This week in artificial intelligence (AI) news, we take a look at the environmental impact of AI systems, a review of Stuart Russell’s new book on controlling AI, and Yoshua Bengio’s work developing an algorithm to help machines learn cause and effect.
When posing questions on the negative consequences that artificial intelligence (AI) may have on society, the economy, and culture, we generally think of questions like “Will robots take my job?” and “What are the costs of developing these new tools?” However, the concerns that need to start being raised are “How is AI impacting the environment?” and “What can we do to make AI systems more efficient in terms of the environment?” In a recent article from Slate, Roy Schwartz, a researcher studying deep learning models at the Allen Institute for Artificial Intelligence and at the University of Washington, spoke on the issue of energy consumption when it comes to developing, using, and, more importantly, training, AI systems. Schwartz elaborates on the importance of researchers reporting numbers to show how much energy has been consumed when running their programs. “Once people start reporting it, many people will be encouraged to work to improve to find greener or more efficient solutions,” stated Schwartz. Read the full interview by April Glaser on Slate.com by clicking here.
In “Human Compatible: Artificial Intelligence and the Problem of Control”, computer scientist and founder of the Center for Human-Compatible Artificial Intelligence (CHAI), Stuart Russell, warns of “the problem of control” and the misuse of AI, creating a somewhat dystopian outlook for the future. However, the problem is not with AI as we know it today, but “superintelligent” AI, which has the capacity, according to Russell, to surpass the general cognitive abilities of humans, thus having the capability of operating outside of our control. Read more on this critique of Stuart Russell’s new book from Nature.com by clicking on this link.
The development of deep learning techniques has led to advancements in areas like facial recognition, self-driving cars, and medical imaging. However, Yoshua Bengio, a professor at the University of Montreal and recent recipient of the Turing Award, says that deep learning needs to be fixed in order to unlock its full potential. Machine learning and deep learning systems are trained in highly-specific tasks, such as classification and voice recognition, but fall short when it comes to cause and effect relationships. The ability of AI systems to understand causality would lead to smarter machines and new possibilities. Read more about Yoshua Bengio’s quest to change the way machines learn in this article from Wired.com.