What is AI ethics?

Examples of AI ethics issues include:

• Data responsibility and privacy
• Fairness
• Explainability
• Robustness
• Transparency
• Environmental sustainability
• Inclusion
• Moral agency
• Value alignment
• Accountability
• Trust
• Technology misuse

Most organizations today use AI tools and big data in some capacity to support automation and enable data-driven decision-making. While the intention is usually to improve business outcomes, companies can experience unforeseen consequences in their AI applications, often due to poor upfront research and design and biased datasets. These issues help highlight the importance of the ethical use of AI systems.

The research and data science communities have developed guidelines for the ethics of AI to help with mitigating these unforeseen consequences and unfair outcomes. Leading AI companies have also taken an interest in shaping these guidelines, as they know that failing to uphold ethical standards can harm their products, services and users. Lack of diligence in AI ethics can result in reputational, regulatory and legal exposure, leading to costly penalties.

Innovation tends to outpace government regulation, particularly in the case of emerging technologies such as AI. Over time, one can expect that more governments will issue legally mandated AI protocols to protect human rights and civil liberties. While some regulation does exist­—such as the EU AI Act—many existing AI ethics guidelines are voluntary frameworks put forward by scientists and industry groups.

The academic community has long relied on the ethical principles outlined in the 1979 “Belmont Report”¹  to guide ethics in experimental research and algorithmic development.

These principles can offer a basis for the ethical research, design and use of AI and machine learning. However, because they were devised with human subjects in mind­—and long before modern generative and agentic AI systems—they are not sufficient on their own. They should be viewed as foundational, but not a comprehensive set of AI principles.

The principles of the Belmont Report are:

There are several issues at the forefront of ethical conversations surrounding AI technologies in the real world. These concerns reflect the broader potential impacts of AI on society. Some of these concerns include:

The release of ChatGPT in 2022 marked a true inflection point for artificial intelligence. The abilities of OpenAI’s chatbot—from writing legal briefs to debugging code—opened new possibilities for what AI can do and how it can be applied across almost all industries.

ChatGPT and similar AI tools are built on foundation models, AI models that can be applied to a wide range of downstream tasks. Foundation models are typically large-scale generative models, comprising billions of parameters, that rely on unlabeled training data and use self-supervision. Foundation models, often referred to as large language models (LLMs), can apply what they’ve learned in one context to another, making them highly adaptable.

Yet there are many potential issues and ethical concerns around foundation models, such as bias, generation of false content (including unreliable or misleading output) and lack of explainability. Many of these issues are relevant to AI in general but take on new urgency in light of the power and availability of foundation models.

Foundation and generative AI (gen AI) models also introduce ethical challenges that stem from their scale, general‑purpose design and ease of reuse across domains. Because these models can be deployed in contexts far removed from their original training intent, questions of responsibility and accountability become far more complex. It is often unclear how ethical obligations should be shared between model providers and their downstream deployers. Whose responsibility is it to prevent a model from, for example, misusing sensitive data or leaking intellectual property: the provider, the deployer or the end user?

In addition, generative AI systems can produce convincing but incorrect or misleading outputs and enable misuse at an unprecedented scale. As a result, ethical guidelines for foundation models often emphasize clear documentation, usage constraints, ongoing monitoring and human oversight throughout deployment.

The technological singularity is a theoretical scenario where technological growth becomes uncontrollable and irreversible, culminating in profound and unpredictable changes to human civilization. While this topic garners a lot of public attention, many researchers are not concerned with the idea of AI surpassing human intelligence in the near or immediate future.  

Strong AI (AI that would possess intelligence and self-awareness equal to humans) and superintelligence are still hypothetical. However, the ideas raise some interesting questions and ethical implications concerning how we use AI in autonomous systems, such as self-driving cars.

For example, it’s unrealistic to think that a driverless car would never get into an accident. But who is responsible and liable under those circumstances? Should we still pursue autonomous vehicles, or do we limit the integration of this technology to create only semiautonomous vehicles that promote safety among drivers?

