The AI Hype Cycle: Insights into the Rise and Fall of Artificial Intelligence
When OpenAI’s ChatGPT launched in late November 2022, it revolutionized the tech world like few innovations before it.
Not only did it send shockwaves through Silicon Valley, as companies rushed to develop their own AI projects, but it also propelled generative AI into the mainstream, sparking widespread conversations about artificial intelligence.
However, over two and a half years after the initial surge in AI interest, evidence suggests that the excitement may be waning. Since peaking last month, the stock prices of major tech companies driving the AI revolution have dropped by 15%. Increasingly, investors are questioning whether AI can deliver the substantial returns they had anticipated.
Simultaneously, experts are scrutinizing the limitations of large language models (LLMs), the engines behind technologies like ChatGPT. Despite the tens of billions of dollars invested by major tech firms since the AI boom began, very few companies are actually using these models in practice.
Recent data from the US Census Bureau reveals that only 4.8% of American companies currently use AI models to produce goods and services, a decline from a high of 5.4% earlier this year.
While long-term predictions for AI remain positive, with tech giants continuing to invest heavily, this drop in adoption suggests a recalibration in the AI boom. For the first time, it appears the AI hype machine is slowing down.
Understanding the AI Hype Cycle
The AI hype cycle is a recurring pattern of intense enthusiasm followed by disillusionment surrounding artificial intelligence. It mirrors the trajectory of other technological innovations, as outlined by Gartner's Hype Cycle. The typical stages are:
- Technology Trigger: A breakthrough in AI generates significant buzz and media attention.
- Peak of Inflated Expectations: Overhyped claims and unrealistic expectations about AI's capabilities emerge.
- Trough of Disillusionment: As initial excitement fades, the challenges and limitations of AI become apparent, leading to decreased interest.
- Slope of Enlightenment: Practical applications of AI begin to surface, leading to a more realistic understanding of its potential.
- Plateau of Productivity: AI becomes widely adopted across industries, delivering tangible benefits.
A Gartner report from June indicated that most generative AI technologies are currently at the peak of inflated expectations or still on the rise. The report suggested that these technologies are two to five years away from becoming fully productive.
However, as the limitations of current AI technologies become clearer, we may soon enter a period of disillusionment before AI matures and reaches its full potential.
The Reality of AI's Capabilities
Breakthroughs in generative AI, particularly in LLMs, have captivated the world and led to ambitious claims about AI's potential. Yet, many of today’s AI models are not powerful enough to meet these expectations. A recent study by the American think tank RAND revealed that 80% of AI projects fail, a rate more than double that of non-AI projects.
"The unusual nature of generative AI's limitations is one of its biggest barriers to success. While AI systems can solve complex problems, they often fail at simpler tasks, making it difficult to gauge their true potential."
Examples of failed AI projects, such as McDonald’s attempt to automate drive-through orders and the US Government’s efforts to summarize public submissions, illustrate the disconnect between AI’s advanced capabilities and its performance in basic tasks.
Moreover, a study by Cornell University showed that the abilities of large language models like GPT-4 do not always align with user expectations, especially in high-stakes scenarios where incorrect responses can have significant consequences.
Successful AI projects demonstrate that it is challenging to achieve accurate responses from generative models with human prompts. Even systems like Khan Academy’s Khanmigo tutoring tool, designed to aid learning, often struggle with providing appropriate responses.
Where Is AI on the Hype Cycle?
As more organizations implement AI solutions, the real-world challenges of deploying these technologies are becoming more evident. This growing awareness may signal that we are approaching the trough of disillusionment, where the gap between AI’s capabilities and overblown promises becomes clear.
This phase is a natural and necessary part of AI’s development. Once the initial hype fades, the tech industry can focus on practical applications, address limitations, and ultimately reach the plateau of productivity, where AI delivers meaningful benefits across various sectors.
Will AI Lose Its Hype?
The current excitement surrounding AI is driven by recent breakthroughs, especially in large language models. However, as with any new technology, initial enthusiasm often gives way to a more measured assessment of its capabilities and limitations.
It’s important to note that a decline in hype doesn’t mean AI’s importance is diminishing. Just as the internet and smartphones endured their own hype cycles, AI is likely to become an integral part of our lives, even if public interest wanes.
AI is also rapidly improving. Studies show that increasing the size of language models, as well as the data and computing power used for training, contributes to better performance. These models also exhibit emergent abilities, where unexpected skills arise once a certain level of complexity is reached.
As tech companies continue to invest in building larger, more sophisticated models, AI's capabilities will likely improve. According to recent estimates, generative AI will need to generate $600 billion in annual revenue to justify current investments, a figure that could grow to $1 trillion in the coming years.
Ultimately, AI’s success will be measured by its practical applications and the tangible benefits it delivers to society. For now, those real-world applications are still on the horizon.