Analysis: AI needs its iPhone moment

Contents

If artificial intelligence were already where some of its prophets saw it a year ago, this analysis would be written quickly. Just throw a few ideas and chunks of sentences into the funnel, press the button, done. The Thermomix for texts will do the trick.

An analysis by Malte Kirchner

Malte Kirchner has been an editor at heise online since 2022. In addition to technology itself, he is interested in how it is changing society. He pays particular attention to news from Apple. He is also involved in development and podcasting.

Instead, even in August 2024, the same mediocre quality of text, inspiration and ingenuity would still come out of the machine. And whether it will ever be possible to generate a top text in this game of mathematical probabilities is more than questionable. It is by no means the case that the author claims to be able to meet all three criteria in the following. But, and this is the crucial difference: humans at least have a chance. AI – and this is about current generative AI and not about applications in medicine and industry – continues to have its limits.

Subdued expectations

And if you read the reports of the past few months, these limitations are unlikely to change any time soon. Expectations for OpenAI's GPT 5 are already being tempered. Daring gadgets such as the Rabbit R1 or the AI Pin are falling far short of expectations or - as in the case of the AI Pin – are already being returned en masse. Collective disillusionment is spreading, at least among those who were caught up in the hype.

The market research company Gartner draws trends in curves, known as hype cycles. The AI hype, as the majority of the editorial team agrees, has passed the peak of exaggerated expectations and is in free fall into the valley of disappointment.

There are many reasons for this: you don't have to be a prophet to predict that the high pace of innovation of the past two years will be difficult to maintain in the long term. R1 and AI Pin are examples of how quickly half-baked products were thrown onto the market just to achieve the next advance. The two attempts at AI endpoints have thoroughly failed to make AI acceptable in everyday life. They are not even good as a feasibility study.

So far, it has barely arrived in everyday life

In addition, there is an increasing lack of fresh, high-quality data for training the large language models. Humanity will soon simply have nothing more to offer AI. There is no other explanation for the fact that people are even looking for fresh water in YouTube's septic tanks. Moreover, if AI-generated content is increasingly being thrown into the maw of the LLMs because it simply cannot be identified as such, then fears are justified that even the existing content will deteriorate. There is also the fear that AI models could even collapse. The uncertainty alone about whether this could happen and whether countermeasures are necessary raises doubts about the reliability of artificial intelligence. The lack of suitable AI chips, on the other hand, appears to be a problem that is almost easy to solve.

The biggest problem, however, is that despite all the hype and speed, AI has so far barely arrived in people's everyday lives. At least not to an extent that justifies the immense investments and high stock market values that make us believe, like the start-up bosses, that AI will soon be everywhere.

There is no doubt that artificial intelligence has long since conquered spaces. Programmers get the right answers to problems faster than searching the web – and that will certainly remain the case. And chatbots from support systems are sometimes a little less annoying than they used to be. Some also swear by the ability to summarize texts briefly or the smarter translation services.

In many more cases, however, AI is still a technology with potential. Just as the capabilities of the UMTS network only paid off when the masses started using smartphones, AI is still waiting for its iPhone moment. And Sam Altman can be in such a hurry to release new technology with OpenAI: As long as companies struggle to recognize tasks that can be completed more efficiently with AI and lack the courage and conviction to put this into practice, the AI train will only move at a crawling pace.

Courage to remain calm

It is fitting that even the iPhone manufacturer is still working on a suitable response to the hype: Apple Intelligence, which, like the iPhone, is supposed to do some things differently in the existing world, could, with its focus on data protection and local processing of data, initiate the path of enlightenment that Gartner sees in every hype cycle and lead AI to the plateau of productivity. But it could just as easily be a dead end. It is significant, of course, that Apple has so far taken a fairly conservative approach to implementing AI in its devices. Perhaps calmness is the key to anchoring AI in everyday life.

The LLMs urgently need a period of consolidation. Instead of chasing the next wild movie generation sow around the place, it would be more important to remove uncertainties and deliver what has already been promised up to this point. For productive use in a business environment, it is also crucial to ask whether AI really leads to cost savings - or whether the human supervision effort is so high due to hallucinations and inaccuracies that the employer ends up paying twice: for the employee and for the AI tokens. Only when such reservations have been dispelled will confidence perhaps grow that AI is a solid foundation on which something can be built.

(mki)