PERSONAL ASSISTANT WITH COMPONENTS OF ARTIFICIAL INTELLECT

Abstract

The article is concerned to the area of information science, to say more specifically – it is concerned to personal assistant with components of artificial intelligence and to the way of its application.

This project can find its application by creation and exploitation of computer, managing and production systems of different purpose where the user is in need of intelligent dialogue with a computer system and knowledge accumulation.

In comparison with other ways of communication with computer systems complexity of making retrieval requests reduces. As far as questions are formulated on natural language and adequate choice of search results takes place, teaching and knowledge accumulation, taking experience and user’s references into account are possible. Active dialogue of system and the user is carried on, including messages on system’s initiative. The article was written as part of a grant from the Russian Foundation for basic research № 14-07-00783.

1. General introduction to the problem

All known historical person and humanity as a whole is trying to create a helper to facilitate labour – first physical, and then intellectual.

At the same time success in facilitating labour processes primarily related to increases in power and acceleration of production, and the success of collective intelligence – with the growth of information capacity.

However, the growth of information capacity does not mean growth and ability to comprehend and process information. We see this in the example of the modern Internet – enormous amounts of information do not allow to navigate it in a reasonable time, to make sense of it and correct use, we also often can't ensure the authenticity of information.

Existing search engines are only partially solve this problem, offering us to act on the principle of Robert Sheckley [1]: To properly ask a question, you have to know more than half the answer!

We begin to stagnate and, figuratively speaking, everyday to "reinvent the wheel" - to repeat already done the research and work, unable to find them in a huge array of diverse information.

It is quite obvious that the idea is to simulate the human brain has exhausted itself, so unproductive to attempt to build a modern means of transport based on the modeling work of the legs, when the technique confidently uses the wheel.

Remembering that "in the Beginning was the Word", it is wise to pay attention to semantic algorithms, simulating the processing texts consisting of words, so similar to the main structural processes of thinking and mental effort.

While it is unwise to rely on probabilistic algorithms, e.g., neural networks, as a result of their work is often unclear and unpredictable.

We set and solve the problem of creating artificial intelligence first step – artificial intelligence assistant.

And the basic principles of its work – the clarity, predictability, learning, personality and reliance on existing information resources.

Adding to any modern search engine and thus using already accumulated knowledge, deterministic methods of analysis of the text we will achieve full clarity and predictability of the decisions of our intelligence assistant by entering individual preferences of the owner – will have the personality and trainability, typing evaluate the usefulness of the work of the assistant, we will close the feedback learning and individualization of our "second I" that can help us.

2. Introduction and Discussion

This development can be used by creation and exploitation of computer, operating and producing systems of different purpose where a user requires intelligent dialogue with the computer system and accumulation of knowledge.

Current level of acknowledgement of this problem is realized in every searching system.

But, up-to-date searching systems have following fundamental imperfections:

·   Complexity of retrieval requests compilation and search data choice

·   Impossibility of teaching and accumulation of knowledge, accounting of experience and the user’s preferences

·   Absence of dialogue of the system and the user. Modern intelligence projects-helpers

Google Now [4], the search engine without requests - the system analyzes the user action and its location and generates an "information card" - advice, recommendations and assistance to the user

Siri – voice organizer and maintaining dialogues with the user – reminds about business, gives advice and answers questions [3].

The shortcomings of these projects are the following:

•   "Black box" - the user does not know how the system works, he is looking for and what information about it takes into account,

• The need for continuous communication channel with the Internet,

• The secrecy and pressuremost project

• The lack of feedback and assessments utility Basic solutions (what we need and what we offer)

• "White box" the user knows or has a clue how the system works and what information about it takes into account,

• Learning and feedback to the user,

• Local or corporate use,

• Openness and complementarity of the project

In personal assistant with components of artificial intelligence and method of its application tasks of project are solved and fundamental imperfections of prototype in up-to-date searching systems are removed.

3. The main ideas of the project

Personal assistant with components of artificial intelligence is intended for dialoguing with the user by means of questions and answers exchange, representing by itself texts on native language and forming pieces of advice for the user in a form of texts on native language. It consists of operation executing components implemented on unconditioned computational resources. Moreover, operation executing components include at least the following: a component for text recognition, inputted by the user; a component for searching in data array; a component for answer forming for the user; a component for feedback, estimating the answers formed for the user; a component for settings which manages forming of answers; a component for connection with other personal assistants; a component for connection with searching systems, and data array, including, but not limiting the following: data array of the user’s questions; data array of answers to the user; data array of settings, containing information about user’s preferences and mode of operation for personal assistant; data array of events, describing the information about past or coming events.

