The Clearview AI has a special role to play in AI research. Almost all of the literature and practical applications come with questions around the accuracy of a machine learning system’s ability to distinguish a certain, ambiguous picture based on the labeled data set. It has been our role in the AI labs of the human mind to do the studying and writing about the capabilities of AI-derived models, but the research must ultimately undergo a form of testing, at least in a more tangible form than commonly being performed via open-source algorithms or reports.
Now the Clearview AI has taken part in its first, large-scale accuracy test by taking part in an interoperability exercise designed to simulate a real-world scenario.
Shizuka Inoguchi of MIT and the Voice and Video Computation Lab at Rensselaer Polytechnic Institute designed the exercise. In it, they attempted to simulate an actual production-line setting in which people are responsible for selection and captioning of a series of distinctive images. The Clearview AI was designed with the express purpose of working with unencumbered data in which real objects – be they human, computer-generated, or any combination of both – could be accurately labelled and visualized.
See For Yourself
While this experiment doesn’t directly affect human life in any way, it is an example of how AI research has to move from abstract theory and imagining to dealing with the real world, where machines will inevitably be put to work. In the past few years, the AI development community has been celebrating unprecedented progress towards large-scale implementation of systems that are capable of remarkable abilities. As we move into a world where access to AI is at least equal to a lot of human-resourced and labor-intensive labour, these systems may have to be acceptable both to their users and to society as a whole.
While solving real world problems has been my goal since I moved to the scientific community from academia, I have also been interested in the ethics of our technology and its potential to impact human life and human civilisation. Now we will see what the AI community can do beyond conceptually tackling these issues. The Clearview AI is just one tool out of many – it is, however, an important one that can demonstrate its usefulness even when it has clear advantages over human-trained systems.
Shizuka Inoguchi of MIT and the Voice and Video Computation Lab at Rensselaer Polytechnic Institute created and led the creation of the Clearview AI.
Michael Dedmon, Director of Rensselaer’s Uridium Center for AI, and the Leïla Collier Department of Mathematics taught the Clearview AI.
This article contains references to K and a s a r i l s h l m t l h an h a i l h r i r as “due to the nature of the experiment.” Readers are strongly advised to consult these materials for the details of the experiment.