“The future is already here — it’s just not evenly distributed.”
— William Gibson, science fiction writer
The practice of law stands on the brink of a radical transformation. AI-powered tools promise to reshape traditional workflows, save time, and reduce costs. From incumbent giants like Westlaw or Relativity to recent efforts by iManage and Litera, along with a legion of emerging startups like Harvey, Legora, or DeepJudge, the legal tech landscape is evolving rapidly. And so is the technology itself.
This surge of innovation has captivated lawyers, technologists, entrepreneurs, and venture capitalists alike, suggesting a future of greater efficiency and new players in the legal field. Academics have already studied the potential gains in efficiency and accuracy in the law of the future.1
The legal industry presents a paradoxical environment for AI development, simultaneously offering the most promise and the greatest challenges.2 The current focus on enhancing efficiency and accuracy within existing workflows, i.e. streamlining existing processes, is valuable but iterative. The drive for incremental improvements risks overshadowing AI’s true potential to fundamentally reimagine the
practice of law.
“Technology is neither good nor bad; nor is it neutral.”3
— Melvin Kranzberg
We are on the cusp of what Reid Hoffman, founder of LinkedIn, called a “Cognitive Industrial Revolution” where the shift from mechanical
processes to AI-enhanced systems will drive innovation and enhance human decision-making in unprecedented ways.4
AI’s ability to process vast amounts of data and identify patterns at unprecedented speeds promises to be a powerful tool across many industries. Transformer models, the technology that underpins most frontier large language models, are designed to pattern and generate human language by analyzing vast amounts of text data at unprecedented scale.5 Their abilities with translation, summarization, and conversational AI make them a game changer for legal applications. They will empower legal professionals, including those with limited technical skills, to build and refine their own idiosyncratic workflows. And for the very best legal professionals, they will use these tools to laser-focus on client value, rather than efficiency.
For power users of these AI systems, it’s evident that their capabilities are remarkable, often astounding, even if they don’t truly think or understand in a human sense. Users experimenting with these systems frequently find themselves captivated by AI’s linguistic prowess, ability to follow instructions, and interpret ambiguous requests. This fascination is not unfounded; Bill Gates considers AI one of the most revolutionary technologies he’s witnessed in his lifetime.
In various use cases — from client intake, conflicts, communications and marketing, and billing guidelines to advanced research and data sorting and simulation — we are in a new age where natural language processing can drive new processes and innovation. We can tackle problems that were impossible only a short few years ago.
And this is just the beginning. Or to put it in Silicon Valley marketing parlance: What you see with today’s tools truly is the worst this technology will ever be.
Startups are pouring into less-explored areas of the law such as predictive outcomes or analyzing a judge’s entire corpus of opinions to detect detailed semantics, outliers, or even judicial philosophies. Imagine insights like, “This judge responds well to sports analogies” or “This judge is a legal positivist at heart.”
The legal industry’s penchant for precedent makes pattern-recognition and predictive machine learning tools transformative solutions to complicated problems. The main challenge? The legal field hasn’t systematically prepared high-quality datasets for sophisticated legal questions or tasks. This data gap explains why benchmarking and evaluation tools will continue to proliferate in this space, becoming increasingly specialized for specific practice areas and use cases.
AI technology will revolutionize the legal profession by enabling unprecedented innovations in workflows and methodologies. While the field has previously benefited from advancements like word processing, simple document comparison tools, timekeeping tools and basic search capabilities, the impending AI-driven transformation will be far more profound.
The most sophisticated AI tools offered to lawyers in the past decade (adopted with notable hesitation) were machine-learning solutions like Kira, Eigen, and Luminance, all built on earlier-generation ML techniques. Now, amid the current AI revolution, these established platforms face pressure to rapidly integrate large language models (LLMs) into their products and workflows or risk obsolescence.
The scale of this change is likely to eclipse even the monumental impact that spreadsheet software had on finance and accounting, fundamentally reshaping how legal professionals approach their work and deliver value to clients.
“Any sufficiently advanced technology is indistinguishable from magic.”6
— Arthur C. Clarke, author
Current AI systems in law primarily rely on two key technologies: natural language processing (NLP) and ML. While these tools have demonstrated impressive capabilities,7 they fundamentally operate based on statistical correlations derived from existing data. At their current best, these systems can generalize patterns to some extent, aiding in the analysis of situations that may be at the periphery of direct training data.
