Ethics Aware AI Software

• This folder contains all code, collected data, and presenttation work from my Bachelor's Essay (499A) on Ethics Aware AI Software. It highlights models created using NUSMV and models examined under ethical dilemmas that we created.

Background

• Emerging artificial intelligence systems are making more decisions in our society than ever before – and it is important to examine the impact that these decisions have on humans' daily lives. As technology advances, the intent of such systems is to improve human lives, not worsen them.

Objectives

• Demonstrate the ability to read and understand technical papers on current AI systems.
• Familiarize the components of ethics aware artificial intelligence software.
• Construct and evaluate a theoretical framework for ethics aware artificial intelligence software.

Research Methods

• Ethics can be defined as a moral decision making in order to protect living things. Why is ethics aware AI software critical? As technology continues to grow in our society, it is crucial that we consider how this technology affects our decision making and whether it is ethical or not. Technology must be developed to respect our morality and make decisions without bias or discrimination. What are the pitfalls if ethics is not integrated in the software decision making process? Without having ethics integrated in the software decision making process, this could lead�to unethical principles and potentially immoral behavior from the user's perspective. When examining our research methods used throughout this study, we must first know the definition of model-checking – which is making sure that a specific software meets the given requirements or properties. It involves exploring different states of a given system and verifying whether such states satisfy formal specifications. Two different types of model checking include Computation Tree Logic (CTL) and Linear Temporal Logic (LTL). CTL model-checking creates a tree of computation paths whereas LTL reasons about temporal properties over a linear sequence of states. Although both CTL and LTL are used in model-checking to verify desired properties, this research will mainly be utilizing CTL to examine the consistency of rules, beliefs, and actions within existing models and models created.

Results and Future Work

• In order to ensure that software is developed with ethical considerations in mind, we must create transparent algorithms – meaning, ways in which AI makes decisions should be clear and available to the public. This will allow others to hold such systems accountable and prevent biased or discriminatory decision-making within systems/models. Secondly, incorporating ethical considerations within the development process of such systems would allow stakeholders to visualize the potential impact of the AI on society – and to avoid any negative effects. Finally, ethical training of AI software that recognizes ethical dilemmas and makes ethical decisions would be beneficial, along with regular assessment and monitoring of the software for potential problems. Overall, this research has increased awareness of such issues, and aims to continue creating models to impose dilemmas on for research.

Contact Information

• Channing Smith, smithcs@g.cofc.edu, College of Charleston
• Dr. Kris Ghosh, ghoshk@cofc.edu, College of Charleston