“DIALOGUE” ROBOTS?
The following excerpt is taken from a contribution to the conference proceedings Desiring Machines. It was written together with my daughter Miriam Stuke at the invitation of Sandor Goodhart and Thomas Ryba, the editors of the volume.
As a specialist in nanoscience, she was the perfect complement to my expertise as an educational scientist – and thus an ideal partner for exploring the conference's theme. Together, we challenged two fundamental concepts in the current discourse: the ability of robots to engage in "dialogue" and the ability of machines to "learn". Here is the first part of our contribution.
Although our presentation had already been published online during the COVID-19 pandemic, we were finally able to hold the finished book in our hands only at the beginning of this year. Many thanks to Sandy and Thomas for their patience and perseverance.
The current discourse regarding “dialogue robots” deals with a wide variety of phenomena and products. Among these, one distinction is particularly helpful. On one hand, we find digital representations and interfaces such as chatbots or avatars. On the other hand, we find analog robots and androids that occupy physical space (Ozaeta and Graña 2018). Companies like Soul Machines (in New Zealand) go to great lengths to develop hyperrealistic digital representations, even implementing mirror images of the human user in the pupil of their avatars.1 In addition, these companies label their products with terms like “identity,” “personality,” and even “digital humans,” or, last but not least, “soul.” On the other hand, companies like Hanson Robotics try to convince us that their products are “socially intelligent,” “good and understanding friends,” or even “a new sort of life form.”² They try to make the facial expressions of their androids look as human-like as possible. To be quite precise, current “dialog programs” are, after all, nothing more than machines that mimic and imitate human language. The main question, therefore, is whether this imitation is convincing enough to make us respond to them in a satisfactory way. For example, do they address us efficiently as customers of their products in web applications? Communicative practice with robots requires either that the human indulges in the experience despite accepting that the robot is only mimicking human behavior or, logically speaking, that the robot “comes alive.” In science fiction movies, the trap of psychological projection is part of the daily business. For example, Ridley Scott (born 1937), film director, commented on the stunning visual effects in his movie Alien I: “The most important thing … is not what you see, but the effect of what you think you saw” (Scanlon, Gross, and Lippincott 1993). Nevertheless, the fundamental ontological questions cannot be decided by resorting to categories of functionalism. Therefore, dialogue, as an intersubjective communicative practice, can never be proved by a utilitarian approach like the Turing Test (Turing 1950).
Facts and Fiction
Since facial expressions are fundamental for human communication, the emotional facial reactions that robots manufactured by Soul Machines manifest in their so-called digital humans are paramount in this development. We also see advances in the technical realization of androids. Hanson Robotics humanoid robot Sophia is a great example. What is important to notice is that for a convincing presentation, the simulation of personality is imperative. In our opinion, the most interesting element of such a machine is its distinguishability from humans. We do not expect to see androids that can fool someone familiar with modern technology into thinking they are human anytime soon, not only because of the strange speech patterns and pauses between responses but also because of the lack of idiosyncrasies. Another issue is the set of presuppositions that humans make in communication, which help humans to adequately react to illocutive speech acts.Thus, the embodiment, gesture, and general appearance of a human is technically feasible, for example, as a perfect doll with a puppeteer acting remotely. An example from the late eighteenth century is the Maelzel’s automaton chess player described by Edgar Allen Poe in his famous story (Poe 1984). In Japan, Ishiguro Lab is producing realistic androids for commercial use.³ An example is “ERICA” (Glas et al. 2016; Lala et al. 2021), which is essentially a desk assistant that very convincingly mimics human facial expressions. The next steps in the development of this android are to install modules that emulate intention and desire to get seemingly emotional behavior (Cowley 2006).
Communicative Patterns
Returning to our present consideration of dialogue-reality, it seems to be helpful to start by differentiating dialogue from other communicative practices. We propose a four-field matrix systematizing “four types of communicative interaction along the axis of process-orientation vs. result-orientation and person-orientation vs. topic-orientation. Each type has its own strengths and weaknesses and is appropriate for reaching different goals” (Stuke 2021, 83). Dialogue, in our understanding, is appropriate for listening, perceiving, expressing, and thereby strengthening mutual understanding among all participants as human beings in coexistence. In addition, a discussion is useful for enlarging argumentative richness by gathering and categorizing arguments within a complex question requiring necessary co-orientation. A debate focuses on agonistic attack and defense to clarify strengths and weaknesses of alternative positions in a preclarified structure of coordination. If a decision must be made, communication aims to reduce complexity between distinct alternatives to a responsible degree for practical cooperation. Although none of these patterns occur in absolute purity, it is obviously necessary to clarify the context in which they occur (Habermas 1984) in order “to know where we are” (Chiara Lubich, quoted in Torno 2012, 49). We need to become aware of the communicative setting as well as the main aim pursued within the actual situation. This matrix is shown in Figure 19.1. A probable or useful implementation of AI is indicated with colors in the following illustration.
Communication programs cannot satisfy the requirements of a “self” which is necessary for a dialogue,⁴ Sophia is only ever a simulation of the capability of wisdom and friendship.⁵ A debate, on the other hand, can well be accomplished by an automatic system because of its clear structure. IBM has installed such a system under the name “Project Debater” (Slonim et al. 2021). It contains automatic speech analysis, including argument detection, segmentation, and classification with respect to a defined position for a given topic. Discussions are similar to debates in that they contain classifiable arguments. For this reason, the core of discussions will remain human, while AI can aid in the hierarchical storage and labeling of arguments. For ethical reasons, and because a machine is built by humans, the responsibility for decisions must remain with the human, as this is the only way to ensure that no subliminal bias or hidden design flaw causes harm (Krafft, Zweig, and König 2020). A machine can anticipate what it has been trained on, but only humans can imagine different futures outside the data horizon. There are three possible variants for dealing with this issue: making the machine responsible, making the creator responsible, and leaving responsibility with the user. The first variant is illogical as the machine is an object that is not reflective only reactive. The second variant is challenging as the corpus of people behind an AI program is enormous so that either no one is prosecuted or one person is singled out (Dremliuga 2020). In this context, one thinks of scapegoating in Girardian theory. In the third variant, the AI affects us but is never required to justify itself and not culpable. This issue obviously opens a whole other can of worms, one that we do not address in this chapter.
excerpt 149-152 from: Stuke, T./ Stuke, M.: “Dialogue” Robots and Machine “Learning”. The Mimetic Trap of Wishful Thinking. In: Goodhart, S., Ryba, T. (Hg): Desiring Machines: Mimetic Theory and Artificial Intelligence in Technology, Philosophy, Film, and Fiction” Bloomsbury Academic, Purdue University 2026, S.149-156, doi: 10.5040/9781350383951
1 See; https://soulmachines.com/wp-content/uploads/2020/04/Sam-Square-1024x1024.png. See also https://soulmachines.com/
2 https://hansonrobotics.com/. Compare also: “Of Objects and Affect Artificial Empathy, Pure Sociality, and Affective Coordination” (Dumouchel 2017) on pure sociality. See video on Sophia: “I’m a new sort of life form ... I definitely need to have the human experience to be a good and understanding friend.” (https://yout ube.com/watch?v=XIIiwTfHLpU, accessed July 4, 2021).
4 related to Greek dialogesthai “converse,” from dia “across, between” (see dia-) + legein “to speak,” https://etymonline.com/word/dialogue, accessed July 8, 2021
5 gr. wisdom
Comments