Domain Overview

In the recent years Artificial Intelligence (AI) has become frequently present in proposed robotic solutions by introducing machine learning capabilities in complex applications. While AI is not yet close to reaching its full potential, as it continues to advance, the field of robotics will benefit alongside. Machine intelligence is progressing at an astounding rate, powered by “deep learning” algorithms.

Today’s AI programs can recognize faces and transcribe spoken sentences, can spot subtle financial fraud, find relevant web pages in response to ambiguous queries, map the best driving route to almost any destination, beat human grand-masters at chess, and translate between hundreds of languages. We’ve been also promised that self-driving cars, automated cancer diagnoses, house-cleaning robots and even automated scientific discovery are on the verge of emerging.

The lack of human-like capability in machines is underscored by recent cracks that have appeared in the foundations of the modern AI. While today’s programs are much more impressive than the systems of 20 or 30 years ago, a series of research studies have shown that deep-learning systems can be unreliable in decidedly un-human-like ways.

There are many examples demonstrating that the best AI programs can be unreliable when faced with situations that differ, even to a small degree, from what they have been trained on and for. The errors made by such systems range from harmless and humorous to potentially disastrous. Even more worrisome are recent demonstrations of the vulnerability of AI systems to so-called adversarial examples. In these, a malevolent hacker can make specific changes to images, sound waves or text documents that while imperceptible or irrelevant to humans will cause a program to make potentially catastrophic errors.

Our own understanding of the situations we encounter is grounded in broad, intuitive “common-sense knowledge” about how the world works, and about the goals, motivations and likely behavior of other living creatures, particularly other humans. Additionally, our understanding of the world relies on our core abilities to generalize what we know, to form abstract concepts, and to make analogies — in short, to flexibly adapt our concepts to new situations. Researchers have been experimenting for decades with methods for embedding A.I. systems with intuitive common sense and robust human-like generalization abilities, but there has been little progress to date.

AI programs that lack common sense and other key aspects of human understanding are increasingly being deployed for real-world applications. While some people are worried about “super-intelligent AI”, the most dangerous aspect of AI systems is that we will trust them too much and give them too much autonomy while not being fully aware of their limitations.

For a current AI system to learn, for example doing image recognition, it needs thousands, or millions of photos all labelled with the objects in the image. Starting by feeding an image into the system, a neural net would make a guess, for example, whether there’s a dog in the photo. If the computer guesses right, then the system reinforces the pathways that led to that guess, and over time, repeating this process millions of times, the neural network can get quite good at making accurate guesses. This framework for learning is “supervised” because it relies on labelled data, telling the computer each time whether it got the guess right or wrong.

This technique has been enormously successful for specific tasks, such as image recognition, but it relies heavily on having the right kind of data and focusing on a very specific, narrow task. Only within the past decade the technology has really proved its usefulness, but it is still early for figuring out how to structure different business processes and technical problems in a way that deep learning can tackle them. Using the image recognition example, current supervised deep learning-based AI tools will be able do tasks currently done by humans.

However, supervised learning is limited, and more work is done on “unsupervised learning” where the AI provides its interpretation of situations based on ‘trial-and-error’ repeated attempts to perform a desired task. Unsupervised learning is associated to learning without a teacher, and modelling the probability density of inputs. This can be done by the ‘cluster analysis’, a branch of Machine Learning that groups the data that has not been labelled, classified or categorized. Instead of responding to feedback, cluster analysis identifies commonalities in the data and reacts based on the presence or absence of such commonalities in each new piece of data.

Compared to supervised learning where training data is labeled with the appropriate classifications, models using unsupervised learning must learn relationships between elements in a data set and classify the raw data without help. This hunt for relationships can take many different algorithmic forms, but all models have the same goal of mimicking human logic by searching for indirect hidden structures, patterns or features to analyze new data.

In the context of robotic systems, AI is a methodology that allows robots to perform complex tasks that cannot be done based on the state-of-the-art know-how. AI provides the means of capturing user requirements that are continuously changing and control the robot in the presence of dynamic changes in the environment and in the presence of parametric uncertainties.

The robotics business is bound to be transformed to include applications where the AI plays a key role. The business development already focuses on applications where the interaction between machines and humans is merging to provide better services to the humankind. The emerging Intelligent Robots will eventually co-exist with humans in such a way that the humans would benefit extensively from their association with Intelligent Robotics.

A wide range of Intelligent Robots applications present significant business opportunities such as in the context of Intelligent Home, Intelligent Community, and Intelligent City that could be the basis for related product developments and business growth.

The domain of Intelligent Robots requires skills in Robot Design, Control Systems and Artificial Intelligence. The AI skills are critical as they link the AI with the Robotics in such a way that the new domain of Intelligent Robots can generate and provide new technology to advance business interests.