mercredi 4 juin 2014

Will Transcendence Transcend Science Fiction?

A lire sur:

According to futurists, the Singularity will be the singular most important incident to ever unfold. The term describes technology’s version of a black hole: that incredible moment when artificial intelligence will surpass our own. Transcendence is the newest dystopian interpretation of this A.I. benchmark. The reviews have been less than glowing, but it still brings a pivotal question into sharp focus: Are we nearing true machine intelligence?
Transcendence tells the story of a dying technology visionary, played by Johnny Depp, who uploads his brain to a computer right before he dies. The story takes place in the present, compelling many to question the possibility of an apocalyptic singularity to transpire. While the technology depicted in Transcendence is still very much science fiction, it alludes to some remarkable developments that are unfolding in the artificial intelligence community.
The film centers on Dr. Caster, a lauded researcher who proselytizes about strong A.I.; a type of machine intelligence that can accomplish whatever a human can. In the past few years, strong AI has steadily gained a stronghold outside of the silver screen. One prominent milestone occurred in 2012, when Google X lab created a neural network composed of 16,000 computer processors. After exposing the network to 10 million image captures from YouTube videos, the system eventually taught itself to recognize cats. The researchers didn’t instruct the neural network to look for a specific object; according to researcher Jeff Dean, “it basically invented the concept of a cat,” based on the sheer volume of cat videos on the site. The amazing part of this experiment is that by immersing the machine in data, and forcing it to incessantly examine the pixels, it was able to learn what a cat was without an ounce of human knowledge.
The company Vicarious is taking a more Transcendence-like approach to advancing the field of artificial intelligence. They recently received funding to help them chip away at technology that emulates the human brain. More specifically, Vicarious is working on replicating the neo-cortex, the section of the brain that controls sensory information. Using computer vision, Vicarious was able to build software that could solve CAPTCHAs nearly 90% of the time; a task that even humans find difficult to do. CAPTCHAs are warped codes that help websites identify bots, and are an important method for calculating machine intelligence. The technology Vicarious is developing is quite similar to the computer vision aspect of PINN, the system that enables Dr. Caster’s wife to upload his brain in Transcendence. Vicarious’ belief that the human brain can unlock the computer brain may be A.I.’s missing link.
Another company contributing to a stronger kind of A.I. is DeepMind Technologies. Google recently acquired the company after they taught a computer to play vintage video games on an Atari console. DeepMind developed an algorithm that harnesses sensory information to help computers accomplish specific tasks. To build the algorithm, they used reinforcement learning, a type of machine learning steeped in neuroscience fundamentals. Similar to the YouTube cat experiment, the computer digested the pixels on the screen, and learned the rules based purely on the pixel movements, and not from human feedback. The system was never programmed to understand the game’s strategy; it reached that understanding all on its own. By analyzing the actions happening on-screen, it was able to construct a game plan on how to win.
What is so impressive about DeepMind’s algorithm is that it exhibited human-level reasoning. When we watch a game, we intuitively piece together the steps needed to win. Crystallizing what makes that intuitive for us is how we’ll be able to create machines that can think on their own.
Artificial intelligence is no longer an artifice. Transcendence will happen within our lifetime if we continue developing machines inspired by the human brain. According to computer science pioneer Carver Mead, “a fly’s brain doesn’t look that complicated, but it does stuff that we to this day can’t do. That’s telling you something.” Vicarious, DeepMind, and Google X lab are pushing this futuristic vision forward by tapping into neuroscience for a deeper understanding of human and machine learning.
An integral part of learning involves vision; infants only learn about gravity when they actually see an object fall. Similarly, all of these projects incorporate computer vision; understanding an image so well that the system can then respond with an appropriate action. These companies are taking baby steps in the right direction, but we still have a long road ahead of us. Despite what Transcendence says, until we build machines that can learn in an unsupervised way, they won’t be truly sentient. We still haven’t reached the technological black hole quite yet, but we’re standing on the precipice.

Timothy Tuttle is the founder & CEO of Expect Labs, a technology company that is building a platform to power a new generation of intelligent assistants.  

Aucun commentaire:

Enregistrer un commentaire