Prompt engineers – do we need them?

Around 85 per cent of the jobs needed in 2030 have yet to be invented. This was the claim in a 2017 report by the Institute For The Future, a US-based think tank. It's still uncertain whether that prediction will prove correct, but we're certainly seeing an explosion of new roles created by our changing environment and evolving technologies.

Artificial intelligence is one of these technological advancements. Although there is concern that AI will make large swathes of the workforce redundant, it is equally likely to create opportunities and new roles, although how long these roles will be relevant in this fast-moving world is up for debate. 

One of these new roles is that of a prompt engineer - an individual who has a deep understanding of how to harness AI, particularly the Large Language Models (LLM) wrappers like ChatGPT (based on the OpenAI engine). Prompt engineers can redefine the queries posed to an AI system to get optimal results. In these instances, 'optimal results' can refer to getting the best answers and getting these answers efficiently, with minimal back-and-forth prompting. 

However, AI models are rapidly evolving. The prompts we require today are likely to evolve. The models available require structured prompts to create useful, structured output. As the models grow, more unstructured natural language may suffice to generate the same useful output. And in some ways, the need for professionals to attain the skills required to generate effective prompts is not dissimilar to the longstanding need for professionals to understand the basic requirements to create Boolean or conceptual search queries.

Fundamentally, knowing how to search rapidly expanding and more complex data sets will continue to be necessary. Perhaps more interestingly, the real area of need is problem formulation. This skill involves properly diagnosing and defining a problem so the AI model knows what to work on. 

Problem formulation will be key to successfully adopting and using AI for corporate legal teams and law firms. At its core, problem formulation requires a diagnosis of the issue and, by getting to the heart of the problem, we define the key objective. Once this is recognised, breaking the problem into digestible portions for the people, process, and protocols involved is important.

Next, it's about reframing the problem. Thinking about the issue from multiple perspectives and how the problem impacts different stakeholders can be hugely helpful. Finally, including boundaries to a problem can also be helpful when using AI tools to find a solution.

There's no doubt that AI tools, particularly those such as ChatGPT, are an exciting area that creates opportunities for innovation. This technology's most significant short-term opportunities are optimising internal processes and creating simple templates, contract management, administrative automation, and document review. All of these require a full understanding of the issues that need fixing and begin with problem formulation. 

At Epiq, we focus on how to use these tools both for corporate legal departments and for law firms. Many of the use cases we’ve identified involve summarisation and explanation, and that spans commercial and contracting; regulatory and compliance; and litigation. We continue looking for new ways to incorporate AI into our solutions and interactions, including our end-to-end eDiscovery application, Epiq Discovery, and Epiq Access, which provides our clients with intuitive and interactive dashboards and reports across their legal team.

There's a lot of hype surrounding AI and generative AI, which often needs an expert eye to interpret, t0 leverage deep and evolving search expertise, and to take a serious look at problem formulation. Doing so creates a competitive advantage for organisations that are looking not only to utilise AI tools, but improve efficiencies across the business.  

Eric Crawley, senior vice president and Shah Karim, chief technology officer, at Epiq Legal Solutions.