AIP human-centered automation is an approach to the development of balanced automation systems between a person and the software and technical environment, based on the methods and tools of the AIP Institute to maintain the skills of users in any conditions and extreme situations, and the development of human cognitive functions through continuous training.
The global phenomenon of total automation in connection with the development of the level of development of neural networks (AI) is more manifested in the direction of partial or complete replacement of human activity with autonomously executed functions (autonomous modules). In most cases, the control automation used is executed in the background without direct control by the operator, thereby eliminating the human factor from the system. This type of automation can be considered as completely autonomous [1].
In this case, a person is no more than one of the subjects taken into account in the automation system, including with the use of AI, along with animals or static objects, which occasionally take into account individual parameters (weight, dimensions, speed of movement). It is obvious that with this approach, there is an effect of excessive automation, which leads to the occurrence of emergencies.
A report by the National Transportation Safety Board (US National Transportation Safety Board, 2017, about Tesla) concluded that the use of ubiquitous automation does not necessarily lead to an improvement in human life. Accidents with autopilots are one of the clearest examples of the results of the use of automation “because we can”, without due consideration of the human factor. An untrained driver expects more from a system called “autopilot ” than it can give, as a result of which unforeseen accidents and errors occur.
Questions about the use of autopilots and the level of automation are quite acute in aviation. The publication of the organization Safety Alerts for Operators (SAFO) reflected concern about autopilots: “The constant use of automatic flight systems does not strengthen the knowledge and skills of the pilot when performing manual flights.” Thus, there is a growing problem with the training of pilots and the loss of proper qualifications due to relaxation and not using skills in flight. An increase in the dependence of pilots on autopilots leads to a weakening of their ability to fly without it and can lead to critical consequences in cases of unforeseen outages (source: “Autopilot: An Accident” by JPS Hawkeye).
In vital areas, such as transport (cars, trains, planes), medicine (surgery), and even maintenance of nuclear power plants, the introduction of fully autonomous systems without due consideration of the human factor leads to a decrease in the vigilance of operators and the loss of their qualifications, which in turn adversely affects the behavior of such specialists in critical situations when there are failures in the automation systems.
Automation systems implemented using neural network technology are a “black box”, whose decision-making algorithms are incomprehensible to users (source: IBM AI Guidelines) Despite the achievement of high efficiency in solving problems, there is a significant disadvantage in the widespread use of such systems – the inability to extract (extract) knowledge suitable for transmission and even more so for training the user.
In this regard, we see a promising path for the development of automation as a human-oriented system, where algorithms built around human knowledge, skills, and physical abilities take into account the balance between the necessary computer control for rapid automation and the human desire for improvement when necessary.
The purpose of the human-oriented approach in automation [2] is primarily to increase efficiency, expand the capabilities and compensate for the limitations of the operator, and not to replace the functions performed by a person with autonomous intelligent systems.
Using the example of vehicle automation in this paradigm, the software will adjust the system to the skills and physical condition of a person, being at the same time a backup assistant capable of intercepting control in the event of an incident. This solution gives you the freedom to practice and train the driver. The vehicle becomes both a simulator and an extension of the functional capabilities of a person. Such a symbiosis, while reducing the delay factor of data exchange between a person and an AI system, possibly even through implants, will allow us to realize the original goal of automation – expanding the capabilities and compensating for the limitations of the operator, and not freeing from any activity with a decrease in cognitive functions as a result.
Such human-oriented automation is implemented through the introduction of continuous assessment and training processes. The AIP Institute has launched and is improving several initiatives that jointly create an ecosystem for human-centered automation. These initiatives are also aimed at solving acute issues that arise when creating automation systems, for example, solving the problem of skill reduction, which undermines the human skills that may be required in case of automation failures, and the difficulty of maintaining vigilance when user actions become less frequent.
[1] Mitchell, C. M. (1996). Human-Centered Automation: A Philosophy, Some Design Tenets, and Related Research. In Human Interaction with Complex Systems (pp. 377-381). Springer, Boston, MA.
[2] Abbott, K. H., & Schutte, P. C. (1989). Human-centered automation and AI-Ideas, insights, and issues from the Intelligent Cockpit Aids research effort.