Federal Minister of Labor Hubertus Heil expects a rapid change in the world of work through artificial intelligence (AI). However, he does not believe that jobs will be destroyed overall. “Individual jobs will be lost, but many new ones will also be created,” the SPD politician told the Tagesspiegel newspaper. He added: “From what we know and can gather, we as a society will not run out of work in the future either.” But it will be different work in many ways.
The experts in the ministry predicted that from 2035 there would be no more jobs that have nothing to do with AI. In order to alleviate the shortage of skilled workers, Heil sees “a huge potential in artificial intelligence that we must fully exploit”.
The chat AI ChatGPT and Google’s competitor software Bard are currently causing a stir, as are programs that can generate images based on text descriptions, such as Midjourney. At the same time, there are concerns that such AI-based technology could be misused to spread false information, for example.
Heil does not believe in bans on such programs, but strongly promotes transparency. The minister warned that people should know which data are used to train these learning AI systems. This not only prevents conspiracy theories, but also prevents the systems from producing undesirable results. He warned: “If the systems are fed with wrong data, it will be problematic.”
Training for chat AI
The AI text tool ChatGPT, for example, comes from the company OpenAI. It can understand human input and react to it comparatively naturally – for example by answering questions, writing poems or even outputting code. The basis is OpenAI’s language model GPT-3.5. This is trained with terabytes of text from the Internet, in which it tries to recognize patterns. This definitely has its pitfalls, because such language models are only as good as their training data: if there is nonsense in it, the AI reproduces it.
To prevent this, OpenAI relies on “Reinforcement Learning from Human Feedback” – reinforcing learning with human feedback. So, first of all, people answered inputs. On this basis, ChatGPT itself answered inputs and created several outputs. A human evaluates these expenses. This also creates a ranking from the best to the worst output. ChatGPT takes this feedback into account and tries to optimize its output. In addition to ChatGPT, there have long been other AI chatbots such as Google Bard.
(ssi)
