Follow Datanami:
April 11, 2023

Harnessing Hybrid Intelligence: Balancing AI Models and Human Expertise for Optimal Performance

Vahe Andonians


Measuring metrics on AI models is important because it enables stakeholders to gain valuable insights into how well their algorithms achieve desired outcomes and uncovers potential biases, limitations, and areas for improvement. This process facilitates the iterative refinement of models leading to more accurate and fair predictions; moreover, consistent and transparent assessment of AI models fosters trust and credibility. Measuring metrics is an essential practice underpinning the ongoing development and deployment of responsible, high-quality AI systems.

The four key metrics to monitor AI models for classification tasks, such as extracting information from unstructured documents, are recall, precision, F1-score, and accuracy. Recall, also known as sensitivity, is the fraction of relevant instances the model correctly identifies among all relevant instances. On the other hand, precision is the fraction of relevant instances among the instances predicted as relevant by the model.

AI models can be optimized for either recall or precision by employing various techniques during the development process. Tuning hyperparameters, such as learning rate, regularization strength, and the number of layers or neurons in a neural network, can impact the balance between recall and precision. Other strategies such as adjusting the training data or using specific techniques during training can also help optimize for recall or precision. Techniques such as oversampling the minority class, undersampling the majority class, or using synthetic data generation can help balance class representation and impact the model’s performance.

Adjusting the decision threshold is the most commonly used technique because it is simple, intuitive, and can be applied to various models and scenarios with minimal adjustments. By default, many classification models use a threshold of 0.5 to determine class membership. However, this threshold can be adjusted to optimize either recall or precision. Lowering the threshold increases recall at the expense of precision while raising the threshold improves precision but may lower recall.

An ideal precision-recall curve would have a high precision value at all levels of recall, resulting in a curve that touches the top-right corner of the plot. In practice, however, there is often a trade-off between precision and recall, and the curve will have a more complex shape.

Hybrid Intelligence

The combination of AI and humans plays a crucial role in enhancing the effectiveness and reliability of outcomes by harnessing the complementary strengths of both parties. AI models can process vast amounts of data quickly and efficiently, uncovering hidden patterns and making predictions that would be impossible or time-consuming for humans. Humans possess domain knowledge, critical thinking, creativity, and intuition, essential for making sense of complex situations and ensuring that AI-generated insights align with real-world contexts.

By working alongside AI models, humans can provide the necessary oversight and judgment to validate and refine AI outputs, resulting in more accurate decision-making and improved problem-solving.

Moreover, the collaboration between AI and humans is essential for addressing ethical concerns and ensuring that AI systems are aligned with human values. As AI models become increasingly integrated into various aspects of our lives, it is crucial to balance automation and human intervention to maintain transparency, fairness, and accountability.

In addition, the fusion of AI and human expertise paves the way for new opportunities in innovation and productivity. The synergy between AI’s data-driven insights and human creativity enables us to tackle complex challenges and develop novel solutions that would be difficult for either party to achieve independently. This collaboration fosters a dynamic environment that promotes continuous learning, adaptation, and growth, benefiting individual organizations and society.

Integrating AI models with human workflows requires careful consideration of recall and precision to create a harmonious work distribution between AI and humans. For example, suppose a model demonstrates high precision but lower recall. In that case, humans should be used to review and address the missed relevant instances rather than double checking found instances. Conversely, if a model has high recall but lower precision, it could be employed to cast a wider net, identifying potential relevant instances that humans can then verify for greater accuracy.

In the last six months, it has become clear that the power and value of AI will permeate virtually every software application and bottom line of every business. However, in order to optimize its positive impact, it is just as clear that defined metrics must be established in order to understand and fine tune the AI models to deliver desired outcomes. By selecting the right metrics, combined with driving real collaboration between AI and humans, we can realize the enormous potential of AI beginning today.

About the author: Vahe Andonians is the Founder, Chief Technology Officer, and Chief Product Officer of Cognaize. Vahe founded Cognaize to realize a vision of a world in which financial decisions are based on all data, structured and unstructured. As a serial entrepreneur, Vahe has founded several AI-based fintech firms and led them through successful exits and is a senior lecturer at the Frankfurt School of Finance & Management.

Related Items:

AI Is Coming for White-Collar Jobs, Too

How AI Boosts Human Expertise at Wolters Kluwer

Why AI Shouldn’t Be Deemed the ‘Workplace Enemy’