Developing AI that takes human needs into account
While artificial intelligence (AI) has the potential to transform manufacturing processes through increased automation, it is critical that the human element is not forgotten. Humans are inevitably involved all along the production chain, and a synergistic relationship between robot and worker is essential to ensure smooth operations. “You cannot have AI behaving irrespective of human actions,” notes TEACHING (A computing toolkit for building efficient autonomous applications leveraging humanistic intelligence) project coordinator Davide Bacciu from the University of Pisa in Italy. “As humans, our reactions and well-being are influenced by our cognitive and psychological states.” In order for AI to boost operational efficiencies and lighten workloads along the production chain, it is vital that the introduction of AI does not place excessive pressures on humans.
Autonomous applications that empower humans
The EU-funded TEACHING project sought to address this challenge by developing autonomous applications that leverage human feedback. “We wanted the system to empower humans, and to be dependable and secure,” says Bacciu. To achieve this, the project brought together AI and machine learning specialists, as well as reliability engineers and software developers. “We wanted to develop safe and dependable applications in which AI is running, and then demonstrate the potential of this in end use applications,” adds Bacciu. The project used autonomous cars as one test case. Just as in manufacturing, a human-centric approach is needed here, to ensure the smooth handover and takeover between vehicle and user. A passenger’s stress levels and psychological state can greatly influence comfort when being autonomously driven. AI therefore needs to take into account not only the state of the vehicle, but also that of its passengers. “Our aim here was to try to personalise the service and ensure that AI could react to the user,” says Bacciu. In TEACHING’s model, data from sensors monitoring the physiological state of passengers is fed to the AI, which provides feedback to adapt the driving style of the autonomous car. “The idea was for the AI to be responsive, and eventually anticipatory of the specific user,” explains Bacciu.
Methodologies and models for distributed AI
This work enabled the TEACHING project team to successfully develop new methodologies and models for distributed AI. On top of this, the AI was shown to be able to constantly learn and adapt to the reactions of a specific user. The idea is that the AI is fully responsive to the needs of the human, and not just focused on its own production task. “We developed guidelines and a library for developers on how to structure this type of AI,” adds Bacciu. “This will make life simpler for those wanting to develop autonomous distributed AI applications.” While this technology holds potential in the autonomous car sector, Bacciu sees this as a medium- to long-term aspiration, given the regulatory and technological hurdles that must still be overcome. Of more immediate interest is the manufacturing sector. “There is major potential here because of the need for effective human-robot collaboration,” he remarks. “This kind of distributed AI, which takes account of humans and learns, would allow robots to operate autonomously, while being respectful of human needs.” In the long run, this could help to make the workplace safer and less stressful, and help manufacturers to make significant efficiency savings.
Keywords
TEACHING, AI, artificial intelligence, automation, manufacturing, robot
