Power and Prediction: the Disruptive Economics of Artificial Intelligence by Ajay Agrawal

Author:Ajay Agrawal
Language: eng
Format: epub
Publisher: Harvard Business Review Press
Published: 2022-11-15T00:00:00+00:00

Feedback Systems

Feedback loops are built deliberately. AIs that anticipate the value of feedback ensure that outcome data can be collected. In chapter 6, we discussed the system solution of an AI that predicts the best learning content for an individual on a particular day. This would personalize education, allowing students to move at the appropriate pace for them, and everyone to learn more. We discussed system-level challenges in terms of teacher allocation and social development. Feedback loops suggest the need for further system-level change. The AI requires data on whether the content improved student performance. The sooner the AI receives that data, the better. The challenge is how to design an academic curriculum to ensure that students deeply understand and remember concepts, while keeping feedback loops fast enough to improve the AI. This will require overcoming regulatory barriers to accessing student data combined with technological advances in protecting student privacy. Like the other parts of this system, the feedback part of the AI system solution for personalized education isn’t ready.

While point solution AIs generate a prediction, the power that comes from being an early mover with AI in an industry is a result of feedback. An AI must have access to outcome data in order to learn. An autonomous driving AI needs access to accidents. Every autonomous driving system would ensure that kind of feedback. Accidents, mercifully, are rare. To work well, an autonomous driving system would need access to near accidents. The more such near accidents, the faster it could learn. This requires a system of identifying when a near accident occurs and then building a learning process for avoiding such near accidents in the future. Avoiding accidents isn’t enough. Passenger comfort is also important, so an AI system solution that creates an advantage for early movers would also benefit from a way to measure comfort. AIs may, therefore, need to be designed to learn from, and weigh, multiple outcome measures.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.