Put AI in the Human Loop
Martha G. Russell , Rama Akkiraju
The algorithmic age is here. Algorithms describe the mathematical relationships of chosen attributes and their values; they also describe the rules by which a sequence of specified actions will take place. Attributes, values, relationships, and thresholds are selected and brought together with rules to create algorithms that run on data. In this way, algorithmic models constitute a perspective on reality. They provide the artificial intelligence for risk analysis, outcome prediction, image analysis, natural language processing, and other routine and non-routine tasks in which statistical computations assess likelihood and predict outcomes. By their formulaic nature, algorithms represent an encapsulation of the real world, but not the whole; the encapsulation includes the biases inherent in the formula and the data, as well as in the model’s objectives. Because of this, algorithmic bias in AI is ever present.
Depending on the objectives of the model, some biases may support the model’s effectiveness. However, many biases, particularly the implicit ones, can be detrimental to the model as well as to the people affected by its outcomes. For example, companies are beginning to incorporate AI-based prediction models as decision support tools in many scenarios such as insurance and loan approvals, and college and job application screening. The growing reliance by human intelligence systems on data-driven models using AI signals an urgent need to understand biases in AI, educate current and future AI developers and users, and fine-tune the human oversight on algorithms for AI. With this perspective, we are putting AI into the networks, loops and relationships of humans, rather than keeping humans in the AI loops.
With the increase of statistical machine learning models in real-world business applications, concerns about the predictions made by biases are on the rise. Recent academic studies have documented discriminating biases in identifying people of certain race or color from various…
Martha Russell is Executive Director of mediaX at Stanford University and Senior Research Scholar with the Human Sciences Technology Advanced Research Institute at Stanford. Dr. Russell leads business alliances and interdisciplinary research for mediaX at Stanford University. With people and technology as the intersecting vectors. Russell’s background spans a range of business development, innovation and technology-transfer initiatives in information sciences, agriculture, communications, and microelectronics – for businesses, universities and regional development organizations. With a focus on the power of shared vision, Russell has developed planning/evaluation systems and consulted regionally and internationally on technology innovation for regional development.
Rama Akkiraju is a Director, Distinguished Engineer, and Master Inventor at IBM’s Watson Division where she leads the AI mission of enabling natural, personalized and compassionate conversations between computers and humans. In her career, Rama has worked on agent-based decision support systems, electronic market places, and semantic Web services, for which she led a World-Wide-Web (W3C) standard. Rama has co-authored 4 book chapters, and over 50 technical papers. Rama has over dozen issued patents and 20+ pending. She is the recipient of 3 best paper awards in AI and Operations Research.