Ethics is a set of moral principles that help us discern right from wrong.
The important question of how human designers, producers, and users should act in order to minimize any ethical harms that AI may cause in society—harms that may result from subpar (unethical) design, unsuitable application, or misuse—is at the heart of AI ethics.
There are three subfields of ethics: Meta-ethics, Normative Ethics, and Applied Ethics. AI ethics is a subfield of Applied Ethics. Applied ethics, also called Practical Ethics, is the application of ethics to real-world problems.
AI Ethics asks what actions should be taken by creators, producers, regulators, and operators to reduce the ethical risks that AI poses to society.
Ethics of AI: Major Themes

1.Bias

Humans create AI, which means it is susceptible to bias. Algorithmic bias is a lack of fairness caused by a computer system’s performance.

AI models have demonstrated their ability to replicate and institutionalize bias. In 2015, Amazon found out that the algorithms they used for shortlisting resumes were biased against women. The model was trained based on resume submissions over the past decade where most of the applicants were men, so it favored men over women. An AI model called COMPAS, which is used across the United States to predict recidivism, was found to be biased against black people.
Human bias has been extensively researched in psychology for many years. It results from an implicit association that reflects a bias that we are unaware of and how it can influence the outcome of an event.
Human prejudices have been found to contaminate algorithms and data through AI in recent years. In order to avoid bias in AI systems, we must understand the underlying causes. Read more on AI bias in one of our previous posts.

2.Transparency in AI

One reason why people might fear AI is that AI technologies can be hard to explain.
An AI transparent system solves the “black box dilemma” when technology must explain how it reached its conclusions.
Organizations don’t always have the expertise to create and maintain fair and accurate AI-based systems. For most companies, the supply chain for AI is very long and complicated.
Transparency can be difficult to achieve with modern AI systems, particularly those based on Deep Learning systems. Deep Learning systems are built on artificial neural networks (ANNs), which are a collection of interconnected nodes inspired by a simplified model of how neurons connect in the brain. A feature of ANNs is that once trained with datasets, any attempt to examine the internal structure of the ANN to understand why and how the ANN makes a particular decision is nearly impossible. These systems are known as “black boxes.”
To make models more transparent, the following can be done:
Make use of simpler models. This, however, frequently sacrifices accuracy in favor of explainability.
Combining simpler and more complex models. A sophisticated model allows the system to perform more complex computations, but a simpler model can provide transparency.
Inputs should be modified to track relevant input-output dependencies. It is possible that input manipulations may have an impact on classification if they change the overall model results.

3. Privacy

This is based on the public and political demand to respect a human’s personal information. Data privacy, data protection, and data security are frequently brought up when discussing privacy.
For instance:
The biggest concern for now: Personal Data
Corporations collect personal data for profit and profile people based on their behavior (both online and offline). They know more about us than ourselves or our friends — and they are using and making this information available for profit, surveillance, security, and election campaigns.
One of the major practical difficulties is to actually enforce the regulation, both on the level of the state and on the level of the individual who has a claim. Hence, aside from government regulation, each company developing AI should prioritize privacy and generalize their data so that individual records are not stored. An example of a legal framework that ensures data protection and privacy is
GDPR
The General Data Protection Regulation (GDPR) is a legal framework that sets guidelines for the collection and processing of personal data from individuals who live in the European Union. It was developed in 2016 to give people more control over their data while protecting their personal information.

4. Safety

The first question to ask about any technology is whether it works as it should. Will AI systems deliver on their promises, or will they fail? What will be the consequences of their failures if and when they occur?
Take the example of self-driving cars. Each decision the car makes (whether to turn, stop, slow, increase speed) is all governed by neural networks. Neural networks are black box models that are hard to interpret.
The British political consulting firm, Cambridge Analytica, was revealed to have harvested the data of millions of Facebook users without their consent in order to influence the US elections, raising concerns about how algorithms can be abused to influence and manipulate the public sphere on a large scale.
Several companies today like Meta and Tiktok heavily focus on attracting audience attention on their platforms because ads account for a nice chunk of their revenue. In order to maximize people’s attention, they optimized their algorithms by gathering information and recommending content through AI.
An MIT research revealed that tech giants are paying millions of dollars to the operators of clickbait pages, putting the world’s information ecosystems at risk.
To ensure that your AI system operates safely, you must prioritize the technical objectives of accuracy, reliability, security, and robustness.