Seminario Práctica del Derecho e IA

Name: Seminario Práctica del Derecho e IA
Uploaded: 2025-04-02T09:10:27.000Z
Duration: 3 h 48 min 57 s

Welcome to the Seminar

Introduction and Context

The seminar is hosted by the Pontificia Universidad Católica de Chile, focusing on "Legal Practices and Artificial Intelligence" with case studies from the United States, Mexico, and Chile.

Notable speakers include Lozar de Terman from UC Berkeley and Baker McKenzie LLP, Daniel Villanueva Placencia from Tecnológico de Monterrey, José Ignacio Mercado from Keri, and Matías Aranguis from UC.

Opening Remarks

The speaker expresses gratitude for being invited to Chile and shares positive impressions of the country’s culture and cuisine.

Emphasis is placed on discussing what constitutes artificial intelligence (AI), highlighting its significance in current legislative challenges.

Defining Artificial Intelligence

Understanding AI

The term "artificial intelligence" is widely used but lacks a clear definition; different interpretations complicate regulatory efforts.

A distinction is made between "artificial" (human-made) and "intelligent" (problem-solving without violence), suggesting that not all technology labeled as AI meets these criteria.

Regulatory Challenges

Reference to the EU AI Act illustrates how broad definitions can encompass simple devices like watches or calculators under AI regulations.

The speaker critiques overly inclusive definitions that could classify low-risk tools as high-risk AI systems based on their use in consequential decisions.

Comparative Analysis of Regulations

Critique of Existing Laws

Colorado's approach attempts to refine definitions by excluding basic calculators unless they are used for significant decisions affecting labor or education.

The speaker questions the logic behind classifying everyday tools as high-risk AI systems when they do not exhibit true autonomy or adaptability.

Future Directions

Legislative Perspectives on Artificial Intelligence in Mexico

Overview of AI Definitions and Implications

Legislators in Mexico are considering the implications of artificial intelligence (AI), noting similarities with European approaches but also influences from the United States.

A definition discussed in Chile describes AI as resembling human intelligence, raising questions about what constitutes "human-like" characteristics and potential complications.

The anthropomorphization of machines poses ethical dilemmas; for instance, if a machine makes decisions akin to human intellect, should it be considered a form of homicide if it is turned off?

Legal complexities arise when defining AI; future regulations will impose responsibilities and compliance requirements that could affect how technologies are labeled and utilized.

Caution is advised against broadly labeling non-AI technologies (like calculators) as AI, as this could lead to regulatory risks and misinterpretations by legislators.

Narrowing the Definition of AI

It is suggested that countries without clear legal definitions should adopt narrower definitions of AI to avoid confusion and misapplication in legislation.

A distinction is proposed between deterministic systems (not classified as AI) and true AI systems that operate autonomously, solving problems without direct human input.

True AI creates significant legal challenges regarding liability, ownership of intellectual property from outputs, and protection of trade secrets related to inputs.

Understanding Traditional vs. Generative Programming

The speaker reflects on their language learning journey while emphasizing the importance of understanding generative versus deterministic programming models in software development.

High-level autonomy in AI refers to systems capable of generating text or images independently, which developers cannot predict or control reliably.

Practical Applications: Self-driving Cars

Traditional programming involves straightforward instructions ("if this then that"), contrasting with the complex decision-making required for self-driving cars.

Self-driving technology necessitates teaching vehicles through experiential learning rather than rigid programming due to the multitude of real-time decisions they must make on the road.

Understanding AI Training and Recognition Systems

The Evolution of AI Training Methods

Human drivers are trained to take control of vehicles, illustrating the difference between high and low autonomy in AI systems.

Traditional predictive AI struggles with complex image recognition tasks, such as identifying objects in CAPTCHA tests, which are designed to be challenging for machines.

Early software developers opted for a child-like learning approach by feeding machines numerous images instead of programming explicit definitions for objects like fire hydrants.

