Attending the 4S in Boston, we also had time to visit the city and its unique research institutes. Andrea used the time to revisit her former colleagues at the Action Lab (Northeastern University), where she stayed during a research internship in 2010. Amir Farjadian, a former colleague, is now working on models for human-machine cognitive interaction under anomaly at the MIT Department of Mechanical Enginieering and facilitated contact to the MIT Media Lab. The MIT Media Lab, founded in 1980 through the efforts of Professor Nicholas Negropante and former MIT president Jerome Wiesner, is an interdisciplinary research laboratory at the Massachusetts Institute of Technology. The lab is working on projects at the convergence of technology, multimedia, sciences, art and design. On one of the last days of the 4S conference Otkrist Gupta, a Ph.D. candidate at MIT Media Lab, invited us to visit MIT Media Lab and gave us insights in his research on inventing new algorithms for deep learning for health screening and diagnosis.
Relating to his own research in the area of machine learning we discussed the rapid developments in the area of artificial neural networks that have taken place during the past couple of years. Every other week news of breakthroughs in this field are fueling the imaginary of a general artificial intelligence to emerge soon. Even though all current AI applications are domain specific, Otkrist voiced worries about the ethical and social challenges these technologies will pose in the future. A future, however, that nobody exactly knows when it is to become our present. But researching new approaches in machine learning at one of the world leading universities Otkrist, for sure, has a better understanding of what is currently under way than most people. For example, he shared insights from a project that he himself conducted together with Bowen Baker, Nikhil Naik and Ramesh Raskar developing a machine learning procedure that is able to design neural network architectures. Remarkably, they came up with a solution that outperforms human-made architectures (their approach is reported in “Designing Neutral Network Architectures Using Reinforcement Learning”). They essentially build a learning machine that is better at building other learning machines that human programmers are.
Against this background one indeed begins to wonder about what’s next to come in the areas of Artificial Intelligence, Machine Learning, Deep Neural Networks and the like. For STS and digital media studies this poses interesting challenges as well: If agency is never just located in solitary human actors, but distributed across complex networks of human and non-human actors, we need to ask if and how such technologies result in a redistribution of agency.