There is currently much discussion on how to ensure Artificial Intelligence(AI) applications do not lead to undesirable outcomes, particularly bias in algorithms that power recommendations such as who is a suitable job candidate. In most of these instances it is important to be able to establish what data the algorithm was trained on to ensure the data is representative as well as how the algorithm reaches its conclusions. Much of the proposed regulation around Artificial Intelligence(AI) – such as the European Commission’s ‘Proposal for a Regulation on a European approach for Artificial Intelligence’ – is focused on ensuring fairness and ‘explainability’ in Artificial Intelligence(AI) algorithms.
Robots are not currently used in situations that involve bias. The main concerns for robot manufacturers and users when it comes to Artificial Intelligence(AI) are safety and certification. Additionally, robot manufacturers using vision systems must be able to ensure that any data transferred from the vision system does not contravene data privacy laws. For example, by enabling the identification of a specific individual.
Because the output of an AI algorithm is not known in advance, users may be understandably cautious about the impact on safety. However, AI can never be used alone in a robot, as there are many other program layers needed to control aspects of the physical robot system – such as the movements of the robot’s axes. This means robot programmers have to determine which layer in the ‘stack’ has priority over others. Priority can be given to the hard-coded, deterministic layer responsible for actions such as ensuring a robot stops if 10cm away from any object, no matter what the AI algorithm would otherwise determine to be the best course of action. This is almost always the case in the robotic applications using AI in operation today. However, as applications advance – especially in applications where the robot is physically interacting with a person whose position and movement are not pre-determined or in unpredictable environments such as public spaces – we can expect to see AI employed in safety-critical components of the application.
Enabling more variance in the robot’s reaction to safety-critical situations could also offer productivity benefits in industrial environments. For example, if a mobile robot can identify that a worker is within a safety- critical zone for the robot, but on the other side of the robot arm, it could keep moving, whereas in a hard-coded application, it would always stop. Research is being directed at how to ensure safety in these situations. For example, some research focuses on ‘learning controllers’ which continuously process data to determine the robot’s next move, followed by processing of the data generated as a result of that action, in order to determine the next move.
Currently, ISO safety standards exist for industrial robots – including a technical specification for industrial robots used in collaborative applications – certain service robots and mobile robots. These standards enable robot manufacturers to certify their robot as inherently safe. Since the safety of a robot application is dependent on many other factors – it is the production cell, is the robot manipulating a part with sharp edges – users must carry out risk assessments of their applications and are responsible for the safety of their workers, governed by national health and safety laws. It is currently unclear how safety standards and regulations will evolve for scenarios where AI is used in safety-critical components of the robot application.
manufacturers and systems integrators to demonstrate that specific safety or other standards are met and, as a result, to differentiate themselves from non-certified competitors.
There is currently considerable debate and research on the possibility and benefits of software certification or verification for programs partly or entirely composed of AI algorithms. Certification relies on certainty that the desired outcome will be met, which is difficult in the case of AI where the outcome may not be known in advance, or possible to explain after the fact.
Beyond certification, understanding why an algorithm reached a decision, and what data was important in making the decision, could also support better re-usability of algorithms for other purposes. Again, research is being devoted to this. For example, it would be possible – but computationally expensive – to determine and monitor parameters for the neural network governing the robot’s motion, ensuring the robot’s actions remain within a pre-determined safety range, even when the precise action that the robot will take is not pre-determined or explainable.
Artificial intelligence(AI) opens up new possibilities for robotic automation, particularly in environments with high variability. In manufacturing, AI is enabling the automation of a number of tasks involving the picking, placing and manipulating of objects, from machine tending to assembly, that have previously only been possible to do manually. Employees are spared strenuous, heavy or unergonomic tasks. AI is also becoming well- established for robotic quality inspection.
AI-driven robotic applications bring greater efficiency to logistics and retail, enabling companies in these sectors to cope better with peaks in orders, high product variability and an often-unreliable labor supply.
Robots are increasingly making their mark in public domains, from hospitals to shopping malls. In future, AI will enable better interaction between robots and the people and objects they encounter. AI will also help drive robotic automation in sectors such as agriculture which have not previously adopted robots.
AI will help reduce the resources and cost required to program and re-task a robot, opening up the possibility of automation to many companies for which automation has not previously been economically viable.
However, it’s important to note that these changes will take time. There are significant advancements in enabling increasing generalizability in AI algorithms – such as in the types of objects the robot can recognize when performing a particular task. However, a complete robot application involves many more program components and interfaces that are specific not only to the task, but also to the broader, company-specific automation architecture. For each application to be commercially viable, the cost of automation has to be outweighed by productivity or other gains. In many cases it will therefore be years before developments in research labs gain widespread commercial adoption.
Artificial intelligence is attracting increasing scrutiny from regulators and advocacy groups, particularly regarding the question of ensuring the avoidance of bias in AI software. In robotics, the key issue is ensuring safety. In most commercial robot applications using AI today, the robot either does not come into contact with people, or uses deterministic algorithms that override any AI in order to protect humans. For example to ensuring that the robot stops if the distance from an object or person falls below a certain threshold. Nevertheless, efforts to create frameworks and models for ‘explainable AI’ in industrial applications are important, particularly in enabling certification of applications using AI algorithms.
The notion of ‘general artificial intelligence’ in which a robot would be able to apply learning from one task – such as opening a door – to another – such as shutting a cupboard – without further input, as a human would be able to, is unrealistic in commercial environments. As it is observed, a robot application must be tailored not only to a specific task, but also to the specific environment. The AI algorithms used are therefore ‘narrow AI’ and the application will always need to be programmed, even if it re-uses portions of other applications or existing blocks of code.
However, programming is becoming easier, faster, and more intuitive. We can expect strides in reducing the overall time, and skills level, needed to create a tailored robot application, or re-task an existing one. AI will play an increasingly important role in enabling faster application development and re-tasking, and will expand the range of tasks it is economically viable for a robot to perform. This opens up the prospect of automation to new industry sectors and to many small-to-medium-sized companies.
OBSERVERTIMES GLOBAL NEWSNETWORK PRIVATE LIMITED reserves the rights to all content contained within its official website www.observertimes.in /Online Magazine/ Publications