Machine Learning Techniques

Machines learn to recognize patterns through feedback loops; when they misidentify an object, their internal weights adjust based on human correction.

This method allows machines to identify various objects accurately, including faces from significant distances, showcasing advancements in recognition technology.

Generative AI and Its Implications

Generative AI operates on probabilistic methods, predicting the next word in a sentence without understanding context or content deeply.

The surprising outputs generated by these systems raise questions about liability and ownership—who is responsible if an autonomous system fails?

Regulatory Challenges and Definitions

Current regulations may encompass all software under broad definitions, potentially stifling innovation while failing to address genuinely problematic systems effectively.

A proposed narrow definition aims to focus on the most relevant 2%–3% of AI technologies that require new laws or assessments rather than applying existing laws universally.

Future Considerations in AI Development

There is a call for practical discussions around specific issues related to generative AI output copyrights and regulatory frameworks.

As explainability improves over time, future systems may not be classified as "AI" if they can be fully understood and controlled by humans.

Understanding AI and Its Implications

The Future of AI: Determinism and Focus

The speaker suggests that in about ten years, the understanding of intelligence and magic in AI may become deterministic, allowing for precise control over its functionalities.

Emphasizes the importance of focusing on technologies that pose real risks rather than all technology indiscriminately, highlighting autonomy levels as a critical factor.

Defining Intelligence in Machines

Raises concerns about defining AI based solely on human standards of intelligence, questioning how to classify machines that communicate in their own languages.

Discusses the challenge of establishing boundaries for what constitutes intelligence from both human and machine perspectives.

Intellectual Property Concerns

Highlights the complexities surrounding intellectual property rights related to AI-generated content, particularly regarding moral rights and artistic styles.

Warns against attributing human characteristics to machines, suggesting it complicates legal definitions and ethical considerations.

Autonomy vs. Human Control

Proposes measuring autonomy by assessing how much control humans have over machine outputs; if users can influence results, regulation may not be necessary.

Advocates for simpler terms in legal discussions around AI to facilitate understanding among lawyers and policymakers.

Ethical Considerations in Defining AI

Questions whether including a human intelligence component in definitions could lead to different risks when considering AI's capabilities.

Critiques Alan Turing's 1950 proposal for defining AI through comparison with human intelligence as potentially misleading due to its focus on deception.

Discusses whether generative AIs should be granted copyright protection; argues they lack true creative capacity akin to humans.

Discussion on AI Image Generation and Legal Implications

Concerns About Energy Consumption and Legal Rights

The recent public release of image generation technology raises concerns about energy consumption and the potential dangers associated with its use.

Questions arise regarding the legality of using images for training AI models, including issues of royalties and rights, which could make it financially unfeasible for businesses to comply.

Ongoing Legal Discussions in the U.S.

Current legal cases in the United States are addressing these issues, but resolutions have yet to be reached.

The conversation shifts towards automated decision-making, highlighting its growing relevance in various sectors.

Automated Decision-Making: Historical Context and Current Legislation

Legislative Background

France has included a right against solely machine-based automated decision-making in their data protection laws since 1976.

The European Court of Justice recognized credit scoring as a form of automated decision-making only recently, indicating slow progress in regulatory frameworks.

Risks Associated with Automated Decisions

Credit scores can lead to automatic loan denials based on statistical patterns that may reflect systemic discrimination rather than individual merit.

If machines are trained on biased historical data, they may perpetuate existing inequalities by denying loans based on race or gender.

Implications for Employment and Education

Discrimination Risks in Hiring Practices

Automated systems used for filtering job applications can inadvertently reinforce biases if not carefully managed.

A deterministic filter (e.g., GPA thresholds) is acceptable; however, fully automating resume evaluations risks excluding qualified candidates due to past discrimination patterns.

Comparison Between U.S. and European Regulations

The U.S. has historically regulated credit scoring under the Fair Credit Reporting Act since 1970, focusing on preventing bias without broad data protection laws like those in Europe.

