Volume 2 (2022-23)

Each volume of Journal of AI, Robotics and Workplace Automation consists of four quarterly 100-page issues. Articles scheduled for Volume 2 are available to view on the 'Forthcoming content' page. 

The articles and case studies just published in Volume 2 are listed below:

Volume 2 Number 1

  • Editorial
    Christopher Johannessen, Editor, Journal of AI, Robotics & Workplace Automation
  • Papers:
    Artificial intelligence (AI) from a regulator’s perspective: The future of AI in central banking and financial services
    Melvin Lopez-Corleone, Senior Delivery Manager, Bank of England, Sholthana Begum, Head of Data and Strategy, Financial Conduct Authority, (FCA) and Gracie Sixuan Li, Innovation Associate, Bank of England

    Artificial intelligence (AI) is unlocking enormous opportunities. For central banks, AI has the potential to enhance regulatory efficiency and improve the data basis for monetary policy decisions. Machine learning (ML) can provide comprehensive, instant, granular information to complement existing macroeconomic indicators as well as having the capability to analyse big data efficiently, which can facilitate monetary policy decisions. As countries and companies conduct AI research and deploy the technology to the public, several financial authorities have recently begun developing frameworks, outlining their expectations on AI governance and use by financial institutions. This paper illustrates the current advancements in ML techniques and highlights the future trends in the adoption of AI by central banks and companies in financial services. It looks at the use of cloud computing and ML by companies and regulators to develop cost-efficient automation tools that better understand user needs, and presents how this will likely enable companies to adapt to rising trends in customer expectation in the future. The paper also explores the growing use of AI in anti-money laundering (AML) procedures, blockchain technology, and the development of Central Bank Digital Currencies (CBDC).
    Keywords: artificial intelligence (AI), CBDC, central banking, regulation, fintech, diversity and inclusion (DEI), COVID-19, machine learning (ML), data, Bank of England (BoE)

  • Using AI to minimise bias in an employee performance review
    Liz Melton, Strategic Partnerships Manager, Coco and Grant Riewe, Chief Technology Officer, Vibrant Emotional Health and Executive Fellow, University of St Thomas – Opus College of Business

    Performance reviews are intended to be objective, but all humans experience bias. While many companies opt for group reviews as a way to de-bias and challenge the status quo, what is being said in those meetings, how those comments are said and the context for those remarks are just as important. At the same time, most people’s attention span is of shorter duration than a review and being promoted depends on what bosses remember about their direct reports, their subjective measure of employee success, and their ability to convince others that employee accomplishments are deserving of a reward. As a result of these compounding factors, meta-bias patterns emerge in company culture. Combine those limitations with the fact that reviews are often a breeding ground for subtle — and not-so-subtle — bias, and it begs the question: Why are we not using technology to help? With developments in natural language processing (NLP) and conversational AI (CAI), computers can identify biased phrases in real time. Although these technologies have a long way to go to match human nuance, we can at least flag problematic phrases during something as significant as performance reviews. And with the right inputs rooted in social science and normalised based on geography, contextual relationships and culture, we could be surfacing insidious bias throughout organisations. This paper examines how a future CAI tool could reduce bias and, eventually, teach people to re-evaluate and reframe their thinking. In a performance review setting, the system would flag problematic phrases as they are said, and committee heads would stop the conversation. The committee would then evaluate the comment, ask the presenter for further information, and only continue once there is sufficient clarity. Once the discussion concludes, the review cycle would continue until another phrase is identified. The system serves to be persistently aware throughout all conversations and highlight potential bias for everyone to learn from. Beyond pointing out biased phrases during a performance review, a combination of NLP and CAI can serve as a foundation for company-wide analytics. Organisations can track who is speaking in a majority of meetings, what was said, who challenges biased phrases, whether or not certain types of people are misrepresented in reviews more or less frequently, and so on. All this information gives a fundamentally new picture of what is happening inside a company, laying the groundwork for human resource (HR)-related metrics that individuals (and the company as a whole) can improve over time.
    Keywords: bias, bias detection tool, bias detection system, AI, performance evaluations, review process, performance reviews, performance review

  • Human–machine collaboration in transcription
    Corey Miller, ASR Research Manager and Migüel Jetté, Vice President of AI, Rev and Dan Kokotov, Chief Technology Officer, CNaught

    As automatic speech recognition (ASR) has improved, it has become a viable tool for content transcription. Prior to the use of ASR for this task, content transcription was achieved through human effort alone. Despite improvements, ASR performance is as yet imperfect, especially in more challenging conditions (eg multiple speakers, noise, nonstandard accents). Given this, a promising way forward is a human-in-the-loop (HIL) approach. This contribution describes our work with HIL ASR on the transcription task. Traditionally, ASR performance has been measured using word error rate (WER). This measure may not be sufficient to describe the full set of errors that a speech-to-text (STT) pipeline designed for transcription can make, such as those involving capitalisation, punctuation, and inverse text normalisation (ITN). It is therefore the case that improved WER does not always lead to increased productivity, and the inclusion of ASR in HIL may adversely affect productivity if it contains too many errors. Rev.com provides a convenient laboratory to explore these questions. Originally, the company provided transcriptions of audio and video content executed solely by humans (known as Revvers). More recently, ASR was introduced in an HIL workflow where Revvers postedited an ASR first draft. We provide an analysis of the interaction between metrics of ASR accuracy and the productivity of our 72,000+ Revvers transcribing more than 15,000 hours of media every week. To do this, we utilise two measures of transcriptionist productivity: transcriber real time factor (RTF) and words per minute (WPM). Through our work, we hope to focus attention on the human productivity and quality of experience (QoE) aspects of improvements in ASR and related technologies. Given the broad scope of content transcription applications and the still elusive objective of perfect machine performance, keeping the human in the loop in both practice and mind is critical.
    Keywords: speech recognition, transcription, accuracy, productivity, artificial intelligence, postediting

  • High-impact AI: How to achieve business goals while making the world a better place
    Michael Griffin, Chief Data Scientist, Insight, et al.

