Should robots be designed like humans

Artificial intelligence

Christoph Kehl

To person

holds a doctorate in the practice and theory of biomedical memory research and works at the Office for Technology Assessment at the German Bundestag (TAB). [email protected]

The latest advances in artificial intelligence (AI) are undoubtedly impressive: it is now normal that computers have been vastly superior to humans in chess for decades. However, the fact that an AI system is capable of quickly reaching a level in the much more complex and unpredictable game of Go that does not give the world's best players the slightest chance - as happened recently - was until recently considered unthinkable.

Algorithms control more and more areas of work and life, from Internet searches to computer-aided medical reports to financial transactions. As a rule, these are highly specialized software agents that operate in a purely virtual environment. But work is also being carried out on "intelligent" systems that are able to move in the analog world, intervene in it and come into physical contact with people. Machines of this type have the potential to completely redefine the human-technology relationship: Insofar as technology begins to act autonomously and thus undermine simple thinking in terms of means-ends (as is decisive for the use of conventional tools) , the demarcation between people and the tools they create is becoming increasingly blurred. This phenomenon is referred to below as human-machine delimitation.

The societal explosiveness of the topic is increasingly becoming a political consciousness. In 2016, the European Parliament felt compelled to recommend civil law regulations in the field of robotics to the EU Commission. [1] In the Bundestag, "Artificial Intelligence" and "Robotics" were the subject of various hearings, and the Office for Technology Assessment at the German Bundestag (TAB) was commissioned to examine the technological basis and social perspectives of this development. In the following, results and considerations from the associated TAB project are presented, which in the exploratory phase dealt with technological and visionary aspects of man-machine delimitation [2] and in the deepening phase defined the societal challenges of development in more detail, using the example a specific application context - maintenance.

Between vision and reality

Above all, the developments in two fields of technology - robotics and neurotechnologies - are contributing to the fact that the formerly sharp dividing lines between man and machine are becoming increasingly blurred.

Advances in robotics meanwhile make it possible to machine even complex human actions. While the classic industrial robot is committed to the automation of repetitive industrial processes and performs its services largely isolated from people for safety reasons, a wide range of application perspectives also open up for modern service robots outside of industrial production halls - be it simple tasks in private households (in which vacuum cleaner robots are already being used millions of times ) or complex, also personal care services. This step into the everyday world was made possible in particular by the following features, with which service robots can be distinguished from classic industrial robots: on the hardware side, the lightweight construction, which - supplemented by complex sensors and actuators - enables ever closer interaction with humans; On the software side, the implementation of highly developed learning and planning processes (based on AI and machine learning), which should enable the systems not only to unwind a fixed program of action, but also to find their way largely autonomously under new or changing conditions. Service robotics is now regarded as a future-oriented growth market whose global volume could overtake established industrial robotics in a few years, according to forecasts. [3]

On the one hand, if robotics ensures that machines become more and more human-like in their possibilities for action, advances in the field of neurotechnologies, on the other hand, open up completely new options for technologically interacting with humans. The standard clinical repertoire already includes stimulating systems that transmit electrical impulses to the brain - such as sensory neuroprostheses such as the cochlear implant, which can be used to technically compensate for impaired hearing through targeted stimulation of the corresponding nerve fibers. In addition, intensive research is also being carried out on dissipative applications with the aim of obtaining signals from the nervous system and using them to control artificial limbs, for example. This makes use of the fact that mental activity generates electrical potentials that can be detected both invasively (by means of implanted microelectrode probes) and non-invasively (by means of electrodes fixed on the scalp). Both approaches have their specific limitations: The non-invasive data acquisition suffers from a rather poor signal quality and can therefore only be considered for the binary control of simple communication systems. [4] With invasive recording methods, higher data rates can theoretically be achieved, which brings the fine control of artificial limbs into the realm of what is technically possible - however, limiting factors here are the still very limited long-term stability of the implanted probes, which are usually after the foreign body reaction of the brain (encapsulation) lose their function for a short time.

Even if the derivation of data from the brain and thus the intuitive thought control of complex devices has not yet got beyond the experimental stage, further application perspectives are already emerging. Should it one day succeed in establishing a stable bidirectional feedback system by integrating dissipative and stimulating interfaces, "feeling" hand prostheses could be implemented that give the user sensory feedback during the grasping process. The delimitation of human-machine boundaries is finally taken to extremes by the combination of neurotechnological applications with robot technology - for example in the form of "intelligent", self-controlling implants or robotized limbs - whereby autonomously acting machine systems merge almost inseparably with humans. It is clear that in such a constellation it is no longer clear who controls whom: the human being the machine or vice versa?

The visionary potential of these developments is beyond question. Both the advances in robotics and neurotechnologies are accompanied on a societal level by far-reaching ideas about the future, which, interestingly, tie in with older, culturally extremely powerful motifs. [5] On the one hand there is the vision of a (neuro-) technological optimization of humans (human enhancement), [6] which first appeared at the beginning of the 20th century in the course of biological-evolutionary thinking. [7] On the other hand, there is currently intense discussion about the possibility that robots could outstrip humans in terms of intelligence in the not too distant future and either subjugate them or lead them into a glorious future. This idea also goes back at least a hundred years and was first formulated in Karel Čapek's dystopian drama "R.U.R. - Rossum’s Universal Robots" (1920). Since then, these visionary debates have been driven and fueled to a large extent by high-profile science fiction, in which different varieties of these visions of delimitation are strikingly portrayed. [8]

The fact that hybrids of humans and machines are obviously no longer a mere vision of the future is already clear from the fact that around 30,000 people in Germany currently have a cochlear implant. Nonetheless, the question arises to what extent the aforementioned futuristic visions are suitable as authoritative reference points for the social debate about the dynamic of demarcation. To cut a long story short: on the basis of a sober assessment of the performance of the underlying technologies, the TAB study comes to the conclusion that greater doubts are in order. [9] The established neurotechnological interventions in the brain are sometimes associated with considerable health risks that only appear ethically justified to compensate for severe physical deficits. The main argument against broader use in healthy people is that the technologically achievable additional skills (e.g. sensory enhancements such as hearing ultrasound using a cochlear implant) do not promise any real benefit in the foreseeable future, as would be necessary for optimizing people. Service robotics is also currently in a stage of development that banishes any thought of a "takeover" of intelligent machines in the realm of science fiction. The autonomy of the available systems is still extremely limited and is already being put to the test by simple manipulation tasks - for example recognizing and grasping common household items under real conditions. When - and if at all - a strong AI can be realized that equals human intelligence is a question that can still only be speculated about.