A Guide to Real World Artificial Intelligence & Machine Learning Use Cases

Machine learning and artificial intelligence are driving major changes in the global economy.

AI & Machine Learning Will Drive Industry 4.0

This article looks at the ways in which firms across the various sectors of the economy adopt Artificial Intelligence (AI) techniques. However, before we review the sectors affected it is important to note the underlying drivers that are fuelling the growth in the influence and reach of Machine Learning across the sectors of the economy will only grow as we move forwards. This is because Big Data is only getting larger, velocity of data faster, plus the availability of cheaper data storage plus the arrival of powerful Graphical Processing Units (GPUs) to enable Deep Learning algorithms to be deployed. Furthermore, new research in areas of Deep Learning and other Machine Learning areas will continue to emerge into real world production over the next few years leading to new opportunities and applications.

The DLS team strongly believe that the advent of 5G around 2021 will be a transformative and revolutionary moment in human history. The enhanced speed of 5G over 4G will enable technologies, that struggle today with latency requirements, such as virtual reality and autonomous systems, to perform with real time efficiency.

This will be a world of intelligent Internet of Things (IoT) on the edge (meaning on the device) where the data is processed at the place where it is generated and Deep Learning models can run on the device itself rather than on a remote cloud server. This will obviate the need for an autonomous agent such as a robot or vehicle to wait receiving a response from a remote server before it can take an action.

We believe that this world will lead to the creation of new opportunities and businesses that do not exist today.  AI techniques such as Convolutional Neural Networks (CNNs), Deep Reinforcement Learning and Generative Adversarial Networks (GANs) will play an important role in this world alongside traditional Machine Learning methods (and including the rise of XGBoost, Light GBM and CatBoost). In particular Transformers with Self-Attention have been literally transforming Natural Language Processing as well as making an impact in Computer Vision with Vision Transformers (ViT), Time-Series, Chemistry and Biology. Furthermore exciting potential exists with emerging techniques such as Neurosymbolic AI (also known as Composite AI), Neuroevolution and Neural Circuit Policies.

In addition, there is an important role for Federated Learning with Differential Privacy to enable Machine Learning to scale across sectors where data is often siloed and (or) there is a need to ensure that data privacy is protected whilst also enabling Machine Learning to scale via collaborative learning. Federated Learning will be increasingly important in a world where standalone 5G networks scale and Edge Computing across the AIoT scales. It may also play a key role in the Metaverse with 5G enabled glasses whereby one may not want to share everything one is seeing and doing with a third party. Furthermore the same applies to the world of the smart home and the smart city as well as wearable devices. Again policy markers should note the important role that Federated Learning may play to protect data security and privacy whilst also allowing Machine Learning to scale across decentralised devices (including mobile phones). Frameworks such as Flower and OpenMined may play key roles to enable Federated Learning to become a mainstream approach in Machine Learning.

Furthermore, there is vast potential for Machine Learning to play a key role in mitigating the impact of Climate Change and enabling sustainable economic growth as shown by PWC and Rolnick et al (2019).

Sectors of The Economy Affected by Artificial Intelligence  

Financial Services: The use cases here include applications relating to Regulatory Technology such as leveraging face detection and verification for “Know Your Customer” from documents. This has often been powered by CNNs albeit Vision Transformers may potentially impact in this area too. 

Financial firms can also use AI for extracting text from financial documents such as company filings and documents with personal information such as driving licences and passports (Optical Character Recognition (OCR) with LSTM). 

In the investment management field, AI can be applied to passive portfolio ETF Index replication and tracking (Evolutionary Genetic Algorithms). Fraud detection is powered by CNNs and Gaussian Mixture models. In addition, Financial Services firms apply Machine Learning techniques to Credit Risk for lending and other counter party risk evaluation decisions with Linear Regression, Random Forests, Gradient Boosting and Neural Networks. Trader oversight and fraud detection maybe implemented using Variational Autoencoders (VAEs) for anomaly detection. In retail sector, there is a great deal of scope to apply AI to areas such as payments with face (using a CNN). Transformer models with Self-Attention are applied to sentiment and text analytics.

Insurance: AI has enormous potential in this sub-sector of Financial Services. Examples include automating the claims verification process whereby for example if the policy holder has made a claim for a car that was insured under the policy, then a CNN can used to verify that the make and model is that same as that in the insurance policy. Chatbots have been used as a virtual agent to converse with the end user. Furthermore, Variational Autoencoders can also be used for checking for the presence of outliers.

