On a normal day, humans encounter Artificial Intelligence (AI) numerous times. Artificial intelligence has become a part of human's routine in many ways. They enter our lives in the form of smartphones, appliances in our homes and technology in our cars. They aid humans in different aspects starting from making appointments to diagnose illness.

Since humans are at the verge of accepting robotics into society, the question that lingers in everyone's mind is 'Can robots be trusted?' Human has a mythical illustration of robots turning aggressive once they are provided with all the features of humans. We are pushed to such conclusions through movies, soap operas and dramas. So keeping all these confusing thoughts away, we should focus on the ways human-robot relationship could reach the milestone of trust.

Explainable artificial intelligence

Explainable artificial intelligence is a branch of artificial intelligence research that examines the ways an artificial agent could be presented in a more transparent and trustworthy form. Human-robot relationship building and leading it to a stage of trust is really important in the technology world. It is a mandatory move when humans start working with robots. Robots seek to develop systems that attract human attention and at the same time, it should perform well to fulfil the designated task.

A team at Center for Vision, Cognition, Learning, and Autonomy in the University of California, Los Angeles is researching on the factors that make machines more trustworthy. It is also looking into different algorithms that provoke building trust in humans. The lab is using a model knowledge representation that could interpret the surroundings and make decisions which are easily understandable by humans. This naturally improves trust in humans citing the clear explanation and transparency.

In recent research, the team experimented different ways a robot could explain its action to a human observer and which among them made humans feel more trustworthy about robots. Remarkably, the form of explanation that improved trust did not correspond to the learning algorithms that produced the best performance. The robot's performance and explanation are not proportionally dependent on each other. But optimising for one of them is not a good idea. So the team is focusing on building robots that take into account both good task performance and trustworthy explanation.

Unravelling the performance and explanation skills of robots

Educating a robot to accomplish a task: The team undertook a study to know learn how robots perform particular tasks and how people respond to the robot's explanation of the action.

The team taught a robot to learn from human demonstrations on how to open a medicine bottle with a safety lock. To make the robot scan and understand the moves, a person wore a tactile glove that recorded the poses, moves and forces involved in opening a bottle. The information was conveyed to the robot. The robot understood the moves in two ways: symbolic and haptic. Symbolic understanding in robot means the grasp of its representation in human action. Haptic is connected to body posture and motions during the moment.

As a primary step, the robot was made to learn a symbolic model that encodes the string of steps needed to complete the task of opening the bottle. Then, the robot learned a hectic model that features the robot to 'imagine' itself in the role of a human demonstrator. This move encourages the robot to predict the action a person would take while encountering a situation of opening a bottle. The robot also analyses the poses and force it needs to put in to finish the task.

Remarkably, the robot was able to achieve its best performance with the combination of symbolic and haptic process. With the help of symbolic representation, the robot was able to use its recorded knowledge for performing the task and real-time sensing and brought it to action through haptic by predicting the outcome.

Earning human trust through explanation: The robot is now aware of the actions it should make to accomplish a task. The next step would be explaining what it did to the humans to bring clarification and earn trust.

The robot has the ability to draw on its internal decisions and behaviour. The step-by-step description of robotic actions through symbolic models and the feeling and predictions through haptic model are unleashed. To make the experiment more lively and clear, the team has enabled the provision to make the robot do a write-up text of its actions. They wanted to know if the text summary is as effective as the rest.

The program involves 150 human participants, divided into four groups who were asked to observe the robot opening the medicine bottle. The types of explanations were divided as Symbolic, step-by-step, haptic expression of arm position and motion, text summary or symbolic and haptic together were given to the groups of people. One of the groups observed the robot opening the medicine bottle without getting any explanation.

The conclusion turned out to be that people who get to experience the robot explaining both symbolic and haptic process got to grow more trust in them. The other group that got explained by a text summary felt the same as the group that plainly watched the robot do its job. Both the process didn't seem to foster much of human trust.

Take away for the future: Through the research, it was unfolded that people are not putting their trust in robots that only do actions without explaining. So the robots need both symbolic and haptic components to soothe humans.

The research gave an outlook for the future of artificial intelligence and promotes the researches to focus equally on the robots establishments and its ability to explain its actions. Both performance and explanation are a must while building an artificial intelligence system. The human-machine relationship should be further shaped in order to move it further.

Making robots a part of everyday life is not a new idea. But to accomplish the move, humans need to put their belief on robots. Trusting robots through the explanation they provide for their actions is an important step towards enabling human and robots to a further level.