Defining AI & Machine Learning: Why It Is So Difficult

Artificial intelligence litigation is not a brand-new category of “AI law.” It is classic business litigation: contracts, IP, labor and employment claims, unfair competition, and investigations, to name a few civil areas of law. Any discussion of AI litigation must begin with a deceptively simple question: what is artificial intelligence? There is no unified, universally accepted definition. At its broadest, Artificial Intelligence refers to computer systems designed to perform tasks that ordinarily require human intelligence — recognizing patterns, making predictions, generating language, interpreting images, or making decisions.

The challenge deepens at the subset level. Machine Learning, AI’s most commercially significant subset, refers to systems that learn from data rather than following explicitly programmed rules, improving through iterative adjustment of internal parameters during training. Deep Learning, a further subset, uses layered neural networks and powers the large language models, image generators, and speech recognition systems at the center of most current AI litigation. Each definition captures a different slice of the technology, creating different boundaries for compliance and liability. For companies building, deploying, or licensing AI, careful attention to definitional precision in contracts, terms of service, and corporate policies is a litigation-avoidance strategy. And when disputes arise despite that care, the ability to argue persuasively about what a term means, grounded in both technical fluency and legal authority, is often dispositive.

The “Black Box” Problem in AI Litigation

The “black box” problem refers to the inability to fully explain how an AI system arrives at a particular output or decision. AI is not just software; it is an industrial stack, and legal disputes can arise at every layer. NVIDIA CEO Jensen Huang has described AI as a five-layer cake: energy at the foundation; chips and computing infrastructure including GPUs and servers; cloud data centers that host and run that hardware at scale; AI models trained using that compute; and the applications where users experience the technology. It is through the repeated adjustment of internal parameters called weights, a process known as backpropagation, that a model learns.

The architecture that enables learning, however, also gives rise to one of AI’s most pressing legal challenges. Because decisions in deep neural networks emerge from millions or billions of weighted interactions, no single node may be “responsible” for a given output. In copyright disputes, this opacity makes it hard to establish whether a model’s output derives from specific copyrighted training data. In employment discrimination cases, it complicates efforts to prove an automated tool rejected candidates based on a protected characteristic. In discovery, it raises questions about what records exist and what constitutes a sufficient explanation of a system’s decision.

This is the technical root of the “black box” problem, and it runs through nearly every type of AI litigation. Litigating these disputes effectively requires understanding not just the application layer where harm may manifest, but the infrastructure, data, and training processes underneath it.

Agentic AI & the Accountability Gap

The challenge sharpens with agentic AI: systems that act autonomously without a human in the loop to complete a task or series of tasks. In commercial settings, the stakes are high. A Canadian tribunal has already held that an airline AI chatbot made false representations to a consumer, rejecting the argument that an AI system acted “independently” and could not be considered the company’s agent. That ruling signals broader accountability questions that will only intensify as AI agents operate with greater autonomy. Companies deploying agentic AI should evaluate now whether their contractual frameworks, terms of service, and oversight protocols are prepared for the liability exposure these systems create.

AI Copyright, Licensing, & Training-Data Disputes

The foundational question in AI copyright law is whether using copyrighted works to train machine learning models constitutes infringement or fair use. Federal courts are actively adjudicating this question in cases involving publishers, authors, visual artists, music labels, and software developers. A key ruling has established that even factual editorial compilations warrant copyright protection when copied to develop competing AI products. In parallel, pending litigation is testing whether near-copy outputs generated through targeted prompting constitute compelling evidence of infringement. These disputes are reshaping licensing economics across publishing, music, visual arts, and open-source software.

Trade Secrets & Confidential AI Assets

Model weights, training datasets, proprietary prompts, fine-tuning methodologies, and evaluation benchmarks represent enormous value with enormous vulnerability. Trade secret disputes in AI are rising as employees move between competitors, partnerships dissolve, and vendors gain access to sensitive model architectures. Habibian Law litigates to protect these assets and to defend startups and emerging companies facing misappropriation claims, with a focus on the technical specificity these cases demand.

Algorithmic Discrimination & Employment Claims

Automated decision-making tools used in hiring, lending, and housing are under increasing legal scrutiny. Federal enforcement agencies have already secured consent decrees against companies whose automated recruitment software systematically rejected applicants based on age, establishing that existing anti-discrimination law applies fully to automated systems. New York City’s Automated Employment Decision Tool (AEDT) Law requires annual bias audits, public posting of results, and advance notice to candidates before any AI-driven screening tool is used. Companies that deploy these tools face real litigation risk. 

Biometric Privacy

Facial recognition and biometric identification technologies deployed in commercial spaces have triggered significant litigation in New York. The deployment of AI-driven identification tools raises complex discovery obligations, and the NYC Biometric Identifier Information Law provides an enforcement framework. As these technologies proliferate, exposure grows.

Antitrust & Algorithmic Market Power

Antitrust claims tied to AI are here. Federal courts have addressed when algorithm-driven conduct by a dominant platform can constitute antitrust injury, including when algorithmic changes are alleged to suppress competition. The DOJ’s continued litigation in the search and AI markets, and the FTC’s ongoing review of major AI investment structures, signal that antitrust exposure for AI companies will expand significantly.

Deepfakes, Synthetic Media, & Digital Replicas

AI-generated intimate images, impersonation, and fraud present serious civil and criminal exposure. The federal TAKE IT DOWN Act criminalizes nonconsensual publication of intimate images targeting AI-generated deepfakes with penalties up to two years’ imprisonment. New York provides additional state law protections, including a private right of action for deepfake intimate images. The New York Fashion Workers Act separately requires written consent for any AI-generated digital replica of a model’s likeness.

Discovery & AI Integrity

AI discovery disputes are a new battleground, with courts confronting questions about data, audit logs, model evaluation records, and vendor documentation. The integrity of AI in the litigation process itself is also under scrutiny: courts have imposed sanctions where attorneys submitted AI-fabricated case citations, establishing a non-delegable duty of verification that extends through every stage of litigation, including on appeal. 

Why Habibian Law for AI Litigation?

AI litigation demands lawyers who understand the technology, know the case law as it develops in real time, and can translate complex technical arguments into persuasive courtroom advocacy. We built this practice to do exactly that.