AI Now calls for ban on affect recognition as market expected to surge to $90B by 2024
Science does not justify the use of biometric “affect recognition,” and the technology should be banned, according to researchers at the AI Now Institute at New York University, Reuters reports. The Electronic Privacy Information Center has also filed a complaint about affect recognition company HireVue, which was cited as an example of a problematic application of the technology by AI Now, with the U.S. Federal Trade Commission.
“Many job candidates have benefited from HireVue’s technology to help remove the very significant human bias in the existing hiring process,” company spokeswoman Kim Paone told Reuters.
AI Now’s 2019 report suggests that affect recognition is applied to job screening without accountability, and tends to favor privileged groups. Recent academic research is cited which shows multiple flaws in the methodology for interpreting moods from facial expressions.
The AI Now Institute makes 12 policy recommendations in the report, and concludes that with action from civil society, academic researchers, regulators and others, the future of AI can be determined by people. The groups co-founders said in a conference call prior to the release of the report that harmful uses of AI are increasing despite consensus on ethical principles because of a lack of consequences for violating those principles, according to Reuters.
Companies such as Microsoft and Amazon are marketing facial expression-reading technology, and the market for affective computing is expected to grow at a staggering 32.3 percent CAGR from $22.2 billion in 2019 to $90 billion in 2024, as AI emotion technologies are deployed for retail and healthcare applications and adoption of advanced electronic devices increases, according to a new report from Markets and Markets.
Facial feature extraction is expected to register the highest growth rate among software segments, and Asia-Pacific is expected to have the highest growth rate of any region during the forecast period, according to the “Affective Computing Market” report.
One attempt to deal with the limitations of facial expression analysis is offered by startup Akvelon, which uses similar technology to classify “attitudes” as positive, negative, or neutral, according to a Medium post by Daniel Diroff a data scientist and machine learning researcher with the company.
Akvelon demonstrated its technology at the recent AI and Big Data Expo in Santa Clara, California, with video processed with a deep convolutional neural network in real-time to make frequent judgements.
Diroff says that the technology utilizes a transfer learning technique with a ResNet variant, and was fine-tuned with Facebook’s PyTorch deep learning framework.