The method of application of personal assistant with components of artificial intelligence, described above, consists in the following:

a)    The user forms questions which enter the text recognition component inputted by the user, which in its turn transmits them into the component for searching in data array. Whereupon, the component for searching in data array compares them with data array of answers to the user and if identical or resembling answer is available it is directed to the component for answer forming for the user.

b)    In that case, if the question of the user forms an event, information about the event is data array, describing information about past and coming events.

c)   In that case, if identical or resembling answer in data array of answers is not found, the text recognition component applies to the component for connection with other personal assistants, which transmits a question to another personal assistant and waits for an answer from it, another personal assistant functions by analogy.

d)   In that case, if answer from another assistant is not received or appeal to other assistants is impossible or prohibited by the component for settings, the component for text recognition inputted by the user distinguishes content of a question in the form of text. Whereupon, this text is transmitted to the component for connection with searching systems, answer from which come in the component for answer forming for a user, where the answer is compared with data array of settings and with account taken of this comparison is transferred to the user.

e)    Not obligatory that the user estimates an answer when receives it. The answer of the user is transferred to the component for feedback and the answer is placed into data array of answers with the user’s estimation or into data array.

f)   The component of answer forming for the user independently from user’s questions forms recommendations for the user in that case, when from other personal assistants a question (the question itself or answer on which is indicated in data array of settings as interesting for the user) comes in and also in that case, when questions asked before require an answer in present time, or the event has come, which there is in data array.

Thus, complexity of making retrieval requests reduces. Teaching and knowledge accumulation, user’s experience and preferences accounting is possible as far as questions are formulated on native language and adequate choice of searching results takes place. Active dialogue of the system and the user carry on, including messages on the system’s initiative.

Here given some technological explanations of usage and schematic description, with references on Figure1. It depicts block diagram of personal assistant with components of artificial intellect – personal ARtiFicial Intellect Assistant (ARFIA) in diagram form.

Text comparison is the basic function used by personal intellect assistant. It consists in that texts are compared between each other in set-theoretic sense. General part (intersection) is detected: words which concern only to the first or only to the second text. Comparison of texts allows to single out the sense of questions asked by the user and to realize users’ preferences by means of comparison of preferences with found answers.

Measure of texts similarity is computed in following way – similarity S S = 2P/(t1+t2) where

P – the size of texts intersection, t1 – number of words in the first text, t2 – number of words in the second text. For exampl, the user asked ARFIA the question:

Where is pharmacy №51 situated?

Then ARFIA compares text of the question with text-patterns of questions to single out the sense: Where is situated

Where to buy Who is

What is name of When will be

For the first question (two words are coincide) S=2x2/4+2)=2/3, for the second S=1x2/4+2)=1/3, for the rest coincidence is equal to 0.

Thus, ARFIA identifies that the question on location (the first pattern) was put and single out what exactly is to be found (pharmacy №51) – words containing only in the first text. Suppose there is no result in data array of answers to the user and search in the searching system was executed.

Search results:

1.     Pharmacy №51 of drug chain A5

Address: Moscow, Dmitrovskoe highway, 45/2 Tel.: 8-495-916-86-47

Time of work: 9.00-21.00

Metro station: Petrovsko-Razumovskaya

2.     Address

125430, Moscow, St. Mitinskaya, 40 Business Center “Mitino”, 1 floor Tel.: 8-495-759-37-29

Time of work: 8.00-23.00

3.     Address

Lipezk, St. Belana, building 26, Business center “Real”

Tel.: +7 (4742) 45-1668, +7 (4742) 45-3607 Time of work: Every day 9.00-21.30

If the user has Moscow as the city of location in preferences, then comparison picks out only first two search results and transmits them to the user.

Further there is illustration of ARFIA work with data array of events describing information about past and future events.

Suppose that today is the 7th of March ARFIA prepares the notice to the user:

Tomorrow is the 8th of March. It’s time to congratulate women! Let the user asked qualifying question:

What did I give as a present last year?

ARFIA will actualize searching in data array of events of one year remoteness and will find the following data: Last year presents:

Women co-workers received boxes of sweets “Vechernyi Zvon”; Mother received money;

Girlfriend received a bracelet; For the next question of the user

Where is it better to buy presents on the 8th of March?

According to analogous scheme, the retrieval request (presents on the 8th of March) will be marked out and the search will be carried out:

Its results, for example, will be the name of site: http://www.dostavka-podarkov.ru/holidays/03_08/ Presents for women on the 8th of March.

This year our internet-shop has prepared for women new presents and souvenirs for the 8th of March. We hope that among varied stock of presents for the 8th of March you could find and order an appropriate present. Attention! Make orders of the 8th of March presents in advance, not in the last day.

Also, we would like to remind that we are ready to work with wholesale and corporate customers of the 8th of March presents and souvenirs. Delivery of retail and wholesale customs realizes on the whole territory of Russia.

In our internet-shop wide range of gifts for women is always represented. Afterwards, ARFIA will ask a question in some time intended for its self-learning: What did you choose as presents this time?

It saves data entry in memory to give it to the user in case of need.

