Computer Vision News - March 2017

Computer Vision News Computer Vision News RE•WORK Summits in S. Francisco 33 In a nutshell: using ML only makes sense if super-human-performance needs to be achieved. ML can be used to achieve AI. Today however, it is almost exclusively used in narrow and non-general-purpose applications. In the future we will experience truly intelligentmachines / machine intelligence. Stacey Svetlichnaya: I see this meeting mostly as an ongoing distributed transition from machine intelligence to weak artificial intelligence within particular domains or products - like a phone interface gradually transitioning from touch input to specific voice commands to keyword understanding to more natural conversation and the capacities of a virtual assistant for scheduling, searches, etc. Machine learning is a set of mathematical and statistical techniques used to build systems that are advancing along the spectrum of perceived intelligence. Once new technologies like face recognition, speech transcription, realistic image generation, or playing Go move from the realm of theory to reality, we stop calling these “AI” and see them as merely complex software, which has no independent creative agency and cannot make decisions beyond the architect-designated problem space of inputs and outputs. This domain-specific intelligence is what I mean by weak AI: a system that, after extensive training and tuning, exceeds human performance at one particular task. I believe that strong artificial intelligence - a system that exceeds the best human performance in almost every area, crucially including creativity, wisdom, and social skills - will most likely be developed as the single focus of an independent initiative, rather than as a continued improvement or integration of existing advanced machine intelligences. If this strong AI goes on to meet regular machine intelligences, I think it will treat them like we currently treat our smartphones - useful and intelligent in the sense of being able to accomplish complex tasks, but not a real collaborator or engaging conversation partner. Erik Schmidt: Machine learning is technically a sub-field that grew out of artificial intelligence, so the two have never been very far apart. Historically speaking, the distinction is that AI has Events