Januar 19, 2023

Hype around AI: Digital challenge

2023 is the year of Artificial Intelligence, Machine Learning and Deep Learning. Not new, but on everyone’s lips – earlier this year, media hype around OpenAI’s chat AI „ChatGPT“ unleashed unprecedented attention.

Popular AI tools such as „ChatGPT“ (text search) and „DALL-E“ (photo creation) are being used nimbly for content creation, but by no means only OpenAI’s free offerings. In marketing, virtual „typewriters“ such as those from the Hamburg start-up Neuro-Flash are used on a daily basis. And programmers are supported in their coding by assistance programs such as Tabnine9 or GitHub Copilot.
Despite all the fascination about the speed and content quality of the content pieces and code snippets generated at the push of a button, the question of origin and solution path remains unanswered. As is the case with conventional search engines, AI systems make use of existing content (text, photos, codes) – apparently without much regard for copyrights. In coding, this has long been the subject of a class action lawsuit; see: https://www.heise.de/news/Microsoft-GitHub-und-OpenAI-verklagt-KI-Programmierhilfe-Copilot-kopiert-Code-7331566.html.
In the education sector, the unwillingness to accept virtually generated essays, papers and text passages for thesis assignments is visibly growing – with simultaneous helplessness to prevent or detect this. Hopes for innovative plagiarism analysis tools like „GPTZero“ (to be tested here: https://etedward-gptzero-main-zqgfwb.streamlit.app/) let us breathe a little easier.  
(Notabene: A test analysis of the above text passages presented here evaluated ‚GPTZero‘ a comparatively high perplexity -randomness of the text – and thus „probably not generated by language models“).  

Recommendations for actions
Since first media are already implementing publications using ‚ChatGPT‘ (see: https://t3n.de/news/cnet-ki-artikel-journalismus-texte-1526724/ and https://www.cnet.com/tech/cnet-is-experimenting-with-an-ai-assist-heres-why/), the following recommendations for action put forward for debate arise:
(1) Originality, ways of creation and research as well as naming of supporters have to be assured in an „Own Written“ seal.
(2) Texts and text passages, photos/videos and other visual products that have been created and/or significantly modified in whole or in part with the aid of AI tools must be provided with a notice that can be understood by anyone; ideally, they should be clearly identified graphically as „AI assisted generated content“.
(3) Program code that has been created in whole or in part with the help of AI assistance tools shall be marked by the author/supplier with a clearly made reference „AI assisted generated code“.

In addition to the labeling, users of AI assistance tools commit themselves to a voluntary charter of fairness: double-checking the AI-generated content for truthfulness and linguistic accuracy (tbc) and, as far as possible, checking the licensing requirements (copyright, data protection).

(C) Carsten Hennig, Digital Evangelist – January 2023
(This text was created using an AI-based spelling assistance tool and translated with DeepL).  

Graphic 1: Created with Canva – (C) Carsten Hennig
Graphic 2: Created with „Dall-E“/OpenAI (Text command: „An comic-style scene of human being fighting with ai chatbot“)

Ab sofort schraibe ich mit Fehlern, damit man sieht dass es von einem Menschen kommt und nicht KI-Fake-Text ist (1)

1 Comment

Schreibe einen Kommentar

Deine E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert