Has the AI Bubble Burst or Is AI Primed for a Resurgence?

Over the past few years, we’ve seen an explosion in AI, perhaps best exemplified by conversational AI and large language models. Proponents of the technology were quick to optimistically speculate about a future in which AI would develop exponentially. However, broadly speaking, AI products have fallen somewhat short of expectations.

Has the AI bubble burst or should we prepare for another wave of technological innovation?

The Landscape of Modern AI

Depending on how you look at it, the landscape of modern AI might look impressively advanced and multifaceted, or somewhat stale and repetitive. Using a competitor analysis tool, you can generate hundreds and perhaps thousands of examples of companies using AI to reach new customers and solve new problems. Yet most of these companies are using the same fundamental technology as the foundation for their work.

Nearly all of today’s most advanced AI products are ones built on very sophisticated pattern recognition and replication. The easiest example of this comes in the form of large language models (LLMs), which study millions of examples of written text to better understand how to simulate human-written content in a convincing way. There are also AI models that apply similar fundamental concepts to still images, music, and video.

The Broken Promises of AI

Lots of technologies have referred to themselves as AI in the past, but it’s only recently that we’ve tinkered with technologies that genuinely seem like they’re artificially intelligent. A few years ago, when ChatGPT first started making waves, most of us were astounded at its ability to answer questions, produce content, and even replicate the tone and voice of some of our favorite authors. In some use cases, the AI even spoke or acted as if it was conscious and legitimately thinking.

Quickly, both AI experts and laypeople began speculating about a near future in which we would all have robot butlers and sophisticated AI companions with legitimate thoughts and feelings. We’ve been told through sci-fi and pop culture that the AI explosion itself is unstoppable, snowballing past technological plateaus we long thought would limit our progress.

People everywhere waited both excitedly and anxiously, wondering if the next year would reveal an AI so powerful it could replace half the jobs in the country. In reality, we’ve definitely seen some advancements and new ideas fleshed out, but ultimately, today’s AI isn’t that much more advanced than the original LLMs we saw introduced a few years ago.

In fact, in some ways, it seems like they’ve taken a step backward. Part of this is because the novelty has worn out. Part of this is because we’ve had more experience with the technology, so we know what its limitations and weaknesses are. But the biggest part of this is that AI has somewhat stagnated, and it’s going to take a massive leap of innovation and ingenuity to get past it.

What’s Holding AI Back

So what exactly is holding AI back? Where are the robot butlers?

No General Intelligence

Contrary to some initial assumptions, what we call AI isn’t really intelligent. It doesn’t utilize the general intelligence necessary to evaluate new situations, practice logical reasoning, or come up with truly creative ideas. Even AI products designed to simulate interactions with real human beings with fleshed-out personalities don’t use general intelligence. Instead, most AI products are a form of glorified pattern recognition. It’s a very advanced form of pattern recognition, but that’s essentially all it is. These machines function the same way basic calculators do, following sets of instructions to obey user prompts and generate material that looks and sounds like similar material it has relentlessly studied. It’s a fantastic, impressive, and genuinely helpful magic trick, but at its core, it’s still just a magic trick. Without general intelligence, the realm of potential applications for any AI software is tightly constrained.

Data Dependencies

An AI built with pattern recognition is only as good as the data available to it. Most generative AI models have huge libraries of available data to browse, but there are still limits to what they can access. Most of them don’t have access to data in real time. Accordingly, there’s a kind of ceiling to the development that most AI models can achieve.

Self-Training Feedback Loops

It doesn’t help that some AI products are already caught in a kind of self-training feedback loop. Early generative AI was harnessed to produce large volumes of content, which was then published on the web. Subsequent AI models were then trained on that AI-generated content, reinforcing AI habits and tendencies and reducing the prevalence of the human mind in the equation. It’s hard to say that this has made AI worse, but it certainly hasn’t helped its potential.

Limited Contextual Applicability

The majority of AI products currently for sale or available for use have limited contextual applicability. You might be able to ask it to write literally anything you imagine, but it won’t be able to solve engineering problems, come up with genuinely new ideas for stories, or perform heart surgery. Most AI is locked in a prison, capable of only a narrow range of potential outputs.

Ethical Concerns

It’s also possible that AI progress has slowed in part due to ethical concerns. Some people have expressed genuine trepidations about what general AI might mean for us. In addition to potentially posing an existential risk, it might lead to heightened inequality and other forms of injustice. Although not everyone shares these concerns, they seem prevalent enough to influence the direction of the industry.

Distrust and Skepticism

Not everyone is aboard the AI train. Some people, witnessing egregious errors and obvious biases from mainstream AI platforms, have withdrawn their support from this technology altogether.

Copycats and Stagnation

After the first round of generative AI models entered the market, thousands of entrepreneurs started foaming at the mouth, eager to create AI products of their own to join the gold rush. This led to the development of countless copycats, adding little innovation but instead mostly replicating technologies that already existed.

Talent Shortages

Compounding the problem, there are massive talent shortages in the realms of machine learning and AI. These are very advanced technologies, and only the brightest, most experienced minds have the potential to push it further. We simply don’t have enough engineers skilled in this discipline to help us punch through our current ceiling and develop the next wave.

Buzzwords and False Hype

Finally, all the buzzwords and hype around AI have not been helpful. They’ve set false expectations about what AI is and what it can be. They’ve also led to people overestimating what current AI tools can do.

The Next Wave

AI experts aren’t sure what happens next. This might be a valley of technological stagnation where we stay for many years, or decades to come. This might be a temporary blip on the radar between massive, fundamental leaps forward in innovation. The next wave of AI might be a super-advanced form of what we’ve already seen, or something borderline incomprehensible to us, closer to the realm of true general intelligence.

So has the AI bubble burst? Yes and no. Generative AI may have been a bit oversold and overhyped. We also might have been a bit too eager to predict a future in which AI launches a new era of blindingly fast technological progress.

However, there’s still much potential for this technology. It’s only a matter of time before we begin to tap into it. The hype may have died down for the moment, but it’s likely to return once we turn the corner and enter the next chapter of this story.

The post Has the AI Bubble Burst or Is AI Primed for a Resurgence? appeared first on ReadWrite.