Understanding Automated Decision-Making in Europe

The Regulatory Landscape of Automation

In Europe, there is a tendency to prohibit automation in certain contexts, such as ski lifts, where automated systems may deny access based on outdated information. This raises questions about the definition of automated decision-making.

Broad definitions of automated decision-making could overlook significant issues like credit scoring and resume scanning, which are currently problematic for clients seeking legal advice.

Confusion Among Companies Regarding AI Usage

A study revealed confusion among companies regarding their use of artificial intelligence (AI) for recruitment; while 15% of lawyers reported using AI, 85% from HR departments claimed otherwise.

This disconnect highlights a lack of understanding about what constitutes AI and how these systems function, posing risks for society if not addressed properly.

Legislative Developments in Data Protection

The discussion emphasizes the need for clear definitions around AI and automated decision-making to prevent future issues. Current regulations on personal data protection already address some aspects of automated decisions.

Recent legislation in Mexico introduced new conditions regarding automated decisions related to personal data processing, indicating that this issue is already present globally.

Challenges Faced by Companies

New regulations require companies to implement mechanisms allowing individuals to withdraw consent for their data being processed through automated decisions, leading to potential operational challenges.

Compliance with these regulations will demand significant investment and effort from companies as they navigate the complexities involved.

The Risks Associated with Misunderstanding AI

Case Study: Legal Missteps Due to AI Misunderstanding

An anecdote illustrates how an older lawyer misused ChatGPT to draft court motions without understanding its limitations; this resulted in fabricated case citations leading to penalties.

The incident underscores the importance of comprehending how generative models operate—highlighting that they work probabilistically rather than deterministically.

Implications for Legal Practice

Lawyers must educate themselves and their clients about the functioning of AI systems to mitigate risks associated with errors stemming from misunderstandings.

Understanding Chatbots and Their Limitations

The Nature of Chatbot Outputs

Chatbots are designed to generate draft text using probabilistic methods, not as definitive sources of truth. Disclaimers emphasize this limitation.

Practicing lawyers often use chatbots similarly to how they would ask a secretary for a draft motion, understanding that the output is merely a starting point for further refinement.

User Responsibility in Verification

If a chatbot provides incorrect information, it is not considered a defect; users must verify the accuracy of the information provided.

Complaints about inaccuracies, such as personal data errors highlighted by Max Schrems, illustrate misunderstandings about chatbot functions. They produce drafts rather than factual data.

Importance of Terminology and Training

Clear communication regarding what chatbots can do is essential. Proper terminology in contracts and training for colleagues can mitigate misunderstandings about their capabilities.

The Impact of AI on Legal Practice

Personal Experience with AI

A lawyer reflects on their journey with AI, noting that they were not exposed to AI concepts during their legal education in the late 90s and early 2000s.

The speaker describes themselves as an attorney impacted by client inquiries related to AI without being an expert in the field.

Historical Context of Technology in Law

The speaker recalls limited exposure to technology during law school, highlighting basic computer skills training but no significant focus on AI or its implications.

Evolution of Legal Framework Post-2020

Since 2020, regulatory changes have complicated previously straightforward legal practices. New laws like fintech regulations and data protection laws have emerged.

Discussion on Data Protection and AI Regulation

The Evolution of Data Protection Laws in Latin America

Prior to the Personal Data Protection Law, there was an existing law regulating artificial intelligence systems, marking Latin America's pioneering efforts in this area.

New authorities like ANI (National Agency for Personal Data Protection) and CMF (Financial Market Commission) emerged, but there is concern over adopting European regulations without adapting them to local realities.

Challenges with Current Regulations

There is a growing disconnect between legal frameworks and practical applications, leading to confusion among lawyers dealing with AI-related inquiries.

Over the past 20 years, questions have shifted from straightforward legal queries to complex issues that are difficult to navigate due to evolving technologies.

