Sign In

Communications of the ACM

ACM TechNews

Researchers Are Peering Inside Computer Brains. What They've Found Will Surprise You

View as: Print Mobile App Share:
Artist's conception of a computer brain.

Researchers at research firm OpenAI have developed advanced methods for peering into the inner workings of neural networks, helping to make their notoriously opaque decision-making more interpretable.


Researchers at artificial intelligence (AI) research company OpenAI developed new techniques to examine the inner workings of neural networks to help interpret their decision-making.

As neuroscientists have found in studies of the human brain, the researchers found individual neurons in a large neural network used to identify and categorize images can encode a particular concept.

This finding is important given the challenges of understanding the rationale behind decisions made by neural networks.

The researchers used reverse-engineering techniques to determine what most activated a particular artificial neuron.

Among other things, the researchers identified a bias that could enable someone to trick the AI into making incorrect identifications.

Said OpenAI's Gabriel Goh, "I think you definitely see a lot of stereotyping in the model."

From Fortune
View Full Article


Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA


No entries found