Computer Vision News - March 2017

Computer Vision News Computer Vision News RE•WORK Summits in S. Francisco 35 We would like this “Machine Intelligence” to be able to do many things for us. Is there any task that you prefer it would never learn to do? Nikhil: I would hate to see intelligent war machines. One powerful force that has kept the world relatively peaceful is the fact that both sides in a war literally have “skin in the game”. I dread to think of a world where one side can declare war without having to feel the human consequences of their actions on their own side. Gilles: As of now I think that we as a species are not prepared at all for a certain thing that a very intelligent machine could learn: self-awareness. Once this has been achieved, there might potentially be an intelligence explosion. It is very unclear what will happen to any form of life on our planet when this happens. Stacey: I think dystopian scenarios of machine conquest of mankind receive a lot of attention, and it’s tempting to speculate on how we could maintain control by preventing machines from doing certain tasks or learning certain skills - say, if machines never learn genuinely creative thinking or locomotion at human levels of agility and perceptual awareness, or never engage in global-scale strategic planning, then we don’t need to worry about them taking over the world. I think as they become truly intelligent, machines will inevitably learn all of these dangerous skills and eventually master many of the tasks we consider to be uniquely human. I would instead prefer they never learn how to replace active human engagement with reality and the natural world. This means that machines should not replace certain kinds of often-menial thought processes to the point of humans not needing to exercise - and therefore regressing in their capacities for - long- term planning, memory, sustained attention, structured problem-solving, and decision-making, which is already starting to happen as apps increasingly plan what we read, watch, eat, and do on weekends, who we interact with, what we think is interesting or cool and why, etc. At a larger scale, this means mindful design of products, and especially immersive virtual environments, such that humans are not completely absorbed into the artificial experience of the machine, such that they have final control of how they spend their time. This is the potential future that worries me the most - one where machine intelligence renders human intelligence obsolete. Events What we do NOT want machine learning to learn… War machines Self-awareness Creative thinking and global-scale strategic planning