4. Аlgorithm description

Thus, personal assistant with components of artificial intelligence intended for dialoging with a user (1) by means of questions and answers exchange which represents with themselves texts on natural language and forms pieces of advance for the user on natural language consists of executing operations components realized on unconditioned computational resources. At that, executing operations components include at least the following: a component for text recognition, inputted by the user (2), a component for searching in data array (3), a component for answer forming for the user (11), a component for feedback, estimating the answers formed for the user (12), a component for settings which manages forming of answers (9), a component for connection with other personal assistants (8), a component for connection with searching systems (10), and data array, including, but not limiting the following: data array of the user’s questions (4), data array of answers to the user (5), data array of settings (6), containing information about user’s preferences and mode of operation for personal assistant, data array of events, describing the information about past or coming events (7).

·       The user forms questions which get into the component for text recognition inputted by the user. This component in its turn transmits them into the component for searching in data array, following which the component for searching in data array compares them with data array of answers to the user and in the presence of identical or resembling answer it is directed to the component for answers forming for the user.

·       In that case if the user’s question forms an event, information about the event is data array of events, describing the information about past or coming events.

·       In that case, if identical or resembling answer in data array of answers is not found, the text recognition component applies to the component for connection with other personal assistants, which transmits a question to another personal assistant and waits for an answer from it, another personal assistant functions by analogy.

·       In that case, if answer from another assistant is not received or appeal to other assistants is impossible or prohibited by the component for settings, the component for text recognition inputted by the user distinguishes content of a question in the form of text. Whereupon, this text is transmitted to the component for connection with searching systems, answer from which come in the component for answer forming for a user, where the answer is compared with data array of settings and with account taken of this comparison is transferred to the user.

·       Not obligatory that the user estimates an answer when receives it. The answer of the user is transferred to the component for feedback and the answer is placed into data array of answers with the user’s estimation or into data array.

·       The component of answer forming for the user independently from user’s questions forms recommendations for the user in that case, when from other personal assistants a question (the question itself or answer on which is indicated in data array of settings as interesting for the user) comes in and also in that case, when questions asked before require an answer in present time, or the event has come, which there is in data array.

Additional examples can illustrate the work of ARFIA. In 1950 Alan Matison Turing published an article under the title “Can the machine think?”[2]. Turing considered two fundamental definitions in the article as “machine” and “think”. But as long as scientist’s thoughts on the first problem were mainly actual for his contemporaries, developed by Turing criterion of sapience of computing systems is used nowadays.

For detection of electronic intellect he proposed so called “game of simulation”. Initially three people participate in it: a man, a woman and a judge. The last is separated from two others by room’s walls, everyone communicate by means of telegraph connection (to except cognition by the voice, in modern case it will be communication by means of inputting texts on keyboard). The aim of a person who asks questions is to find out of the rest participants who the woman is and who the man is. The aim of game for the woman is to help to the judge and for the man to put him off his stride. The next step was to make the machine answer instead of the man. Will the person who asks questions make mistakes as often as it was in game where the only participants are people? Alan Turing predicted that in the course of five minute test a computer with memory about 120 megabyte could deceive a human in 30% of cases. Current semantic technologies allow lifting Turing’s plank up to 70 or more percent cases.

Artificial Intellect Assistant (ARFIA) is technically realized as an application which is teachable pseudo-reasonable system connected in the internet-space with other intellect assistants. ARFIA has scalable modular architecture that includes the nucleus and additional modules which are connected to ARFIA by standard interface and process data in standard or document format.

Besides that, indexing of created by the user documents, messages and dialogues in social networks is carried out for their fast-access retrieval, correction of preferences and renewal of everyday dictionary. Description or image annotation can also be offered to the user (overlooked, created and selected by him). At that images or references on them are placed into the collection of local results.

1. Conclusion

Taking into account realization of knowledge accumulation and teaching during 2-3 months ARFIA turns into inalienable alter ego of the user and it allows to minimize technical problems of search and selection of information, and also successfully automatically introduces the user in groups of communication and social networks by means of module of text generation. Besides, the local collection of the most needed informational resources, independent of long-haul networks accessibility, is automatically organized to the user.

In comparison with other ways of communication with computer systems complexity of making retrieval requests reduces. As far as questions are formulated on natural language and adequate choice of search results takes place, teaching and knowledge accumulation, taking experience and user’s references into account are possible. Active dialogue of system and the user is carried on, including messages on system’s initiative.

List of references

1.     Robert Sheckley Ask A Foolish Question, http://www.readbookonline.net/readOnLine/63673/

2.     The Alan Turing Internet Scrapbook The Turing Test, 1950 http://www.turing.org.uk/scrapbook/test.html

3.     Speech Interpretation and Recognition Interface https://ru.wikipedia.org/wiki/Siri

4.     Google Now https://ru.wikipedia.org/wiki/Google_Now