AI: COMMERCIAL CONSIDERATIONS AND THE LAW
By Miguel Vasquez & Aniko Sookoo
To ensure that T&T properly leverages the use of AI, it is essential to understand: (a) What is AI and generative AI; (b) Practical applications and potential uses as efficiency drivers; (c) Risks and shortcomings in order to reframe appropriate regulations and policies; and (d) How other countries have regulated or intend to regulate the use of AI, while refraining from stifling the opportunities it creates.
AI systems are computer-driven tools that process inputs which include data, text or images to generate outputs, such as content, predictions or recommendations, with varying levels of autonomy. GenAI is a subset of AI which builds on that fundamental input-output concept, but creates entirely new content, drawing from learned patterns in large datasets, which include text, images, audio and video. Unlike traditional AI, which is deterministic and relies on past data, generative AI or GenAI is non-deterministic and responds to user prompts with the creation of original outputs.
Prominent examples of GenAI include OpenAI’s ChatGPT and Google’s Gemini, which use large language models to generate coherent, contextually relevant content, enabling human-like responses. These technological advancements have enabled the integration of GenAI into various everyday tasks such as writing, drafting and reviewing documents, coding, design and research and development.
GenAI has the potential to significantly boost productivity at both the individual and organisational levels. At an individual level, AI tools can be incorporated into daily workflows, facilitating automation, fostering augmentation and enhancing operational efficiencies. Further organisations with strong absorptive capacity can benefit from an even greater transformation and simplification of operations. For example, Tesla uses AI to forecast commodity price fluctuations while in retail, Amazon’s AI-driven inventory systems dynamically adjust stock levels based on consumer demand.
Hallucinations
Despite its transformative potential, generative AI has its drawbacks and overreliance on its outputs highlight the need for caution, given the risk of inaccuracies. Indeed, while LLM outputs can be perceived as well-written and tailored to specific contexts, these systems may have limited understanding or appreciation for what is generated. The outputs generated are based on statistical patterns and algorithms, which may lead to inaccuracies or fabricated information; a phenomenon known as “hallucination”.
The risk that “hallucinations” pose has been exposed in one of the gravest of ways, namely the courtroom. In the U.S. case of Mata v Avianca, Inc., Case No. 1:2022cv01461 (S.D.N.Y. 2023), attorneys relied on ChatGPT to draft submissions, citing and quoting legal authorities that did not exist, but rather had been “hallucinated”. This issue is not confined to foreign jurisdictions, but has also been exposed in T&T. In Nexgen Pathology Services Limited v Darceuil Duncan Claim No. CV2023-04039, attorneys similarly cited non-existent cases, thus misleading the court and later attributing the error to a junior assistant. The judge scolded the attorneys and in doing so, cited the principle fundamental error; that the attorneys committed a breach of their professional duty, which was realised through the submission of fictitious and unverifiable authorities, by their failure to conduct proper and independent verification of the output.
With responsible use on the other hand, the potential benefits are significant: increased efficiency, reduced errors, lower operational costs and heightened innovation. The challenge therefore lies in striking a careful balance between leveraging the transformative power of GenAI while managing its risks and limitations.
Regulatory Framework
The regulatory framework for AI remains nascent in T&T. There are no specific laws governing AI, nor a formal national AI strategy. As such, an imperfect application and interpretation of existing laws would be relied on for the purpose of governing AI and its use. These include the Electronic Transactions Act, Data Protection Act, consumer protection and cybersecurity provisions. The lapsed Cybercrime Bill, 2015 would, in any event, be inept at regulating GenAI, a technology that has emerged for public use only 3 years ago.
As the integration of AI into systems and workflows accelerate, be it in financial services, energy, healthcare or even within the government, there will be increased pressure to enact specific rules to mandate compliance with international recommended standards such as the OECD AI Principles and UNESCO's Recommendation on the Ethics of Artificial Intelligence.
Preliminary steps have been taken to develop AI-related policies and oversight. The Ministry of Public Administration and Artificial Intelligence has been established to spearhead AI initiatives and guide national efforts. T&T is also part of the Caribbean Telecommunications Union, which has launched the CTU Caribbean AI Task Force, a strategic initiative designed to coordinate the development and governance of AI across the Caribbean.
The Caribbean Court of Justice has also taken steps to regulate the use of GenAI in legal practice by issuing practice directions providing guidance on the permissible use of GenAI tools. The practice direction seeks to strike a careful balance by allowing its use to draft submissions, summarise legal arguments or conduct basic research, but all outputs must be thoroughly fact-checked, reviewed for accuracy and adapted to meet legal standards. The practice direction further imposes safeguards on confidential and privileged information to prevent unintentional disclosure.
Indeed, disclosure of confidential, privileged or sensitive information, remains a significant inherent risk in the use of AI and GenAI, given its reliance on data (whether inputs or outputs). The European Union has purported to set global benchmarks for consent, data minimisation and cross-border transfers through the existing General Data Protection Regulation, which the EU has reasoned are relevant to AI even though not specifically mentioned—albeit with the intention for the EU AI Act to bridge any gaps. T&T’s Data Protection Act, though not fully proclaimed, reflects many of the general privacy principles but is likely to require significant amendment to properly encompass the regulation of AI, consent and data protection.
Consent
Consent for users of AI tools, from the perspective of the user and if relevant, the owner of the data inputs, is also a critical consideration. Consent obtained for the purposes of a particular use or transaction, may not extend to a separate use or transaction. Accordingly, robust contractual safeguards and risk assessments should be undertaken to ensure parties are optimally protected, both in terms of the inputs and the outputs.
This is critical not just from the perspective of data protection, but also liability for harm caused through the use of GenAI. Whether that be in the form of a misdiagnosis in healthcare or an autonomous vehicle accident. Traditional tortious causes of action such as negligence and product liability will be stretched to encompass these losses, in the absence of specific legislation. Insurance markets will be required to respond, with the creation of AI liability and data protection policies and products, as well as cyber-risk coverage.
The ownership and control of the AI, GenAI and output, is likely to also be tested. In the case of training an AI diagnostic tool, the question will arise as to who owns the resulting model and the derivative insights produced. Indeed, when an AI system generates a novel design, piece of music or chemical formula, the natural question that arises is, who owns that work of art or product. Copyright laws, including T&T’s Copyright Act, presupposes a natural person or human as the author. Jurisdictions such as the United Kingdom provide for “computer-generated works”, but many doubts remain as to its use and functionality. Patent law presents similar dilemmas – the UK Supreme Court dismissed an appeal to allow an AI, DABUS, to be named as a patent inventor. Businesses must therefore adopt clear contractual terms to allocate IP rights over training data, outputs and derivative works. Trade-secret protection and technical safeguards, such as encrypted model weights and restricted access, are becoming equally important.
Beyond the law, however, lies the question of values. Transparency and accountability are key to public trust. Individuals negatively affected by an algorithmic decision should understand how and why the decision was made; this “human oversight” or “meaningful human control”, to prevent over-automation.
AI and GenAI create risks and issues across a diverse range of sectors, but its potential as not merely a tool but a catalyst for more intelligent, effective and efficient operations, cannot be underscored. Companies and businesses will augur well to take the first step toward maximising its benefits, particularly in T&T. Measures to safeguard against the risks and ethical concerns of AI should be adopted however, as measured caution must proceed parallel along the path to digital transformation and progression.
ABOUT THE AUTHORS
Miguel Vasquez is a Partner at M. Hamel-Smith & Co. and Aniko Sookoo is an Associate. They can be reached at mhs@trinidadlaw.com. Disclaimer: This Article contains general information on legal topics and does not constitute legal advice.