These sorts of ethical dilemmas are far from settled, but they are the debates that new, innovative AI technology sparks.

While a lot of public perception around artificial intelligence centers around job loss, this concern can be reframed. With every disruptive, new technology, the market demand for specific job roles shifts.

For example, when personal computers became widespread, roles such as typist and file clerk largely disappeared. At the same time, entirely new job categories emerged: IT support, software development and data entry.

Artificial intelligence should be viewed in a similar manner. AI initiatives are not necessarily going to shrink the workforce. Rather, AI is likely to shift the job demand to other areas. Individuals will be needed to manage AI systems and address the more complex problems that arise in their industries.

The most important aspect of artificial intelligence’s impact on the job market is helping people transition to new areas of market demand.

AI models rely on data. Therefore, data privacy, data protection and data security are core concerns of AI ethics.

Policymakers have advanced significant pieces of data protection legislation over the last decade, and the rules they set forth largely apply to AI tools as well as traditional software.

For example, the EU’s General Data Protection Regulation (GDPR) gives individuals in the European Union and European Economic Area more control of their personal data. In the United States, state-level policies such as the California Consumer Privacy Act (CCPA) require businesses to inform consumers about the collection of their data.

Government bodies are also starting to enact AI-specific rules around the collection, storage and use of personally identifiable information (PII). For example, in 2024, California began drafting rules on AI and automated decision-making technology. Those rules have since been finalized, and businesses must comply by 1 January 2027. 

While companies typically have good intentions for their AI and automation efforts, biases embedded in training datasets and AI algorithms can lead to discriminatory outcomes.

For example, Amazon scrapped an early AI-powered recruiting tool when it was discovered that the tool systematically scored women lower than men. Why? Because the tool was trained on a pool of resumes drawn largely from men.

Bias and discrimination aren’t limited to the human resources function either. They can be found in several applications from facial recognition software to social media algorithms. For example, facial recognition tools routinely have a harder time identifying people of color.

As businesses become more aware of the risks, they’ve also become more active in the discussion.

For example, IBM withdrew its general-purpose facial recognition and analysis products in 2020. At the time, IBM CEO Arvind Krishna wrote:

“IBM firmly opposes and will not condone uses of any technology, including facial recognition technology offered by other vendors, for mass surveillance, racial profiling, violations of basic human rights and freedoms or any purpose which is not consistent with our values and principles of trust and transparency.”

The rapid adoption of generative AI has intensified concerns around synthetic content such as deepfakes, automated misinformation and the erosion of trust in digital information.

As AI‑generated text, images, audio and video become increasingly difficult to distinguish from human‑created content, the need for transparency now extends beyond explainability of models to the traceability and disclosure of outputs. In response, policymakers and technology providers are advancing content provenance mechanisms. These mechanisms can include things such as labeling requirements, watermarking and cryptographic metadata to help users identify AI‑generated material.

These approaches can introduce ethical tensions, including tradeoffs between robustness, privacy and freedom of expression, but they are increasingly viewed as essential to preserving trust in digital ecosystems.

While there is no single global law governing artificial intelligence, AI regulation has moved rapidly from voluntary guidelines to enforceable legal frameworks in multiple jurisdictions.

The European Union’s Artificial Intelligence Act (EU AI Act), which entered into force in 2024, establishes binding obligations for certain AI practices. These obligations include bans on unacceptable‑risk systems, transparency requirements for AI‑generated content and lifecycle governance rules for high‑risk and general‑purpose AI models.

The proliferation of laws such as the EU AI Act signals a broader shift in AI ethics from aspirational principles toward accountability mechanisms that combine ethical standards with legal compliance.

Artificial intelligence performs according to how it is designed, developed, trained, tuned and used. AI ethics is all about establishing standards and guardrails around each of these phases of an AI system’s lifecycle.

