Focus July 2020

Deep learning and deep neural networks Later, Dr Turkbeyler and the team began working successfully with deep learning and deep neural networks because there was now an enormous amount of data available from social media in the form of text and images. While huge advances in the processing power of hardware meant that they could train powerful new algorithms more quickly. Now, it wasn’t just a case of recognising faces in images or identifying objects - industries could use AI to detect impurities in images and reveal imperfections in production line components that are imperceptible to the human eye, reducing future re-work. Today’s applications Today, Dr Turkbeyler oversees a team of Data Scientists and AI specialists who apply AI in a huge range of fields, from satellite data, imagery, text data and natural language processing to cybersecurity. “AI, together with Intelligent chatbots and robotics, are making a big difference to companies and are a relatable example of how cutting-edge technology like Artificial Intelligence and Deep Learning drives Digital Transformation and changes the way companies operate. “There’s lots of hype about AI”, Dr Turkbeyler cautions. “But practical cases can vary widely, from helping companies gather and analyse data from sensors, images, files, audio, video and documents, and collating these into a complete operational picture to using machine learning to digitise technical drawings, documents. This is one of the first steps on the journey towards producing design digital twins. “It can be about helping businesses with small Digital Transformation projects, where they can explore how AI can remove the manual aspects of what people do, so that they can add value in other, more meaningful ways. In this context, human-machine teaming helps to build skills and confidence and drive the challenging cultural changes associated with Digital Transformation.” Algorithmic biases, trust and ethics Other commercial applications of AI include those that determine a person’s age or their sentiment from an image. In the rail industry. Using CCTV cameras to detect ‘destructive and unusual behaviours’ can help prevent incidents by alerting an operative that someone might be about to engage in risky behaviour. But, while these applications may set out to save lives, they do start to raise some very ethical questions. “At BMT, we’re particularly mindful of how biases can pervade each stage of the data journey, from the origination of the data itself, to those tasked with using it to programme an algorithm. The team pays attention to the training data given to the algorithm to learn, to ensure inputs are varied and unbiased. If left unchecked, these biases can reinforce, rather than correct, the systems and practices that discriminate against vulnerable communities on the basis of race, ethnicity or disability, among other factors.” One such example is the idea that future crime can be predicted based on police data of arrests and incarcerations, where it has been well-documented that these have disproportionately targeted the black community. In this age of Digital Transformation, it’s more vital than ever that our scientists and engineers are as neurologically, ethnically and linguistically aware and diverse as possible. AI in smart cities and disaster relief “On a more positive note, we see local government using AI to improve services by detecting graffiti and potholes, while utilities companies can optimise energy usage by predicting demand and regulating the amount of renewable energy supplied to the grid. “We’ve been able to use Deep Convolutional Neural Networks and Generative Adversarial Networks in Rapid AI Mapping project work. This turns satellite imagery into usable maps that allow geospatial analysts to automatically extract map features, like roads and buildings to support urgent disaster relief operations.” Biomimicry in AI and Swarm Technology “By mimicking nature and applying deep reinforced learning algorithms, our scientists have taught drones to perform a perched landing, as a bird would do and are now teaching a swarm of underwater vehicles to collect information collaboratively and return it for analysis”, Dr Turkbeyler shares. “This is something that would previously have been complex and expensive for human operators to do.” AI in design and engineering BMT research and development also extends into applying AI towards traditionally computationally complex problems like fluid dynamics. By using our wind speed data sourced from thousands of fluid dynamics projects, we are working towards shortcutting the often expensive and time-consuming task of scale model testing, reducing the number of design iterations needed. The future of AI And this brings Dr Turkbeyler to the future of AI. “Robotics and AI will keep making our daily lives easier”, she suggests, “and we won’t even notice it’s happening. But trust and ethics will remain paramount, a principle she describes as ‘explainability’. In other words, we shouldn’t use AI algorithms if we can’t explain how they came to their decisions. At the same time, less scrupulous groups are using AI technology for political and financial gain, whilst ignoring ethical standards, so this issue must be understood and addressed and testing and verification will dominate this field, especially for safety-critical systems. “So, whilst this is an interesting time to be working in this field” says Dr Turkbeyler, “our focus remains on helping our customers to shape a future in which AI is ethical, accountable, responsible and transparent.” 23 22