In the legal tech space, where precision is critical, the choice of the best AI model for contract intelligence is more than academic—it's essential. 

With the 2023 OpenAI rollercoaster and looming annual budget reviews, law firms are at a crucial juncture: deciding whether to adopt generative language models (LLMs) with their expansive abilities, or extractive AI models known for their sharp accuracy. This decision lies between embracing generative models' wide-ranging potential and opting for the narrower but more reliable power of extractive AI.

Regardless of which model may be best suited, the emergence of AI in the legal space seems inevitable. A recent Thompson Hine survey, as reported by the Thomson Reuters Institute, reveals that 82 percent of attorneys now view generative AI as a practical tool for legal tasks, signaling a notable shift towards tech adoption in a typically cautious sector. 

It also found that 96 percent of law department leaders and legal operations staff are not just open to but are actively seeking to integrate generative AI, seeing its potential to boost efficiency and cost-effectiveness in legal services.

Embracing generative AI doesn’t mean accepting the technology for all it is, however. The critical question for me is: with generative AI presenting a clear lack of factuality, are we putting too much hope on this model, which cannot deliver the reliable results needed in legal work?

Generative LLMs, celebrated for their wide-ranging comprehension and ability to process natural language, are front runners for most firms. But this adaptability carries inherent, fairly damaging risks – a likelihood of ambiguity and inaccuracy. 

Teams reviewing legal documents, which demand exactness and precision, can’t afford the potential mistakes and biases intrinsic to generative models.

An example of what can go wrong was reported in a Forbes article, detailing an incident in the US where a lawyer used ChatGPT in court and unintentionally referenced non-existent cases. This shows the potential legal pitfalls of depending on generative AI without proper verification.

In contrast, extractive AI models, a form of Language Models, are crafted for precision and reliability. They don't just scan text, they extract specific, core information, and always provide a reference to the content that’s been used.

Because the answers are extracted, not generated, there’s no room for inventions or hallucinations - meaning they are the better option for legal use cases - for instance, where an exact understanding of a legal clause is needed.

Finally, scalability and adaptability to changing regulatory landscapes are interesting aspects of extractive models. By maintaining focus on specific information, these models can be updated more efficiently to comply with new regulations - which will come in handy for most legal teams, and which generative AI LLMs can’t yet do. 

Related to this, extractive AI models have an edge in ethical compliance, focusing on analyzing and pointing to existing information, thus reducing the ethical concerns often associated with generative models. 

These models, as they tend to be smaller, are generally easy to train, as they require less data and therefore are easier to adapt. This, however, comes with less flexibility than what LLMs offer.

Choosing between generative and extractive AI models is not just about keeping pace with industry trends. It's a decision that affects the heart of legal operations – the precise interpretation and management of contracts. 

Contract intelligence in particular, which refers to the process of using AI, natural language processing, and machine learning algorithms to automatically identify, extract, and analyze relevant information from legal contracts, has become a huge help for organizations to streamline various aspects of contract management including compliance monitoring, risk assessment, and performance tracking. 

It’s also particularly hard to decide whether to go for generative or extractive AI models - or a mix of both. In this context, legal teams must weigh up:

Individual firm needs that assess if the broad capabilities of generative AI align with the firm's operational requirements, or if the specific precision of extractive AI is more relevant.

Hybrid approaches where many firms may benefit from combining the strengths of both AI models, catering to different needs within their legal practice.

The fact that technology should enhance, not replace, the expertise of legal professionals and that integrating AI into legal practices calls for a strategy that respects and leverages the skills of the legal team.

Advancements in extractive AI, particularly in natural language processing and comprehension, are going to significantly enhance its application in contract intelligence.

If legal teams needed any more reasons to capitalize on extractive AI, here they are: we're witnessing a shift towards a more multi-modal approach, where extractive models will interpret not only text but also diagrams and images within contracts. 

This evolution will make these models more versatile and invaluable, potentially setting the stage for extractive AI to surpass the capabilities of generative AI in certain legal contexts.

To make the most of extractive AI now and in the future, the focus of legal teams should be on integrating this technology effectively into their workflows. Challenges in language processing and system integration will be key areas to address, with the main goal being to derive actionable insights from the extracted information and integrate it seamlessly into existing systems.

While the allure of generative AI is clear for the creation or exploration-based tasks, the reliability of extractive AI models aligns more closely with the critical needs of all legal work requiring an analysis, categorization, or standardization of contracts.

As more of this technology infiltrates legal operations, teams must prioritize what matters most in legal practice: accuracy, compliance, and fairness.