Retail Sector: This sector is one that is likely to be heavily impacted by AI and Machine Learning, both on the ecommerce and physical sides of the business. The ecommerce side generates large sets of unstructured data that can be used to generate meaningful insights that in turn enable personalisation and recommendation engines. Examples include usage of Collaborative Filtering for recommendations, CNNs for apparel detection and classification and Natural Language Processing (NLP) for mining descriptive text and next basket recommendations. Transformers with Self-Attention are making inroads with recommendation systems. Retailers have major cash costs with inventory and machine learning techniques can help them optimise the management of that inventory so as to reduce wastage and cost. Furthermore, areas such as visual search (using CNNs or ViTs) enable state of art personalisation with the customer. We are likely to see Virtual Stylists, powered by Deep Learning techniques combined with NLP, emerge to enable retailers to offer combinations to complete the look. Payment with face is likely to take off in the future too. Physical robots powered by AI (Deep Reinforcement Learning or Neuroevolution) will emerge to optimise supply chain management in the warehouses and maybe even in the future with automated delivery and customer service. 

Marketing: Much of this is covered in retail above. We will likely see marketing as the first sector to be transformed by AI as it has lower regulatory barriers relative to other sectors and large sets of data available. Hyper personalisation with customer segmentation and targeted marketing that is highly relevant to the user as the future of marketing. NLP techniques help us mine the vast amount of unstructured information present in social media for automated lead generation.

Health Care: This sector is likely to be the one most transformed by AI over time. Machine Learning will have major impact in areas such as medical imaging (CNNs and Vision Transformers), applying Transformer models to Electronic Health Care Data (NLP), Robotic Surgery using Deep Reinforcement Learning with 5G. Application of Machine Learning techniques to wearable devices allowing us to process sensor data so as to enable preventive health care. Furthermore, areas like precision medicine are powered by techniques such as Variational Autoencoders improving the outcomes for the patient. Drug Discovery and Regenerative Medicine significantly reduces the time take during clinical trials and shortness the time to market for novel drugs. As noted above Federated Learning with Differential Privacy will be an essential tool to scale AI across healthcare given the sensitivity and importance of data privacy and regulations such as HIPAA and GDPR. Moreover, AlphaFold 2 may fundamental transform the drug discovery sector. Liquid Machine Learning systems may also make a significant impact in healthcare by aiding decision making with data streams that vary over time.

Transportation: This is a sector that faces fundamental changes with autonomous cars and drones becoming a reality. 5G will enable connected cars to become an experience pod powered by Augmented Reality and Virtual Reality for those inside. Furthermore, areas such as Multi Agent Deep Reinforcement Learning will enable automated cars to cooperate with each other and the environment around them, thus providing a strong backbone for communication with well-defined standards. This will enable for safe autonomous driving. Also, autonomous drone technology will be able to provide logistical support to search and rescue teams in areas of remote or difficult terrain, and assist farmers in automated irrigation and harvesting. Developments in Neurosymbolic AI may also play a role in autonomous driving by allowing for what Suchan and Bhatt (2021) describe as "explainability, question-answering, and commonsense interpolation". Furthermore, the Neural Circuit Policy approach may have a major impact in this area. As Adam Zewe of MIT note that the Neural Circuit Policy built with Liquid Neural Networks are a type of "Artificial Intelligence agent can learn the cause-and-effect basis of a navigation task during training" and hence may be potentially signifiant in the field of autonomous driving.

Security: This is a sector that will greatly benefit from the adoption of machine learning techniques such as CNNs and ViTs for face detection and recognition, behavioural recognition in videos and assist in detecting suspicious activity in cyber security.  Interesting advances are being made with Multimodal Transformers applied to video for classification that may have future potential in this sector. Federated Learning may have significant potential in the field of Cyber Security in particular in the era of standalone 5G networks with device security being a major factor. Generative Adversarial Networks (GANs) also have potential with Cyber Security applications and also the potential for adverse security impacts with Deepfake technology being used for cyber crime related purposes.

Manufacturing and Industry: The manufacturing sector will be revolutionised by the arrival of Industry 4.0 that will give rise to autonomous agents such as robots that are capable of automated defect analysis and through application of Deep Reinforcement Learning will enable precision manufacturing at much higher speed and scale as well as efficiency gains in supply chain management. Potential may also exist with Deep Neuroevolution based approaches in manufacturing robotics.

The above is not exhaustive and it should be noted that others sectors such as travel and tourism will be heavily influenced as hotels, airliners and car hire companies all adopt various forms of machine learning. Agriculture will see an increasing adoption of autonomous systems. Education will also see the increased usage of AI for assisting teachers. The construction industry will be benefited with the deployment of robots for construction at scale and machine learning techniques assisting architects at the design stage.