This is fundamentally different from human wisdom, which is grounded in infinite context: experience, ethical considerations, and factors beyond mere rationality. The logic underpinning formal systems — such as algorithms and most machine learning models — inevitably lacks the ability to self-reference or relate to realities beyond what the system contains.8
Consider prominent AI researcher Andrej Karpathy’s analogy: Large language models are essentially a form of lossy compression of the internet’s data, much like how an MP3 compresses audio. Just as a live musical instrument produces richer resonance than any analog or digital recording, language models can only offer a degraded representation of reality. To the extent they have any concept of the “real world,” it’s at best a hazy, low-fidelity abstraction. The system could give you perfect turn-by-turn directions from Grand Central Terminal to JFK Airport without ever understanding the feel of pavement underfoot or the sensation of objects in motion. It knows the map, but not the territory.
But our relationship with technology is shifting, along with the technology itself. In the past, most technology operated like an on/off switch: It either worked or it didn’t. The light bulb turned on or it didn’t. We expected deterministic, predictable outputs at scale. The promise and peril of large language models lie in their inherently non-deterministic nature. What lawyers call “hallucinations” is what an engineer might call the model’s generative nature operating without sufficient grounding — a feature, not a bug. That is, until it enters a legal context where it becomes a critical failure. Still, this is a fundamentally different relationship with technology.
Even though we anthropomorphize these systems, and the language outputs are hypnotizingly good, can we truly believe that AI comprehends emotionally charged and complex human experiences like love, compassion, forgiveness, and sacrifice? These are not merely complex equations; they are deeply human and likely resist reduction to any mathematical or “atomized” model.
More and more advanced AI systems are in development today, ranging from sophisticated prompting agents with schemas to orchestrating different models with varying expertise. Agentic AI, for example, employs system designs that not only process information and make decisions but also operate with a degree of autonomy, setting and pursuing their own goals based on learned or pre-programmed courses or objectives. These systems, which are designed to adapt to changing circumstances and execute complex tasks without constant human oversight,9 could theoretically come close to addressing many of law’s most tedious, complicated, and time-consuming tasks.10
Despite their impressive capabilities, AI systems remain fundamentally distinct from human cognition. The intricacies of human thought — our ability to reason abstractly and analogously, empathize, and make nuanced judgments — extend beyond the pattern-based approaches of current AI. Probabilities and pattern-matching, no matter how sophisticated, cannot translate directly into wisdom.
Human life is inherently chaotic and unpredictable; our sense of control is often illusory. This unpredictability matters for law. When we place blind faith in legal rules as mechanical absolutes, we lose access to the underlying wisdom, integrity, and virtue that make laws just in the first place. Laws without human judgment become mere algorithms — precise perhaps, but divorced from the messy realities they’re meant to govern.
This distinction becomes evident when we challenge AI with tasks requiring creative analogical thinking. Consider the prompt: ‘ABC is to ABD as XYZ is to what?’ The mechanical answers—XYA or YZA—simply replace the last letter. But there’s a more elegant solution: ‘WYZ.’ This answer recognizes that just as ABC moves forward to ABD (C→D), XYZ should also shift—but since Z is at the alphabet’s end, the creative insight is to shift backward instead (X→W), preserving the pattern’s spirit while respecting the boundary constraint. It’s easier to appreciate this visually or aesthetically. This solution embodies the type of lateral thinking long celebrated in studies of human creativity and cognition.11 Yet current language models struggle with this leap.
Somewhat impressively, later ‘thinking’ versions showed progress; ChatGPT-4o’s purported chain-of-thought process revealed it had considered WYZ, a remarkable leap. Yet even today’s frontier models still default to more mechanical answers like XYA or YZA. When asked to generate multiple possibilities, WYZ doesn’t even crack the top 10 anymore. The models can follow patterns, but they miss the creative insight that makes the answer beautiful.
For the legal profession, this underscores a crucial point: While AI will undoubtedly become an invaluable tool that revolutionizes many aspects of legal practice, it will never truly replicate the full spectrum of human judgment, creativity, and ethical reasoning that lies at the heart of the law. AI will augment and enhance legal work, but the uniquely human elements of legal practice — persuasiveness, fairness, empathy, nuanced interpretation, and principled decision-making, will remain irreplaceable and likely irreducible.
