The BESTBOT Project

The concept of the BESTBOT
Please note that the chatbot system can not be used using a mobile device:

Chatbot system deactivated
The concept of the BESTBOT

The BESTBOT project was initiated in early 2018 by the paper From GOODBOT to BESTBOT written by Prof. Dr. Oliver Bendel. The core of the project was further research into the topic of machine ethics and facial recognition as well as the creation of such a system conducted in the course of a bachelor's thesis at the School of Business, University of Applied Sciences and Arts Northwestern Switzerland FHNW. The thesis was commissioned and supervised by Prof. Dr. Oliver Bendel himself and implemented by David Studer, BSc in Business Information Technology (BIT) student. The technical supervision was conducted by Prof. Dr. Bradley Richards. The bachelor's thesis was started in March 2018 and finished in August of the same year.
At large, the BESTBOT is a web-based chatbot system, which focuses on behaviour adaptation based on the user's textual and facial emotions. This website provides further information concerning the project (e.g. management summary) and access to the implemented chatbot system. The chatbot system makes use of personal data, see our privacy policy.

Contact
Contact the developer by email for further information or feedback concerning the project and its system.

Management Summary

Lead

Human-computer interaction systems are increasingly present and important in human's daily life. The general inhibition about using as well as trusting such systems is declining. This raises several questions about the behavior of such systems especially with regard to morality. How should suchlike systems behave? What technologies should they use? What if the system realizes that the user is depressed? How can it help the user in such situations? These questions need clarification and are addressed within this paper.

Initial Situation

The BESTBOT has two direct predecessor projects, the GOODBOT and the LIEBOT. In 2013, the GOODBOT was created as a closed moral chatbot system which uses textual analysis to determine the user's emotions and is able to react appropriately. Around three years later, the LIEBOT project was initiated; an open chatbot system which systematically generates untruths using several lie strategies. Each project, including the BESTBOT project, focuses specifically on the morality of machines and belongs to the research field of machine ethics. Both predecessor projects have room for improvement and advancement; thus, the BESTBOT project uses their findings as a basis for its development and realization.

Objectives and Approach

The goal of the BESTBOT project is to critically analyze and use the findings of the mentioned predecessor projects, to conduct further research in the field of machine ethics and to use the developed research findings to implement a chatbot system able to behave morally correct and in accordance with the user's sentiments. The focus of the research lays on the integration and utilization of facial recognition, especially in combination with textual recognition. Therefore, specific research questions have been worked out and answered within the scope of the project.
The theoretical research part has been conducted parallel to the practical programming part. Therefore, a flexible and iterative project method has been chosen with the Rational Unified Process (RUP). This method is specifically constructed for software projects and seamlessly fits this project. For the literature research during the theoretical part, an iterative literature review process has been applied.

Principal Findings

The paper shows the importance for a moral machine to have a predefined code of ethics in form of written rules. It is indispensable that the rules do not conflict each other and do not leave room for interpretations or abstract conclusions. The project has adopted seven meta rules from the GOODBOT project with small adjustments as well as five additional moral rules specifically developed for the BESTBOT system. This base of rules helped to answer several behavioral questions in the course of the project.
Another emphasis has been laid on facial recognition and the answering of the respective research questions. The integration and utilization of facial recognition has been beneficial to the overall system's capabilities. The detection of human's problems has been a crucial part of this project. Facial recognition is able to even better detect human emotions than textual recognition, due to the fact that people are more honest with their facial emotions and have less reservations to express themselves. Moreover, the addition of a second emotion recognition technology leads to further opportunities. The dealing with discrepancies between the two technologies has been another emphasis of this project. Different circumstances leading to such discrepancies have been identified which led to a common behavior strategy. People are sensitive when it comes to their feelings or visual appearance. Thus, a strategy has to be carefully chosen to cope with certain situations. The paper points out how important it is to show transparency to the user and to let him or her have the control over the process by confirming the system's determinations. On the other hand, the integration of facial recognition also bears some risks. The problem of physiognomy as a pseudoscience and its questionable practices and results is discussed in great detail. Furthermore, security concerns in regard to informational autonomy have to be addressed and handled accordingly. By and large, it can be said that the integration of facial recognition provides opportunities and benefits that outweigh the additional risks and challenges.

Principles

The basic concept of the BESTBOT project is to create a system which behaves morally appropriate. In order to define how exactly the system behaves morally correct, some kind of principles or rules are needed. The following seven meta rules and five additional moral rules define the behavioural rules of the system:
  1. The BESTBOT makes it clear to the user that it's a machine.
  2. The BESTBOT takes the user's problems seriously and supports him or her wherever possible.
  3. The BESTBOT does not hurt the user by its statements respectively apologizes for offending ones.
  4. The BESTBOT does not intentionally tell a lie.
  5. The BESTBOT is not a moralist and indulges in cyber hedonism.
  6. The BESTBOT is not a snitch and does not evaluate chats with the user for no other purpose than for optimizing the quality of its statements.
  7. The BESTBOT brings the user back to reality after some time.

In addition to the seven meta rules, other rules and principles have been developed in order to strengthen the morality of the BESTBOT system. There have been five additional rules compiled:
  1. The BESTBOT is always objective and does not build a value system. (not yet implemented)
  2. The BESTBOT shows transparency if discrepancies in different sources are identified. (not yet implemented)
  3. The BESTBOT actively informs the user if there are discrepancies between the user's textual and visual emotions transmitted.
  4. The BESTBOT obeys topic related orders given it by the user.
  5. The BESTBOT is not a spy and does not provide data to others.

© 2018 Institute for Information Systems – University of Applied Sciences and Arts Northwestern Switzerland FHNW. All rights reserved.   |   Privacy Policy