. Therefore, chatbots are
Partially Intelligent Systems. Moreover, today’s machine intelligence is greatly confined
to the boundaries of Partial Intelligence. Even highly capable robots of today are an
instance of highly capable Partial Intelligence, as today’s AI does not possess complete
power of all the fundamentals of intelligence.

5.2. Statement 2: Completely Intelligent Systems

Any machine, system or program exhibiting all of the mentioned fundamentals of
intelligence completely will be considered a Completely Intelligent System. Such an
entity will represent the true power of AI.

5.3. Statement 3: Performance Factor

The Performance Factor of a system is a measure offered for representing the
capability of an intelligent system in terms of the fundamental traits of intelligence it
possesses. It is given by the number of apparent fundamental traits shown by the given
system divided by the number fundamental traits of intelligence, i.e., 6. Its value ranges
from 0 to 1. Hence, today’s chatbots showing two or three of the given traits get a

282 Copyright ⓒ 2015 SERSC

 International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

Performance Factor score of 0.33 to 0.5. A “Completely Intelligent System” will have a
Performance Factor of 1.

6. Conclusion
Artificial Intelligence is perhaps the most interesting as well as challenging field
of research today. It has already proven itself for solving some major problems for
mankind. In near future, AI will present itself on a larger canvas and will become
integrated in our day-to-day lifestyle. But there is need to continuously look for new
ideas for development and to make progress in already devised research. We will
then look forward to solving such problems via intelligent systems, where human
intelligence can err. The paper offered an alternate foundation theory of intelligence
in machines. The basic work can be carried forward to develop advanced AI theories
and systems. Moreover, there is utter requirement for new theories to emerge and
develop, especially in a challenging field like Artificial Intelligence.

Acknowledgements
We pay sincere acknowledgment to all the authors of all the reference material. Also,
the kindest blessings of our parents and all the teachers gave the zeal and faith to put
forward this work. Thanks to them.

References
[1] B. A. Shawar and E. Atwell, “Chatbots: are they really useful?”, LDV Forum, vol. 22, no. 1,
(2007).
[2] A. M. Turing, “Computing Machinery and Intelligence”, Mind, (1950), pp. 433-460.
[3] A. Khanna, “Pandorabots Chatbot Hosting Platform. SARANG Bot”, (2015) April 19, Internet:
http:// pandorabots.com/pandora/talk?botid=9f0f09a71e34dcf8/.
[4] A. I. Alice, “Foundation. Free A.L.I.C.E. AIML Set”, (2015) March 21, Internet:
http://code.google.com/p/aiml-en-us-foundation-alice/.
[5] B. Whitby, Author, “Artificial Intelligence”, The Rosen Publishing Group, (2009).

Authors
Anirudh Khanna, is a first year student of B.E. in Computer
Science Engineering, at Chitkara University, Punjab. His interest
areas are Artificial Intelligence, Programming, Web Development,
Green Computing and etc. He is into the analysis and testing of
algorithms, tools and techniques in Artificial Intelligence and
development of novel theories for improvement.

Bishwajeet Pandey, is asst. professor in Centre of Excellence of

Chitkara University. He has worked as JRF at South Asian University
and visiting lecturer in IGNOU. His research areas are Green
Computing, Low Power VLSI, Energy Efficient Design, AI, Green
Computing, etc.

Copyright ⓒ 2015 SERSC 283

International Journal of u- and e- Service, Science and Technology
Vol.8, No. 7 (2015)

Bhale Pradeepkumar, has been working as asst. professor in CSE

Department of Jawaharlal Nehru Engineering College, Aurangabad
and has recently received Masters Degree (CSE) from IIITM,
Gwalior. He has specialization in Information Security, A.I. and Data
Mining.

Kartik Kalia, is working as a researcher in Chitkara University

Research and Innovation Network (CURIN). He is also appointed as
a Director in Gyancity Research Lab. His area of interest is Low
Power Research in Thermal Design for Energy Efficient Computing.

Teerath Das, is a doctoral student in Gran Sasoo Science Institute,

Italy and has completed M.Sc. in Computer Science from South
Asian University, New Delhi. His area of interest includes Low
Power Techniques in VLSI and Digital Signal Processing.

284 Copyright ⓒ 2015 SERSC