Could an AI winter happen again?

From the peak of the AI hype cycle, the only way to travel is down.

While the AI applications at our fingertips are remarkable — some might say magical — they are not a universal solution for every problem as some enthusiasts proclaim. These tools excel at specifically designed tasks and show promise for diverse applications. But there are legitimate concerns that people may become disillusioned if promised functionality fails to materialize.

This pattern of enthusiasm for advanced computing applications, followed by disappointment, is hardly a novel phenomenon. We witnessed a similar situation during the AI Winter of the 1980s and 1990s, when investment in AI-related computing applications dwindled after the hype subsided due to unrealized anticipated benefits.

While the media fervently shines light on AI's capabilities and potential for future development, there are reports about large employers of computer science and data experts letting go of large portions of their workforces. New graduates in these fields are having a harder time finding jobs. There is some speculation that it is a symptom of a new AI winter taking hold.

According to Kevin Ashley, a professor of law and intelligent systems at the University of Pittsburgh, the slight decline in job opportunities is more attributable to the extensive use of automated tools that allow tasks like computer coding to be accomplished with less labour, than to a decrease in investment or a significant loss of optimism.

Mark Doble, the CEO at Alexi, acknowledges the remarkable progress in AI development. However, he cautions that there is a heightened sense of confidence in these applications, with expectations of immediate capabilities that will likely lead to some degree of disappointment. Doble emphasizes that deviations from long-term trends in technology development are typically not realized as expected.

Doble says that a key factor contributing to the current hype is a misunderstanding of the intended functionality of tools like ChatGPT, which aren't designed as general intelligence systems. Instead, they are language models that predict the most probable series of words. While they manifest some degree of general knowledge, their capabilities are inherently limited to the scope of language models.

However, there are opportunities to enhance AI systems by integrating diverse approaches and combining different models to pave the way for developing more robust systems.

AI systems don't need to be fully autonomous to be useful. Doble advocates for the adoption of systems that incorporate human oversight without compromising reliability. These systems can be programmed to incorporate feedback, enabling rapid improvements over time.

Ashley suggests that, from a research perspective, the first AI winter did not hold the field back as much as it could have. AI was needed in legal practice because of the rise of extensive collections of digital documents that had to be processed for e-Discovery. And research funding was always available, though it became more difficult to obtain.

Before the first AI winter, people were similarly optimistic about the imminent widespread deployment of knowledge bases — systems that perform functions such as information retrieval based on structured data. However, populating and maintaining these systems was prohibitively expensive and difficult. And people's expectations were deflated as a result. Still, the deployment of these systems continues today. The internet is the world's largest knowledge base.

According to Ashley, research is ongoing in many areas favoured in the 1980s. His doctoral thesis focused on developing a case-based system for trade secrets, a topic which still draws interest, except people are now using ChatGPT4 to extract the data from documents instead of doing it manually — proof that current efforts build upon previous advancements.

Looking ahead, it seems inevitable for there to be some correction. However, the industry will continue to witness research and development, leading to the emergence of new working methods. These developments carry significant implications for lawyers' work, firms' profitability, and the accessibility of justice within communities.