News
Comment
The development of artificial intelligence is growing substantially. ChatGPT by OpenAI, followed by LLaMA from Meta and then Bard from Google, demonstrate the significant advances in general-purpose AI and its foundational models. However, other specialized, more limited AI systems are emerging. This is the case of Vincent AI, an artificial intelligence assistant capable of answering questions, constructing legal arguments, and comparing legal issues across Spanish, American, Irish, and UK jurisdictions. The technology is currently in beta phase and has been tested by approximately 15 major law firms in these countries.
The emergence of specialized systems such as Vincent AI responds to a practical observation: general-purpose language models, no matter how powerful, exhibit significant limitations when applied to domains requiring strict standards of accuracy, regulatory up-to-dateness, and legal accountability. In the legal field, an incorrect response or one based on outdated sources can have serious professional and legal consequences. For this reason, the approach of vLex, the company behind Vincent AI, emphasizes aspects that, as we have seen in previous articles, are not always prioritized in general-purpose systems. Vincent AI distinguishes itself from other artificial intelligences by focusing on security, striving to avoid biases, enabling multi-jurisdictional capability, promoting transparency, and granting users control over their actions. Although still in its early stages, it is expected to improve over time through user feedback.
These characteristics are not incidental but define the very architecture of the system. While in models such as GPT-4, response generation operates as a black box where it is difficult to trace the origin of each assertion, Vincent AI is built upon a structured legal knowledge base — vLex’s own document repository — that enables each argument to be linked to the normative, jurisprudential, or doctrinal sources supporting it. This approach, which we may term "source-based AI," aligns with the fundamental principles of Documentation Sciences: provenance, authenticity, and traceability of information.
The traceability of content and its reasoning, grounded and cited in the sources from its knowledge base. However, security and the accuracy of responses take precedence over speed. As indicated, the response time is 1 to 2 minutes. This latter figure is revealing. In a context where general-purpose models compete to deliver answers within seconds, Vincent AI deliberately accepts higher latency in exchange for verification and precision. This is not a technical limitation, but a deliberate design decision: the priority is that professionals can trust the outcome, not that they obtain it immediately. From the perspective of Information Retrieval and Software Development, Vincent AI represents a complementary direction to that of large general-purpose models. While the latter prioritize breadth of knowledge and conversational fluency, domain-specific systems—also known as "Narrow" Artificial Intelligences—can offer depth, source control, and adaptation to the regulatory particularities of each jurisdiction. The beta phase of Vincent AI has already been adopted by 15 major law firms, suggesting that vLex is pursuing a strategy of validation prior to a broader rollout.
For now, Vincent AI presents itself as a relevant experiment at the intersection of artificial intelligence and legal disciplines. Its success will depend not only on the technical capability of its model, but on the trust it can generate among a professional community particularly demanding in terms of accuracy, security, and accountability.