Artificial intelligence: Writing AI generates hype and warnings

Artificial intelligence
Writing AI provides hype and warnings

New software can write text in seconds that can hardly be distinguished from that of a human – that could change the world, with serious consequences for millions. “If your job involves writing emails, drafting documents, writing articles or copywriting, exchanging legal papers: you have to assume that this will have a profound impact.

And not necessarily a good one,” warned computer scientist Sridhar Ramaswamy at the DLD innovation conference in Munich. And Abba musician Björn Ulvaeus predicted that software would write better music than many of today’s songs.

Forecasts that software with artificial intelligence will replace office workers, just as automation once eliminated many factory jobs, have been around for a long time. So far, machine learning has been used for helper applications and seemed far from ready. Then in November the ChatGPT software came out and triggered a hype. ChatGPT can write any text such as essays, business letters, poems, news articles on command – and imitate the style of certain authors if you wish.

How is that?

The software is trained on gigantic amounts of text and imitates what it knows by predicting the plausible next words. The result is always grammatically correct, solid – and somewhat uninspired. But for everyday scenarios such as a letter of resignation or an e-mail, that’s always enough.

Knowledge questions are also answered in full sentences based on the information recorded. If you ask ChatGPT how old the president of Australia is, the software replies: “Australia has no president”. But then ChatGPT also lets you know that Prime Minister Scott Morrison is 54. Only: Anthony Albanese has been Australian Prime Minister since May last year. However, the ChatGPT knowledge base was created in 2021. Sometimes the software refers to it, sometimes not. Worse still, in another attempt, ChatGPT makes Morrison President.

Answer looks convincing but is incorrect

Now ChatGPT is still an experimental project that can and will learn. However, the error reveals a fundamental problem: the answer looks convincing, but it’s wrong – and the user has no clue to judge that.

At the same time, authors of targeted false information also get a powerful tool. The technology creates “endless possibilities to formulate relatively plausible lies very quickly,” warned Silicon Valley veteran Phil Libin in Munich. This year, “a wave of nonsense” is rolling towards us. Over time, artificial intelligence will become more grounded in reality and will then benefit from its capabilities.

Until then, however, one must resist the temptation to make working with programs like ChatGPT easier and to automatically produce poor-quality content, emphasized Libin. That will only “increase mediocrity,” he warned. If something could be written by an AI, it shouldn’t be written that way as a human. “We need to raise the bar on what it means to have something created by a human – with a level of quality and originality.”

Microsoft is investing $1 billion in OpenAI

Elsewhere in the tech industry, language-savvy software with artificial intelligence is also being worked on in many places. While the ChatGPT developer OpenAI made its program publicly available, Google, for example, has kept its language program under lock and key and only uses it internally.

Microsoft could benefit from ChatGPT. The software giant invested $1 billion in OpenAI in 2019. This was followed by another two billion, as reported by the “New York Times” and the website “The Information”. With the money, OpenAI paid for the required computing power. And now another investment of ten billion dollars is under discussion. Microsoft can thus secure a third of OpenAI – and also plans to use the AI ​​​​technology in its previously beaten search engine Bing, wrote “The Information”.

dpa