Organizations, governments and researchers alike have drafted frameworks to address current AI ethical concerns and shape the future of AI. While guidelines can vary, there is some consensus around core AI governance activities.

AI ethics is increasingly understood not as a set of principles, but as an ongoing capability embedded throughout the AI lifecycle. Effective governance requires continuous processes for assessing risk, documenting design decisions, monitoring deployed systems and responding to unexpected behaviors or harms. As AI systems evolve over time, ethical oversight must evolve with them, supported by clear ownership and measurable controls rather than static policies alone.

Governance helps ensure that AI systems are operating according to an organization’s principles, values and responsible AI practices.

Common components of an AI governance program include:

• Defining the roles and responsibilities of people working with AI to ensure appropriate human oversight.
  
  
• Educating all people involved in the AI lifecycle about building and using AI in a responsible way.
  
  
• Establishing processes for building, managing, monitoring, communicating about AI and conducting audits of AI systems.
  
  
• Leveraging tools to improve AI’s performance and trustworthiness throughout the AI lifecycle.

An AI ethics board is another effective governance mechanism. For example, the IBM Responsible Technology Board, composed of diverse leaders from across the business, provides a centralized governance, review and decision-making process for AI ethics policies and practices at IBM.

Organizations can build trustworthy AI by grounding their work in clear ethical principles applied consistently across products, policies and processes. These principles should focus on concrete areas, such as explainability or fairness, around which specific standards and practices can be developed.

When ethics are built into AI from the start, it is capable of tremendous good. Fields such as healthcare are already experiencing this good. AI is transforming disciplines such as radiology, detecting abnormalities in scans faster and more accurately than traditional methods.

Many organizations find that they are better positioned to reap these benefits if they treat the ethics of artificial intelligence as a practical discipline that combines values, governance and accountability to support responsible innovation at scale.

Ethical standards are not the primary concern of data engineers and data scientists in the private sector. However, several organizations have emerged to promote ethical conduct in the field of artificial intelligence.

These organizations include:

• AlgorithmWatch focuses on explainable and traceable algorithms and decision processes in AI programs.
  
  
• The AI Now Institute, a nonprofit at New York University, researches the social implications of artificial intelligence.
  
  
• The Defense Advanced Research Projects Agency (DARPA) of the US Department of Defense promotes explainable AI and AI research.
  
  
• The Center for Human-Compatible Artificial Intelligence (CHAI) is a cooperation of various institutes and universities to promote trustworthy AI and provable beneficial systems.

IBM’s Principles for Trust and Transparency dictate its approach to data and AI development. These principles are:

1. The purpose of AI is to augment human intelligence. This means that we do not seek to replace human intelligence with AI, but support it.
   
   
2. Data and insights belong to their creator. Organizations should not be required to relinquish rights to their data—or insights derived from it—to have the benefits of AI solutions and services.
   
   
3. AI systems must be transparent and explainable. IBM believes that technology companies need to be clear about who trains their AI systems, what data was used in that training and, most importantly, what went into their algorithms’ recommendations.

IBM has also developed five pillars to guide the responsible adoption of AI technologies.

• Explainability: An AI system should be transparent, particularly about what went into its algorithm’s recommendations, as relevant to various stakeholders with various objectives.
  
  
• Fairness: This pillar refers to the equitable treatment of individuals, or groups of individuals, by an AI system. When properly calibrated, AI can assist humans in making fairer choices, countering human biases and promoting inclusivity.
  
  
• Robustness: AI-powered systems must be actively defended from adversarial attacks, minimizing security risks and enabling confidence in system outcomes.
  
  
• Transparency: To reinforce trust, users must be able to see how the service works, evaluate its functionality and comprehend its strengths and limitations.
  
  
• Privacy: AI systems must prioritize and safeguard consumers’ privacy and data rights and provide explicit assurances to users about how their personal data will be used and protected.