THE COMPLEXITY OF LAW AND AI’S CHALLENGES
As AI systems master physical tasks — laundry-folding robots, gardening robots, pool cleaners, autonomous vehicles — they face a far more intricate challenge: navigating our evolving and invisible legal and regulatory landscape. This gap between AI’s technical and physical capabilities and law’s inherent complexity presents both a critical challenge and an unprecedented opportunity.
Consider a thought experiment: Imagine an AI system with perfect memory of every law, regulation, and legal case ever decided.
Could this system predict the future of law? Almost certainly not. Law is not merely a database of past decisions and rules or the dry application of IRAC (issue, rule, application, conclusion). It’s a living system shaped by the infinite context of societal norms, ethical dilemmas, and human judgment. When laws conflict, someone must choose. When culture shifts, law follows (and sometimes leads or lags). An AI system would need judgment, not just memory.
AI’s limitations in this domain become self-evident12 when we consider several key factors:
- Lack of consensus: Lawyers and judges often disagree on what constitutes a “correct” decision, reflecting the subjective nature of legal interpretation.
- Changing societal norms: Legal standards evolve with societal values, making reliance on static historical data inadequate for future predictions
- Novel situations: The law frequently adapts to unprecedented scenarios, especially in rapidly changing fields like technology, where past precedents may offer little guidance
- Ethical considerations: Legal decisions often involve complex ethical tradeoffs that data alone cannot resolve, requiring nuanced human judgment to balance competing interests.
Yet algorithmic justice isn’t some distant possibility. It’s already here. Try generating a politically sensitive image in Midjourney, disputing an Uber charge, or returning an item through Target’s app. These platforms make binding decisions through code, sometimes even offer compromises. Lime Electric Scooters automatically shut down in prohibited zones throughout the country. We’re already living under algorithmic governance, whether we recognize it or not.
“The life of the law has not been logic; it has been experience.”13
— Oliver Wendell Holmes Jr., U.S. Supreme Court Justice.
CURRENT APPLICATIONS AND CHALLENGES OF AI IN LAW
Law is facing unprecedented technological focus from the outside. Billions of dollars are being poured into legal technology, with some start-ups like Harvey achieving unicorn status (multi-billion valuation). There are now hundreds of start-ups focusing on e-discovery, predictive analytics, research, semantic comparison tools, novel work streams, enhanced training modules, data flywheels, Clio-type end-to-end systems, etc.
Contract analysis is a highly promising application of AI in law. With specialized training and prompting, AI can excel at advanced tasks including outlier detection and the holy grail of “what’s market?”- type analysis. This area is particularly well-suited for AI due to the following factors:
- Self-contained nature: Contracts primarily rely on their own content, reducing the need for extensive external knowledge.
- Structured format: Contracts often follow predictable templates and structures, making them easier for AI to process.
- Repetitive nature: Many contracts contain repetitive clauses and language patterns, which AI can readily identify and analyze.
- Rules-based logic: Contractual obligations and terms are typically governed by specific rules and conditions, aligning well with AI’s computational, instructional and semantic capabilities.
While AI is accelerating quickly, each application faces unique challenges rooted in AI’s inability to fully grasp true meaning because
of the infinite context of history, ethics, and the uniquely human elements essential to legal decision making.14
Hallucinations (and biases) — inaccurate or nonsensical outputs from AI that occur when the technology recognizes patterns that either
don’t exist or are imperceptible to human observers — also plague these systems, and it is an area of active research. It is likely that this
problem will decrease over time as new methods and mappings are introduced and tested, such as improved retrieval-augmented generation,
better embedding models, novel tuning architectures and larger and more persistent memory and context windows.
REIMAGINING THE LEGAL SYSTEM
“The more laws a society has, the less justice.”15
— Cicero, lawyer and philosopher
Before we reimagine possibilities, I invite readers to sit in a bit of childlike wonder at the immense potential AI systems hold for our profession and not to dwell on or fear change — even substantial change.
This is a paradigm shift, and the question to ask is not just how AI can manage the existing complexities of the legal system, but how it can help us reimagine the legal system itself with a clear focus on enhancing justice, fairness, and human dignity.
Some will say the proliferation of complex and limitless laws, regulations, and policies is an inherent and unavoidable byproduct of our society’s increasing complexity, technological advancement, economic concerns, political processes, and risk-averse nature. In this view, legal frameworks will inevitably multiply and contracts will inexorably expand. Yet simply layering on more rigid rules and structures fails to address (and could exacerbate) the burden that ordinary people experience navigating today’s labyrinthine legal system.16
One significant concern is that addressing our legal system’s complexities by layering on additional AI-driven systems risks creating an ungovernable tangle of interconnected technologies. Rather than simplifying law’s labyrinth, we might merely digitize its dysfunction. In the worst-case scenario, AI becomes a tool for entrenching existing power dynamics, automating bias at scale and amplifying inequities under the guise of algorithmic objectivity.