    A vast amount of wealth has been generated by the explosive growth of artificial intelligence (AI)-enabled applications, and that wealth generation is accelerating. This paper describes how both monetary and non-monetary wealth — the kind of wealth that makes the world a better place — can be generated by the same intelligent application. The paper introduces the term ‘high-impact AI’, which is the practice of achieving business goals while simultaneously benefiting society. Novel and practical examples of achieving this lofty goal are described in the areas of manufacturing, where AI-powered software enables people with disabilities to perform jobs they would otherwise not be qualified to perform; healthcare, where the incredible complexity of practising medicine has been simplified to potentially improve the health and prolong the lives of millions of people; and psychology, where a tremendous variety of human interactions can be quantified and optimised in real time to assist strengthening and enhancing relationships of all kinds. Finally, this paper makes the case that practising high-impact AI will help businesses attract and retain talented people by giving them meaningful work that matters.
    Keywords: ethical artificial intelligence (AI), assisted communication, manufacturing, healthcare, psychology, enhanced order fulfilment, automated inventorying

  • Improved credit default prediction using machine learning and its impact on risk-weighted assets of banks
    Martin Neisen, Partner and Petr Geraskin, Senior Manager, PricewaterhouseCoopers

    The use of risk models, especially credit risk models, has been a standard for banks for many years. Banks use models not only for business decision purposes but also for regulatory purposes when comparing their risk with the available regulatory capital. Furthermore, banks need to efficiently allocate their capital in the current competitive and regulatory environments. As part of this process, they develop models to predict the probability of default (PD), which are further used to calculate risk-weighted assets (RWA). This paper gives an overview of how banks calculate RWA for credit risk. We compare the performance of traditional PD models based on logistic regression with a machine learning (ML) algorithm based on gradient boosting. This shows that an improvement in PD model performance by using ML algorithms can also lead to a decrease of RWA, therefore releasing additional capital for the banks. We developed and calibrated PD models based on logistic regression and light gradient boosting machine (GBM) approaches and compared them in terms of discriminatory power and the impact on RWA to prove this statement. This paper shows that the use of ML leads in our case study to an improvement in the model’s discriminatory power of 5 per cent in terms of Gini and releasing RWA of approximately 6.5 per cent.
    Keywords: Basel IV, artificial intelligence, machine learning, credit risk, banking regulation

  • Six ways in which AI is delivering value for organisations drawing on real-world case studies and offering actionable insights
    Nitin Mittal, US AI Co-leader, Deloitte Consulting and Irfan Saif, US AI Co-leader, Deloitte Risk & Financial Advisory

    After decades as science fiction fantasy, artificial intelligence (AI) has made the leap to practical reality and is quickly becoming a competitive necessity. Yet, amid the current frenzy of AI advancement and adoption, many leaders and decision makers still have significant questions about what AI can actually do for their businesses. This paper highlights compelling, business-ready use cases for AI across six major industries: consumer; energy, resources and industrials; financial services; government and public services; life sciences and health care; and technology, media and telecommunications. The goal is to give readers a much clearer sense of what AI is capable of achieving in a business context — now, and over the next several years — so that business leaders and AI practitioners can make smart decisions about when, where and how to deploy AI within their own organisations and how much time, money and attention they should be investing in it today.
    Keywords: artificial intelligence (AI), machine learning (ML), innovation, automation, technology adoption

  • On the predictability of long-term stock market returns: Design configuration of deep neural networks
    Manfred Herdt, Research Assistant and Hermann Schulte-Mattler, Professor, Dortmund University of Applied Sciences and Arts

    In 1998, Robert J. Shiller and John Y. Campbell proposed that long-term stock market returns are not random walks and can be predicted by a valuation measure called the cyclically adjusted price-to-earnings (CAPE) ratio. This paper is set to identify the predictive power of long-term stock market returns with deep neural networks and trace the impact of different architectural components of deep neural networks. We present three network types — recurrent neural network (RNN), long short-term memory (LSTM) neural network, and gated recurrent units (GRU) neural network — to ascertain what impact the different networks have on predicting long-term stock market returns and whether a parsimonious neural network model (PNNM) can be identified for practical application. The networks above have different design features that allow returns to be predicted and the effects of the various elements of the networks to be understood. For our study, we use monthly CAPE ratios and real ten-year annualised excess returns of the S&P 500 from 1881-01 to 2012-06, with data from 1876-06 (real earnings) to 2022-06 (real total return price) needed to determine the two datasets. Our results show improved forecasting accuracy over linear regression for all analysed neural networks. Only the complex trial-and-error procedure leads to the network design with the optimal result of minimising the root-mean-squared error (RMSE). This approach is usually associated with a considerable time and cost factor. Therefore, for time series studies of the present type, we propose a parsimonious GRU architecture with low complexity and comparatively low out-of-sample error, which we call ‘GRU-101010’.
    Keywords: cyclically adjusted price-to-earnings (CAPE) ratio, gated recurrent units (GRU) neural network, long short-term memory (LSTM) neural network, neural network’s architecture, neural network’s hyperparameters, recurrent neural network (RNN), time series analysis.