Two years into the AI craze, where does it go next?

We are almost two years into the reality of large-language-model (LLM) artificial intelligence. Now, the future of AI is heralding the return of nuclear power: both Google and Microsoft have announced that they are investing in the revival of power plants to satiate the depthless energy needs of their models. Nearly every startup in the last year has some kind of AI in its pitch deck, if not in its name. There is a lot of money (estimates from Vanguard in 2023 put the number over $60 billion) – and attention – being thrown at the promise of an AI-driven future. That promise really comes down to this: AI will allow people to do more digital work, faster, while thinking about it less. Whether this bears out in reality, however, is still up for debate. 

Can AI live up to the hype?

The answer to whether the majority of AI-generated content can be used for commercial purposes is still tied up in copyright lawsuits. Google’s AI search results are often completely wrong, as are ChatGPT’s answers. Even more specialised, fine-tuned LLMs are prone to complete hallucinations. OpenAI’s Whisper, for example, a translation and transcription tool often used in medical contexts, has been found to simply make up things that no one ever said (one study found hallucinations in eight out of 10 transcripts, according to AP). 

Disclaimers say that LLMs should not be used in high-risk scenarios. Yet if they cannot ease the hardest jobs, their use is limited to creating subtitles for YouTube videos, pitching mid-rate TV shows, generating corporate blog posts, and creating filler noise for ads. Which begs the question: is it really worth all the billions in investment and overhaul of national power grids to fuel what ultimately serve best as meme generators? 

It would make more sense to conclude that AI is so hyped for other reasons entirely. Big tech companies need a ‘next big thing’ to continue their growth narratives. Most startups have short life cycles that end with being bought up by one of the big ones. Many investors do not know their LLM from their algorithm. Everyone leapt at the promise of AI without waiting to see the reality, and are now utterly invested in a multi-billion dollar game of chicken.

AI’s future impacts

The things LLMs do well are relatively boring, and not highly profitable. Yet running them is very costly, with numbers reaching tens of millions according to TIME. The budget sheets need to balance out, but they do not look close to doing so anytime soon. So, as we look to the next two years of AI live on the internet, it needs to be asked: what happens if that gap does not close? Here are some of the impacts we will see:

1. The rising cost of AI

AI is being incorporated into processes across the entertainment value chain, from boosting creator tools for independents to replacing pitch teams at ITV. The purely generative models have copyright issues, but these more niche, process-oriented ones will be less affected by regulation. However, they still require physical servers and the energy to power them. The current costs of operation outweigh the prices paid by companies that use them. They will either have to hike up prices for storage and computing, which will make incorporating AI a cost burden rather than a cost saver, or they will simply go under – leaving companies who have incorporated them high and dry.

2. A sharp decline in quality of content

Entertainment dominated by AI generation, regardless of how, tends to have a certain style. It is also driven by the averages of datasets that have come before, making it unlikely to be novel or interesting to consumers increasingly overwhelmed by a deluge of digital content. AI that continues to train on an internet of AI content will also deteriorate in quality. Entertainment that relies too heavily on AI for its show ideas or sound production might find itself lagging behind better-curated options with more organic creative endeavours behind them. AI can help some companies play the numbers game – but in an over-cluttered digital environment, more does not always equal better, and margins will drop as value dilutes.

3. Companies pivoting to new ventures

If Google and Microsoft have exclusive energy rights to nuclear power plants in a world moving away from fossil fuels, they actually have a new opportunity far outside of AI itself: as energy companies. This would be a pretty big plot twist but a strong long-term strategy, if the promise of AI as the next big tech boom falls flat. Energy never goes out of fashion.