Ethical AI and Business Adoption
Artificial Intelligence is expected to bring significant and diverse benefits to society ranging from greater efficiency and productivity tackling a number of difficult global problems. The issue varies from climate change, poverty, disease, on one side to improvise business standpoint on the other hand by incorporating technology in an innovative way of automating processes and executing a data driven approach that counts on business decision making on the other side. This article is a collection of representative current state of ethical AI with reference from survey, results and content from various forums.
Ethics in AI is not quite new but an evolving area pertaining to modernized needs of its usage
Ethics in AI is not quite new but an evolving area pertaining to modernized needs of its usage. A significant amount of effort and capital is spent on this through Government initiatives, Non-Profit organizations and even private companies to foster a system of formulating ethical guidelines while developing AI products all over the world in understanding the perils of using AI at Scale over the Enterprise. Before having a quick introduction to Ethics in AI, let’s first understand what it means. Though we have differentiated study to understand ethics, its moral values and dealing with a moral driven course of action, Applied Ethics concerns what a moral agent is obligated or permitted to do in a specific situation or in a particular domain of action. AI Ethics is considered to be a sub field of Applied Ethics.
Realizing AI at the scale of enterprise level is considered so beneficial that the MIT survey published in MIT Technology Review Insights archive page, March 26, 2020[1] finds it with the limitation of handling ethical principles. AI at scale suffers from hindering realization because of the availability of unrestricted data, technology interventions like Algorithmic Bias, awareness and adoption on the landing of new AI Business Model expected to grow with the needs.
Importance of Ethics in AI is clearly understood with the below map (Figure 2) showing the adoption strategy of Ethics in AI all over the world (reference cited in INDIAai, affiliated to Ministry of Electronics and Information Technology, India).
Earlier when Issac Asimov’s three laws of robotics were introduced, its significance was negligible and considered fictitious. Nevertheless it has now gained and imprinted a high impact with the modern-day AI products and its development. Consolidating from the trending values and norms derivation from various domains the top five are listed below:
Non-Maleficence — Primarily addresses the negative consequences and risks of AI. The act of “non-harming” is termed as non-maleficence. The algorithmic decision-making capability of AI finds different paths having a possibility to harm human beings.
Transparency — Includes “explainability”, “interpretability”, “understandability”, and “black box”. It is about how much it is possible to understand a system’s inner workings “in theory”. It can also mean the way of providing explanations of algorithmic models and decisions that are comprehensible for the user. This deals with the public perception and understanding of how AI works. Transparency can also be taken as a broader socio-technical and normative ideal of “openness”.
Accountability — The state of being responsible or answerable for a system, its behavior, and its potential impacts. Accountability is an acknowledgement of responsibility for actions, decisions, and products. In AI ethics, there are three different senses or dimensions of accountability:
Dimension 1: The question of determining the responsibility — which individuals (or groups) are accountable for the impact of algorithms or AI? Who is responsible for what effect within the overall socio-technical system?
Dimension 2: A feature of the societal system that develops, produces, and uses AI
Dimension 3: A feature of the AI system itself
Fairness — Philosopher John Rawls remarks, the stability of a society — or any group — depends upon the extent to which the members of that society feel that they are being treated in a just manner. When society members feel that they are treated in an unfair manner, it usually creates a foundation for social unrest, disturbances, and strife. People hold social unity only to the extent that their institutions are fair. This could be achieved by retaining a non-discriminate approach towards biased data and algorithms. One could try to salvage what’s possible from the dataset, and see if it could be made at least less biased. One common technical fix on the data set is called anti-classification, or the removal of explicit protected variables from the data. This means erasing information like gender or ethnicity, and their proxies from the data. Here proxies mean features that are strongly correlated with the protected characteristics. Like in the case of the recruiting algorithm mentioned earlier, if a person’s Curriculum Vitae contains references to maternal leave or a women’s college, the algorithm could still make gendered predictions even if an explicit gender variable is left out.
Privacy — The real issue of privacy is the way the public information flows over the network with or without the knowledge of one. Understanding the term “Contextual Integrity” ensures to get better results on AI product development. We understand better considering an example. Facial recognition as such is considered a privacy breach though the face itself is supposed to be public. Successful implementation of projects like Odysseus carried out in London depicts how all the data could be anonymized ensuring privacy and security.
The whole purpose of these discussions on AI in Ethics, research and capital investments is to initiate the move of designing from a “Reactive” based AI system to “Preventive” based AI system. The above-mentioned ethical principles play a vital role as they have direct impact on the business and its derived output on AI products.
The underlying idea is to have a proactive system fueled by AI technology that solves business problems considering the risk analysis of the ethical principles listed above are substantial. We need to derive a way that we handle the system, the below steps illuminate on the possible adoption methodology:
- Identify and list down actions to perform
- Perform benefit — impairment analysis for each action
- Derive the cost-benefit analysis for the benefited action from above
- Choose and realize the action that provides greatest benefits
Future research is required to identify interventions that do influence decision-making, such as by helping developers identify parallels between their decisions and infamous software news stories.
Reference:
