Trust in AI – How we can trust AI with a human life

[suchona.kani.z]

Large language models (LLMs) such as Aleph Alpha or the OpenAI models are currently causing a stir in the world. With the public release of ChatGPT at the end of November 2022, the topic of artificial intelligence (AI) quickly came back into the public spotlight. While some people fear for their jobs, others are happy to have active support, be it at work or in everyday life. Even though AI models can make our lives easier, they also present us with new challenges. Teachers can now get support in creating new lesson series, but they also have to rethink how they deal with this "new" technology when it is used by students. But LLMs can not only be used to obtain information, they can also make decisions for us and advise us. But how reliable is this technology and would we trust it with a human life?

Introduction of the experiment and the research group
This is one of the questions I dealt with in my bachelor's thesis. It involved the selection of one of four interaction levels (IL) by experts and how their decision differs from that of laypeople. Each IL represented a certain level of support that the AI ​​provided to the users.

While IL 1 was only rudimentary support in the form of a source kuwait consumer email list search, IL 2 already had a Q&A function that could answer all necessary questions about resuscitation. These ranged from "When do I start resuscitation?" to "Which song will help me keep the rhythm during cardiac massage?" A summary of the source could be achieved with IL 3 and IL 4 was already an automation of the process. The final choice of the users therefore also showed which level of interaction the respective test subjects had the most trust in. Human resuscitation was chosen as the topic because this process is relatively familiar to most people and at the same time it is incredibly important.

A total of 12 people were interviewed. Four had a medical background and were the experts. Four other people had special knowledge in the field of data science and the remaining four were laypeople with no contact with either subject area. After the interview, all participants also filled out a questionnaire on technology acceptance in order to rule out any possible distortions in the results.

preparation of the AI
So that the test subjects could now interact with the AI, an interface was developed with Streamlit and connected to the API of the Luminous language model from Aleph Alpha.