[This article was originally published in the OTT Annual Review 2019-2020: think tanks and technology on March 2020.]
Today’s remarkable advances in artificial intelligence (AI) are powered in many ways by the billions of images uploaded to the Internet each day. This rich corpus of imagery and video has enabled industrial and academic researchers to develop models that excel at detecting specific objects or identifying visual themes of photos. These innovations have led to meaningful increases in human welfare. For example, AI systems will soon increase the speed of MRI scans by more than ten times, significantly expanding access to this important health diagnostic tool.
However, the potential of AI systems to benefit humankind is limited in many ways, because the world is not composed exclusively of 2D static images. On the contrary, people perceive the world through continuous streams of information captured in the first-person, or an ‘egocentric’ point of view. Moreover, our world consists not just of visual information, but images mixed with audio and complex signals captured via social context, body pose, or other interactions. AI technologies are currently unable to effectively build on these first-person perspectives.
The AI research community is, however, undertaking a number of bold new initiatives to develop computer vision models that can effectively advance real-world perception. Habitat AI, for example, provides a first-of-its-kind platform for training virtual robots in simulation environments, enabling a cheaper and faster advancement of units that can navigate real-world locations or enhance daily life. Similarly, EPIC Kitchens is a unique effort to advance machine perception through first-person video. It’s enabling the development of AI that can understand people’s interactions with objects, what they intend to do with them, and their attention span for specific tasks.
The tech industry is taking notice of these efforts and developing its own initiatives to generate data and benchmarks that enhance or augment first-person perspectives. Each of these undertakings are part of an emerging field known as ‘embodied AI’, which brings together researchers from robotics, vision, audio, and augmented reality. The new energy around this embodied research challenge stands as today’s cutting-edge for AI research.
While daunting, progress against the challenge of embodied AI will rapidly accelerate, and it will not be long before research breakthroughs lead to significant advancements in autonomous driving, advanced robotics, or augmented reality systems that can be housed in everyday, wearable devices. These innovations are poised to provide users around the world with guidance and personalised assistance as we manage daily life, advance our ability to monitor health, enforce laws, or provide professional training to underserved populations. Vuzix is an early leader in these efforts with a suite of augmented reality devices for manufacturing, logistics, and remote health services.
Before these technologies are deployed en masse, the policy research community should work now to identify mechanisms to support and manage this emerging field of embodied AI. At the moment, however, few think tanks have begun to grapple with how these innovations may change civic life.
As the speed of technology outpaces policy frameworks, this lack of attention will leave governments playing catch-up and constituencies without guidance or policy recourse. Moreover, without public action, technology innovators may miss opportunities to build embodied AI for today’s most urgent development challenges, choosing instead to pursue tasks with immediately-viable business models.
Policy research institutions should embark now on efforts to inform the forthcoming debate on embodied AI technologies. While governments have begun to articulate new standards for data privacy, media integrity, and the regulation of technologies like facial recognition, few have undertaken a thorough assessment of this new direction. Think tanks can bridge this gap and prepare the groundwork for effective policy solutions.
Such an initiative should include thorough assessments of how embodied AI might impact civic life, taking into account considerations including inequality, privacy, and safety. Early efforts to educate policymakers about these emerging technologies are critical as the lack of tech expertise in the public sector has encumbered effective government action. Efforts are also needed to begin to define policy frameworks that can balance the deployment of socially-beneficial embodied AI, continued scientific advancement, and inclusive economic growth.
The development of embodied AI stands as a frontier challenge for both technologists and the think tank community. Policy research institutions in particular should work now to inform public officials and debates as this new technology emerges and begins to enhance daily life.