Examining Autocompletion as a Basic Concept for Interaction with Generative AI

Abstract

Autocompletion is an approach that extends and continues partial user input. We propose to interpret autocompletion as a basic interaction concept in human-AI interaction. We first describe the concept of autocompletion and dissect its user interface and interaction elements, using the well-established textual autocompletion in search engines as an example. We then highlight how these elements reoccur in other application domains, such as code completion, GUI sketching, and layouting. This comparison and transfer highlights an inherent role of such intelligent systems to extend and complete user input, in particular useful for designing interactions with and for generative AI. We reflect on and discuss our conceptual analysis of autocompletion to provide inspiration and a conceptual lens on current challenges in designing for human-AI interaction.

Generative artificial intelligence empowers educational reform: current status, issues, and prospects

“…In generative AI, deep learning techniques mainly include Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Variational Autoencoders (VAE), among others. CNN is a commonly used deep learning method that can effectively extract features from data and is used in fields such as images and speech (Lehmann and Buschek, 2020). In generative AI, CNN can be used for image and audio generation, for example, by training machines to recognize objects in images and then letting them automatically synthesize new images (Oermann and Kondziolka, 2023).…”

Bias in AI Autocomplete Suggestions Leads to Attitude Shift on Societal Issues

“…Alarmingly, our work shows that biased AI models can distort attitudes in a covert and implicit way, without people noticing that they are being presented with a persuasive argument by a different actor, be it AI or human. The context of our study is representative of how people encounter and interact with AI auto-completions offered in everyday communication contexts, such as on email (Hancock, Naaman, & Levy, 2020;Hohenstein et al, 2023) and in search engines (Heer, 2018;Lehmann & Buschek, 2022). The biases shown by the models used in auto-complete settings are not known or well understood, but it has been shown in other settings that LLMs are inherently biased (Feng et al, 2023) and, further, that malicious actors can inject LLM models with specific biases that are triggered by specific topics and keywords (Bagdasaryan & Shmatikov, 2022).…”

“…Figures 1-4 Table 1 Arnold, Chauncey, & Gajos, 2020; Lehmann & Buschek, 2022). Since behaviors are known to shift attitudes (Janis & King, 1954;Festinger, 1962;Bem, 1972), biased AI models' nudge to write different text may also result in shifts in people's attitudes on the issue at hand.…”

Strategyproof Learning: Building Trustworthy User-Generated Datasets

“…Today's most influencing algorithms often leverage such data to learn a common model that fits all users' data [27], or to construct a personalized model for each user [38,15,19]. Autocompletion [29], conversational [39] and recommendation [25] algorithms are examples of such algorithms. To be effective, they require huge amounts of data [7,16].…”