Legal, Ethical, and Technical Issues of AI Systems and Their Solutions

1. Overview

a. Definition

The legal-liability, ethical-trust, and technical-safety problems that arise with the rapid spread of AI systems; managing these in an integrated manner to realize Trustworthy AI is the task.

The reason AI's issues are special is that the three layers are intertwined with one another. Algorithmic bias (ethical) leads directly to a discrimination lawsuit (legal), and hallucination (technical) gives rise to liability (legal) for spreading misinformation. Therefore, addressing only one layer merely shifts the problem to another layer, and law, ethics, and technology must be addressed in an integrated manner.

b. Background and Necessity

As generative AI became popular, anyone can now easily use powerful AI, and accordingly the problems of accountability, bias, and safety have surfaced socially. As decisions made by AI directly affect people's lives—such as in hiring, lending, and healthcare—the question of "how far to trust a machine's judgment and who is responsible" has led to regulation. As systems such as the EU AI Act and Korea's Framework Act on AI are strengthened and social acceptance is demanded, preemptively managing AI's issues has become directly tied to corporate survival.

2. Issue Classification

flowchart TB
  AI[AI system issues] --> L[Legal]
  AI --> E[Ethical]
  AI --> T[Technical]
  L --> L1[Liability / copyright<br/>privacy / compliance]
  E --> E1[Bias / fairness<br/>transparency / explainability]
  T --> T1[Hallucination / robustness<br/>security / adversarial attacks]

Issues are divided into three according to their nature. Legal issues are matters of systems, contracts, and rights, addressing "who is responsible and what is lawful," and ethical issues are matters of norms and values, asking "is it fair and transparent." Technical issues are matters of the system's own defects—"is it accurate and safe." As noted earlier, these three are not independent but interact in a chain.

3. Content and Solutions by Issue

Because the issues of each layer have different causes, the solution approaches must also differ. Legal issues are addressed with systems and contracts, ethical issues with verification and transparency, and technical issues with engineering defense. For example, a copyright dispute over training data (legal) must be addressed by training with data whose license is clear and securing the basis for use by contract; gender bias in a hiring AI (ethical) is mitigated by data rebalancing and verification with fairness metrics; and a chatbot's hallucination (technical) is suppressed with RAG and fact-checking.

In particular, bias is usually the result of social bias already present in the training data being reflected and amplified in the model, so it is important that fixing the algorithm alone is insufficient and one must check from the data stage.

4. Integrated Solution

Beyond individual responses, there must be a governance system that runs through the entire development-and-operation lifecycle so that issues do not recur. Once AI governance establishes responsibility and ethics principles, XAI makes the basis of judgments transparent, MLOps continuously monitors performance, bias, and drift during operation, and compliance aligns all of this with legal requirements.

flowchart LR
  G[AI governance] --> X[XAI / transparency]
  X --> M[MLOps / monitoring]
  M --> R[Regulatory compliance]

The reason one must not validate a model once and be done but continuously monitor with MLOps is that performance and fairness can gradually collapse due to data drift, where the actual operational data diverges from the training-time data.

5. Related Regulations and Standards

Regulations and standards present the baseline for response. The EU AI Act takes a risk-based approach that regulates more strictly the greater the risk—prohibiting uses such as social scoring and imposing strict obligations on high-risk uses (healthcare, hiring, etc.). The NIST AI RMF and ISO/IEC 42001 provide management systems for executing this within the organization.

6. Considerations and Implications

From the professional engineer's perspective, the most important principle is embedding from the design stage rather than after-the-fact regulation (Responsible AI by Design). Because the cost of restoring trust is far greater when one responds after a problem erupts, bias checks, explainability, and safety must be included as requirements from the early development stage. Second, the triangular balance of technology, systems, and ethics is the condition for trustworthy AI, and if any one gets ahead, overall trust collapses. Third, especially in high-risk areas, a control mechanism is essential that maintains human oversight (Human-in-the-loop) so that AI does not monopolize the final decision and places ultimate responsibility with a person. In the end, trustworthy AI is a matter of social acceptance rather than technical performance, and the organization that secures it will have competitiveness in the AI era.

In one line: AI systems have legal (liability/copyright), ethical (bias/transparency), and technical (hallucination/security) issues intertwined with one another, and the key is to integrate AI governance, XAI, MLOps, and regulatory compliance from the design stage to realize trustworthy AI under human oversight.