However, AI also presents us with an opportunity to reassess our legal system’s efficacy. Instead of navigating the current maze of rules, we can leverage AI to discern what truly serves justice and societal needs. This could involve systematically analyzing case outcomes, business results, and the broader consequences of our laws, allowing us to refine and simplify the frameworks that govern us.
The technology also promises to inject dynamism into what are currently static legal instruments. Consider contracts that self-amend based on real-world conditions — adjusting payment terms when supply chains falter, modifying delivery obligations during natural disasters, or automatically triggering parametric insurance payouts when predefined weather events occur. These adaptive frameworks could eliminate countless technical defaults that arise not from bad faith, but from rigid contracts colliding with fluid realities.
AI could revolutionize our understanding of how contract terms evolve in response to economic conditions. For instance, do terms tighten during recessions and loosen during recovery periods? Could we simulate more flexible contract structures to adapt to these shifts? Such insights might lead us to rethink how we structure legal agreements, moving beyond traditional rigid formats.
The ad-hoc nature of law and jurisprudence itself can be refashioned with enough consensus. Why can’t ordinary people read and understand a simple contract or know what is covered by their insurance? And yet law today remains, in some corners, outrageously complex and nearly indecipherable (by fellow lawyers too!).
In today’s practice, contract drafting is often seen as more art than science. Yet with AI, we can apply theories from architecture and computer science — like pattern theory — to identify key terms, boilerplate provisions, and the sections that generate the most conflict or negotiation. In contracts spanning hundreds of pages, how much is truly operative? Can AI help us streamline our approach to contract creation and interpretation, focusing on what matters most?
Beyond contracts, AI could facilitate ambitious comparative studies of justice systems, contrasting punitive approaches with restorative models across jurisdictions and time periods. It could also serve as a kind of devil’s advocate, surfacing unpopular or unconsidered perspectives that human advocates might overlook due to bias, convention, or institutional blind spots. I imagine this leading to more nuanced, technologically informed decision-making that avoids the zero-sum games and rigid formalism that so often produce unjust outcomes. Consider the cautionary tale of White & Case’s technical win for Disney Corporation: by enforcing arbitration clauses in Disney+’s terms of service during a wrongful death lawsuit, the firm achieved a narrow legal victory at the cost of severe reputational harm.17 The rule-bound argument was legally correct yet revealed how mechanical application of contract law can generate morally tone-deaf, even self-defeating, results.
An AI system trained to recognize these dynamics could flag such risks, warning counsel that winning a motion might mean losing public trust, client loyalty, or the broader cause of justice. Rather than being paternalistic in dictating outcomes, such a system could feel almost maternal: nudging lawyers toward more balanced, contextual strategies that acknowledge human consequences alongside legal correctness.
Ultimately, AI could help us build simpler, more principled frameworks with clearly articulated factors rooted not in cold abstraction, but in warm lived human experience. It could move the law beyond mere technical victories toward outcomes that are sustainable, just, and truly in service of those it claims to protect.
CONCLUSION
The future will be very different.
The call to action here is to see technology as both the biggest challenge and the biggest opportunity. To see both the potential and the pitfalls.
The transformation of the legal landscape is inevitable. AI’s potential to enhance efficiency, accuracy, and access to justice is undeniable. Yet we should ensure that the pursuit of technological advancement does not overshadow the fundamental principles of law — fairness, empathy, and the recognition of human dignity.
The legal profession has always been an evolving field, responding to societal changes and technological advancements. The future of law is where human judgment and AI capabilities intertwine, creating a legal system that is both more efficient and more just. This technology is not a replacement for human intellect; it is a tool to augment our capabilities, enabling us to navigate the complexities of the modern world with wisdom, compassion, and an unwavering commitment to justice — at a scale and speed previously impossible.
That’s why I am truly optimistic about the future of law, because optimism in this space ultimately rests on optimism about human potential itself. If we trust our capacity for wisdom, compassion, and principled compromise, we can shape technology to amplify these qualities rather than replace them.
For lawyers who truly love the law, this technological revolution isn’t something to fear—it’s something to embrace.