Human-in-the-loop optimization for artificial intelligence algorithms

Numerous organisations use artificial intelligence algorithm-based products in their different activities. These solutions help with a wide range of jobs, from operational task automation to augmentation-based strategic decision making. The users’ trust in the truth and fairness of a product’s outputs must be built before it can be completely integrated and embedded in an organization’s daily functioning. They would be burdened with more work if Artificial Intelligence (AI) products did not have this feature. A human-in-loop decision-making process is important for building confidence and producing a successful AI-powered solution. In this research, a novel interactive system was created to explore the behaviour of AI-powered products. When designing our framework, we considered the necessity of integrating a human-in-the-loop technique in the design stage, something that had been missed in prior research of a comparable scale. The proposed software can optimise and monitor the AI-powered product process and outputs to involve people directly in the optimisation loop to identify and avoid likely and diverse failures. The Local Interpretable Model-agnostic Explanations (LIME) heatmap was utilised to illustrate decision-making features and mistake details more effectively throughout the improvement phase. The literature highlights the need of taking these issues into account throughout the design stage of an AI-powered product. This article describes how a human-in-loop AI-powered product is created by combining technologies from the AI, risk management, and human-computation domains. The designed system is based on deep learning as its decision-making engine, LIME as its approach explanation module, and the human aspect of knowledge workers. For real-world applications, we show how the created system improves product dependability and understandability by using real data and benchmark datasets.