Machines That Learn in the Wild: Machine learning capabilities, limitations and implications

This report explores the current capabilities and limitations of machine learning algorithms and looks at future challenges.

This report explores the current capabilities and limitations of machine learning algorithms and looks at future challenges.

Key findings

  • Machine learning has been very effective applied to specific problems like image recognition for medical diagnosis but to be used more in fields like medicine the human-machine interface needs to be improved.
  • Subjective decisions are still required to solve a problem with machine learning techniques and these decisions significantly affect the outcome.
  • ‘Crowdsourcing Analytics’ is an effective way to reveal what these subjective decisions are and create a consensus from several different approaches. Methodologies like this should be more widely used.
  • Two of the biggest challenges for the future will be developing the skills needed and effective regulation in such as fast moving field.

Machine learning algorithms are transforming many aspects of our lives and the biggest changes are yet to come. This report goes beyond the current hype to explore what these systems are actually capable of doing and what their limitations are. Learning algorithms are increasingly used in applications like medicine or driverless cars where failure could be lethal. The report looks at some of the key future challenges such as safeguards, regulation, skills and an overreliance on machine learning abilities. It addresses these and touches on innovation that may provide solutions.

Listen to the podcast

Authors

Harry Armstrong

Harry Armstrong

Harry Armstrong

Head of Technology Futures

Harry led Nesta’s futures and emerging technology work.

View profile