Legal Paradoxes and Uncertainties

Existing regulations can address some challenges posed by disruptive technologies; however, new proposals may create further disorder and uncertainty in legal interpretations.

Lawyers face specific questions regarding personal rights related to image use and contracts involving personal attributes for AI applications.

Real-world Implications of AI Training Regulations

Concerns arise about how sensitive personal data is handled during AI training processes under current data protection laws.

The intersection of intellectual property rights with innovation raises critical questions about fair use and knowledge circulation within regulatory frameworks.

Regulatory Conflicts Affecting Innovation

Recent regulations often conflict with established practices, complicating compliance for businesses developing facial recognition systems integrated with AI.

A case study illustrates the difficulties faced when trying to align innovative projects with stringent legal requirements that may not be well-suited for local contexts.

Seeking Balance Between Regulation and Innovation

The key question shifts from whether or not to regulate towards how to create a coherent system that provides legal certainty while fostering innovation.

Regulation and Innovation in Technology

The Centrality of the Human Person

Emphasizes the importance of placing the human person at the center of regulatory frameworks while promoting innovation.

Advocates for clear boundaries on regulation, suggesting that existing legal frameworks should be utilized to resolve issues rather than over-regulating.

Critique of Legislative Initiatives

Criticizes a legislator's claim about being first to propose an AI-related law, arguing that being first is irrelevant if the implementation is flawed.

Warns that poorly implemented systems can create significant barriers and increase transactional costs within legal processes.

Evolution of Regulatory Frameworks

Provides a general analysis of how technology regulation has evolved, highlighting current challenges as a "disaster."

Discusses traditional views on law as composed of rules and authorizations, which assumed rational behavior from individuals responding to incentives.

Shift in Regulatory Logic Post-Crisis

Notes that prior to the 2008 financial crisis, regulations were based on assumptions of rationality among individuals.

After 2008, there was a shift towards risk-based regulation acknowledging limited rationality and consumer protection needs.

Challenges with Current Reporting Requirements

Highlights excessive reporting requirements under European AI laws as burdensome for compliance, creating barriers for smaller entities.

Argues that mandatory reporting does not guarantee rational behavior but instead complicates compliance without addressing all risks effectively.

Principles-Based Regulation Issues

Describes the current state as one where regulations are based on abstract principles rather than concrete mandates.

Critiques Chile's fintech regulation for listing principles without providing substantive content or guidance for enforcement.

Regulatory Challenges in Data Protection and AI

Objective Responsibility in Data Regulation

The regulation indicates that companies can be sanctioned even without demonstrable harm, emphasizing objective responsibility for actions affecting data principles.

There is no need to establish a direct link between damage and the actor; the mere act of risk poses a challenge in determining what constitutes sanctionable behavior.

Legal Framework and Predictability Issues

The legal framework previously allowed clear definitions of compliance, but with risk-based regulations, it has become more complex and less predictable.

Risk-based regulations introduce ambiguity as they do not always incorporate specific types of conduct, leading to challenges in enforcement.

Reporting Requirements and Regulatory Efficacy

Many incidents fall outside reporting requirements due to unpredictability, complicating the ability of regulators to enforce sanctions effectively.

Current regulatory frameworks may lack clarity on which behaviors are deemed compliant or non-compliant under risk-based assessments.

Automation and Decision-Making Regulations

A notable case from Germany highlights that decisions based solely on automated processes are generally prohibited unless explicitly allowed by law.

Unlike EU regulations that default to prohibiting automated decision-making, some jurisdictions allow it with certain rights preserved for individuals.

Principles vs. Rules in Regulation

The challenge lies in defining what constitutes responsible conduct within principle-based regulations compared to rule-based systems.

Principle-based regulation often results in higher costs without improving market safety or efficiency, raising questions about its effectiveness post-financial crises.

Political Implications of Regulation

Regulatory measures often serve political interests rather than achieving their intended goals, leading to increased costs without enhanced protection.

