AI winter: A cycle of hype, disappointment, and recovery

The term AI winter refers to a period of funding cuts in AI research and development, often following overhyped expectations that fail to deliver.

With recent generative AI systems falling short of investor promises — from OpenAI’s GPT-4o to Google’s AI-powered overviews — this pattern feels all too familiar today.

Search Engine Land reported that AI winters have historically followed cycles of excitement and disappointment. The first of these, in the 1970s, occurred due to the underwhelming results from ambitious projects aiming to achieve machine translation and speech recognition. Given that there was insufficient computing power, and the expectations of what computers could achieve in the field were unrealistic, funding was frozen.

The expert systems in the 1980s showed promise, but the second AI winter occurred when these systems failed to handle unexpected inputs. The decline of LISP machines, and the failure of Japan’s Fifth Generation project, were additional factors that contributed to the slowdown. Many researchers distanced themselves from AI, opting to call their work informatics or machine learning, to avoid the negative stigma.

AI’s resilience through winters
AI pushed through the 1990s, albeit slowly and painfully, and was mostly impractical. Even though IBM Watson was supposed to revolutionise the way humans treat illnesses, its implementation in real-world medical practices encountered challenges at every turn. The AI machine was unable to interpret doctors’ notes, and cater to local population needs. In other words, AI was exposed in delicate situations requiring a delicate approach.

AI research and funding surged again in the early 2000s with advances in machine learning, and big data. However, AI’s reputation, tainted by past failures, led many to rebrand AI technologies. Terms like blockchain, autonomous vehicles, and voice-command devices gained investor interest, only for most to fade when they failed to meet inflated expectations.

Lessons from past AI winters
Each AI winter follows a familiar sequence: expectations lead to hype, followed by disappointments in technology, and finances. AI researchers retreat from the field, and dedicate themselves to more focused projects.

However, these projects do not support the development of long-term research, favouring short-term efforts, and making everyone reconsider AI’s potential. Not only does this have an undesirable impact on the technology, but it also influences the workforce, whose talents eventually deem the technology unsustainable. Some life-changing projects are also abandoned.

Yet, these periods provide valuable lessons. They remind us to be realistic about AI’s capabilities, focus on foundational research, and communicate transparently with investors, and the public.

Are we headed toward another AI winter?
After an explosive 2023, the pace of AI progress appears to have slowed; breakthroughs in generative AI are becoming less frequent. Investor calls have seen fewer mentions of AI, and companies struggle to realise the productivity gains initially promised by tools like ChatGPT.

The use of generative AI models is limited due to difficulties, such as the presence of hallucinations, and a lack of true understanding. Moreover, when discussing real-world applications, the spread of AI-generated content, and numerous problematic aspects concerning data usage, also present problems that may slow progress.

However, it may be possible to avoid a full-blown AI winter. Open-source models are catching up quickly to closed alternatives and companies are shifting toward implementing different applications across industries. Monetary investments have not stopped either, particularly in the case of Perplexity, where a niche in the search space might have been found despite general scepticism toward the company’s claims.

The future of AI and its impact on businesses
It is difficult to say with certainty what will happen with AI in the future. On the one hand, progress will likely continue, and better AI systems will be developed, with improved productivity rates for the search marketing industry. On the other hand, if the technology is unable to address the current issues — including the ethics of AI’s existence, the safety of the data used, and the accuracy of the systems — falling confidence in AI may result in a reduction of investments and, consequently, a more substantial industry slowdown.

In either case, businesses will need authenticity, trust, and a strategic approach to adopt AI. Search marketers, and AI professionals, must be well-informed and understand the limits of AI tools. They should apply them responsibly, and experiment with them cautiously in search of productivity gains, while avoiding the trap of relying too heavily on an emerging technology.