Sign In

Communications of the ACM

ACM TechNews

Is Reliable Artificial Intelligence Possible?


Artist's representation of artificial intelligence.

Researchers at the Swiss Federal Institute of Technology in Lausanne say artificial intelligence technology should be openly available.

Credit: ktsimage

Artificial intelligence (AI) technology should be openly available, according to researchers at the Swiss Federal Institute of Technology in Lausanne (EPFL).

"We have to hold AI accountable, and the only way to do this is to verify it for biases and make sure there is no deliberate misinformation," which is not possible if the AI is privatized, according to EPFL researcher Marcel Salathe.

Last year, EPFL researchers created an algorithm to recognize plant diseases, and Stanford University researchers recently showed AI can be trained to recognize skin cancer. Although these diagnostic tools use datasets of images to train and learn, the datasets can be influenced to prevent deep-learning algorithms from classifying images correctly.

In addition, deep neural networks are highly vulnerable to visual perturbations that cause the AI to misclassify images. These vulnerabilities highlight the importance of certifying AI technology and monitoring its reliability.

From Swiss Federal Institute of Technology in Lausanne
View Full Article

 

Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


 

No entries found

Read CACM in a free mobile app!
Access the latest issue, plus archived issues and more
ACM Logo
  • ACM CACM apps available for iPad, iPhone and iPod Touch, and Android platforms
  • ACM Digital Library apps available for iOS, Android, and Windows devices
  • Download an app and sign in to it with your ACM Web Account
Find the app for your mobile device
ACM DL Logo