Transferability of Regulatory Principles

Questions arise regarding whether principles established by entities like the EU can be effectively transferred across different jurisdictions.

Ambiguity in AI Regulations

Discussion on AI Regulation and Ethical Concerns

The Challenge of AI Regulation

The speaker discusses the complexity of applying regulation to artificial intelligence (AI), highlighting that it is often seen as a catch-all solution for various societal issues.

There is a notion of "politics of indignation," where individuals advocate for regulations based on emotional responses to perceived injustices, even if they are not directly affected by those issues.

A conversation with a sociology professor raises the question of whether specific laws for AI should exist, particularly regarding environmental protection, which leads to debates about existing regulations versus new ones.

Misattribution of Blame

The speaker notes that AI is often blamed for numerous problems in society, suggesting it serves as a scapegoat for broader technological issues rather than being the root cause itself.

The concept of "responsible" or "ethical" AI emerges as a vague term lacking legal grounding, complicating accountability in judicial contexts.

Standards vs. Principles in Regulation

There is an ongoing discussion about how standards (like ISO or NIST) provide concrete guidelines but may not ensure complete safety from harm when applied in practice.

The metaphorical reference to technology's blame game illustrates how any issue can be attributed to AI without considering other technologies that might share similar flaws.

Hope Through Standardization

The speaker reflects on the hope brought by standardization discussions, comparing it to having multiple ports on computers and seeking one universal standard.

Technical standards are essential but must be balanced against principles; while standards offer certainty in compliance, they do not guarantee protection from harm.

Confusion Among Standards and Principles

A significant challenge arises from the multitude of available standards and principles related to AI; stakeholders must choose which ones hold authority or relevance based on power dynamics within organizations.

The Intersection of Chaos and Order in Legal Interactions

The Nature of Interactions

The speaker discusses the fascinating interplay between rational and artistic worlds, suggesting that from chaos emerges order through increased interactions.

Emphasizes the importance of legal interactions, particularly how disputes are resolved by courts, highlighting a trend in the U.S. where many cases settle to avoid judicial rulings.

Notes that settling disputes is often cheaper than court judgments, which set standards for behavior and accountability.

Regulation and Standards

Critiques European regulation as primarily risk-based, arguing it has historically failed to provide effective solutions for artificial intelligence (AI).

Suggests that unifying interpretative criteria will only occur through judicial decisions that clarify standards and principles.

Questions on Legal Systems and Definitions

Impact of Different Legal Systems

A question arises regarding the implications of differing legal systems (common law vs. civil law) on definitions within transactional law.

Discusses how definitions can vary significantly across jurisdictions like Chile, Mexico, Europe, and the U.S., affecting certainty in legal interpretations.

Regulatory Approaches

In Chile, regulatory agencies may define terms more strictly compared to courts in common law systems where definitions can be situational.

Raises concerns about how these varying approaches impact legal frameworks globally amidst evolving technology regulations.

The Role of Definitions in Law

Importance of Clear Definitions

Highlights that unclear definitions can lead to poor transactional documents and ineffective laws; clarity is crucial when discussing AI.

Asserts there is little difference between common law and civil law today regarding technology regulation due to extensive legislative detail.

Transactional Implications

Stresses that vague definitions complicate compliance representations during transactions; companies struggle with broad claims about AI usage.

Understanding the Challenges of AI Regulation

The Complexity of Disclosure in M&A Transactions

A plaintiff lawyer may compel the production of confidential documents, which raises concerns for sellers. It is advised to avoid broad representations that could expose issues with a startup company.

Legislative Definitions and Their Impact

Narrow definitions in legislation are crucial; overly broad regulations can stifle innovation and suffocate companies, as seen in EU practices.

The Need for Balanced AI Regulation

California's governor vetoed a broad AI law due to its vagueness, emphasizing the need for careful regulation to protect the state's leading AI companies.

International Perspectives on Regulation

Countries must consider their regulatory approaches carefully; Germany has lost tech leadership due to excessive regulation, while France acknowledges the risk of regulating industries no longer present domestically.

