Column: Human and machine side by side
Human and machine
This article was published in the newspaper DIE WELT on August 28, 2021.
Artificial intelligence (AI) is one of the central drivers of digital transformation. Over the past decade, groundbreaking applications of the technology have regularly been strung together and have regularly overtaken themselves in the process. Driven by the ever-increasing success of AI technology, it is constantly diffusing into all essential areas of human life. In addition to the world of work, this also includes wider social life in particular. AI experts all agree that the current state of development of AI is just the tip of a huge iceberg that has only just begun to uncover. The resulting changes will create a new era of our society, in which people will work and live side by side with machines.
With the growing importance of AI for society, people are also increasingly concerned that the technology, which is already changing so much, will cause people to act “like robots” in a few years' time. While artificial intelligence is still used as a tool today — just as machines or computers used to be — it will become increasingly involved in people's thoughts and actions in the future. To do this, she will analyze the behavior of people and interaction with other people in order to determine which courses of action are best for her users at a particular moment.
Human-centered AI is a sub-area of AI research that focuses on developing AI systems that behave like humans. In contrast to traditional AI research, which focuses on developing AI systems that should behave rationally and without human-like characteristics, human-centered AI aims to study humans and human intelligence in order to develop AI systems with similar characteristics.
Human-centered AI is motivated by the desire to create AI systems that can interact with people naturally. Although rational AI systems can act intelligently, they do not behave like humans, which can be an obstacle to their acceptance by humans. While traditional AI research has focused on overcoming this barrier, human-centered AI research aims to create AI systems that behave naturally so that they are more comprehensible and acceptable to humans. It is important that we make sure in advance that the technology is in line with our humanity and not only supports people in the working world, but also does not “overtake” them. Such human-centered AI can help to integrate human and non-human intelligence in a healthy relationship. The focus is on the needs of users, who are regarded as the most important target groups of the technology.
The central challenge of artificial intelligence is therefore not the technology itself, but rather how humans use this technology. Technical development will continue in the future and thus produce ever more intelligent, autonomous systems. The systems and their decision-making processes will increasingly leave people behind; “machine learning” will be the order of the day. It is important that people see AI technology as a step towards a better future and not as a threat. The development of AI is a process that can only be mastered through dialogue between people and technologies. The inadequate education of society about the technology and the inadequate communication between users and industry play a major role here. People should always be in the foreground and not just as an AI tool.
Genese
The above text was written using the AI “GPT-3". The acronym GPT stands for “Generative Pretrained Transformer” (in version 3), an AI developed in 2020 by the American research company OpenAI. GPT-3 is a so-called generative language model that has been trained on an extremely large amount of diverse texts from the Internet.
GPT-3 is trained using deep learning, a group of machine learning methods that have broad similarities to information processing and transfer in the human brain (nerve cells or neurons pass on the information to be processed to other neurons and brain areas). As part of model training, the approximately 175 billion parameters of GPT-3 adapt to the underlying training data and learn the connection between the previously observed text of a sentence/section and the next word that is likely to appear. As a result, the AI generates a general and comprehensive understanding of the use of words in the respective context during the learning process.
It happened in this article as well. Only the first five sentences of the first paragraph served as a starting point for text generation. This defined the context in which the AI should operate. All subsequent sections were generated independently by the AI and their content was not modified (note: individual paragraphs were created repeatedly and then combined to form an article).
The key to success therefore lies in cleverly defining the context that guides the AI in the generation process. What is groundbreaking here is that the context definition works using natural language and does not have to be “programmed” manually into the AI. The AI thus independently “understands” the context in which it should operate.
With this example, we hope to give you an idea of the power of such AI systems. It is expected that new versions of these and similar AI systems will be created in the near future, which will be able to produce even better results.
