HANNOVER MESSE 2019, 01 - 05 April
switch to:
Machine Learning

Machine learning becomes smart through human intelligence

Researchers at Carnegie Mellon University have developed the AI assistant, Evorus. The voice assistant learns by analyzing answers given by humans.

27 Feb. 2018
Dirk Bongardt
HMI-ID02-040db_Workheld01
Machine learning becomes smart through human intelligence (Foto: Tablet Solutions GmbH)

Developers have high hopes for machine learning -based technologies. Intelligent production management or predictive maintenance systems are therefore also set to implement artificial intelligence and facilitate more cost-effective and productive production processes that are considerably more customizable. In addition to image processing – a basic feature of many ‘smart’ automation components – greater attention is also being paid to speech recognition. After all, for humans, speech is the number one interface of communication.

The clearer the meaning of an utterance, the better the speech control, as is the case with field management software WorkHeld from Vienna-based Tablet Solutions GmbH, for example. With this solution, the artificial intelligence provides technicians and engineers in the field with the appropriate installation instructions. Chatbot system Evorus , developed by Carnegie Mellon University in Pittsburgh, is aiming to go one step further – for the AI to answer questions without any human assistance. To achieve this, Evorus is first working together with humans, to learn from them. Whenever the artificial intelligence is unable to answer a question, it recruits human employees from A mazon Mechanical Turk , who provide appropriate answers and vote on the best answer in that particular case. Evorus learns from these answers and uses them later to answer subsequent questions independently. In the five months since Evorus began learning from the crowd, the costs for human helpers have already fallen by around a third.