Shifting Attitudes Towards Regulation

For the first time, major tech executives express a desire for regulation due to concerns over autonomous AI systems. They advocate for smart regulations that do not hinder innovation.

Redundant Regulations and Their Confusion

Many new laws prohibit actions already illegal without specifying their relation to AI, potentially confusing citizens about what is permissible.

The Role of Different Government Branches in Regulation

A balance of power among branches is essential; Congress creates laws while executive orders can address urgent needs when legislative processes lag behind.

Future Outlook on Legal Practice and Regulation

The Future of Regulation and AI in Law

The Trajectory of Regulation

Discussion on the increasing complexity of regulations, with politicians focused on creating new laws rather than repealing outdated ones.

Mention of Elon Musk's initiative to reduce regulation in the U.S., highlighting his focus on space exploration and how regulations hinder progress.

Musk's involvement in politics is driven by regulatory challenges faced during rocket launches, leading him to acquire Twitter for better communication.

Political Dynamics and Public Sentiment

Observations that many Americans across political lines agree that excessive regulation is counterproductive for businesses.

Acknowledgment of the difficulty in predicting future political trends but a belief that similar reactions to regulation may emerge globally.

The Role of AI in Legal Practice

Emphasis on the potential for artificial intelligence (AI) and big data processing to help manage overwhelming legal regulations.

Personal experience shared about the challenge of keeping up with rapidly changing laws across jurisdictions, particularly in Europe and California.

Enhancing Judicial Efficiency with AI

Suggestion that AI could assist judges in making quicker decisions, reducing backlog issues within the U.S. asylum system.

Concerns raised about automated decision-making; advocates for using AI as a tool for judges rather than replacing human judgment.

Balancing Human Judgment and Technology

Critique of perceptions surrounding AI use in justice as "high risk," arguing instead for its integration into judicial processes.

Call for collaboration between humans and AI, leveraging technology while maintaining oversight from legal professionals.

Case Study: Compass Software

Inquiry into the implementation of AI tools like Compass software within the justice system, which assesses criminal recidivism risk.

Discussion on AI and Racial Bias in Crime Prediction

Historical Context of Crime Statistics

The speaker discusses how historical data, particularly regarding Americans who were descendants of enslaved individuals, influences crime statistics. This raises concerns about the representation of these groups in crime predictions.

Ethical Implications of AI Predictions

There is a belief that while statistical predictions based on race may be accurate, using such data for individual assessments is problematic due to societal decisions against racial discrimination.

Societal Decisions Against Discrimination

The speaker emphasizes America's collective decision to move away from racial discrimination, acknowledging the historical injustices faced by marginalized communities.

Challenges in AI Training

Concerns are raised about how AI systems can avoid perpetuating historical biases. It is suggested that careful examination and testing are necessary before relying on these tools.

Safeguards Against Bias in AI Systems

The discussion highlights the need for safeguards when training AI systems to ensure they do not consider race or other discriminatory factors. Alternative indicators must be identified to prevent bias.

Legislative Measures in New York City

New York City has implemented laws requiring companies using Automated Employment Decision Tools (AEDT) to conduct third-party bias audits, ensuring compliance with anti-discrimination standards.

Learning from Experience with AI Tools

Over time, it is expected that experiences with these tools will lead to improvements, potentially making machines less biased than humans who inherently possess biases.

Comparison with Pseudoscience

The speaker warns against drawing parallels between current predictive models and discredited pseudosciences like phrenology, emphasizing the dangers of basing decisions solely on empirical criteria without rigorous validation.

Importance of Auditing Algorithms

A case study reveals issues surrounding an algorithm used for predicting criminal behavior which was poorly audited and heavily reliant on race-related questions.

Educational Initiatives

An invitation is extended for participation in a diploma program focused on artificial intelligence and law, highlighting ongoing discussions among experts about these critical issues.