The creature stalked around the edge of the metal cylinder protruding from the ground. The cylinder had a smooth black sheen that seemed to absorb the light. As it completed a second circuit the creature had a sudden memory – the memory of a story it had been told when it was growing up. The story had described the descent of their species to this land many thousands of years ago – the creature had assumed it to be a fairy tale. As it paused next to a thin crack in the cylinder a flash of light played across its face. A second later a voice came from the cylinder. “Recognised.” The thin crack started to open.
Researchers at Google have created a large-scale free-form neural network that they have trained using still images from YouTube content. The training was not directed, yet the system still managed to adapt itself to respond strongly to the image of a cat. Suggesting that a lot of the training data had cats in it. Nice to see this kind of raw AI research taking place not just in academia. Read the blog post here: http://googleblog.blogspot.co.uk/2012/06/using-large-scale-brain-simulations-for.html
He shivered. The cold was starting to creep through him. He pushed at the concrete beam again, but it did not give. Thirst welled up inside him as he recalled the water cooler that had been just across the office before the building collapsed four days earlier. He tried to put it from his mind. Suddenly a low-powered drilling noise sounded above him. Dust started to fall on to his face, causing him to cough and raise his free arm against the downfall. When the dust stopped he moved his arm away. In front of him a white plastic shape was pulling itself through the opening. Once through it reconfigured itself into the shape of a humanoid. It popped open a compartment on its chest and retrieved a water bottle. Holding it in front of him the machine said, “Drink this.”
DARPA has announced a competition with a $2 million prize for robotics and software enthusiasts to build a machine that can be used to help out in disaster situations. The winning robot will need to perform a series of tasks including using tools and vehicles. See the press release here: http://www.darpa.mil/NewsEvents/Releases/2012/04/10.aspx Go to the challenge page here: https://www.fbo.gov/index?s=opportunity&mode=form&id=ee8e770bcfe1fe217472342c67d6bd5a&tab=core&_cview=0
The woman walked to the access port and pressed her hand against the panel. It flickered into life and displayed a large red X. She pressed her hand on the panel again, with the same result. Hearing a noise behind her she turned to watch the room’s entrance, but none of the creatures pursuing her had made it throught the maze yet. She pulled on the door, trying to force it open – but it stayed firmly shut. With a rising sense of desperation she pressed her hand on the panel once more. This time a question mark appeared and a voice crackled from the panel. “I’ll let you through if you win my game.” It said. Beneath her hand she saw a grid form and colourful icons appeared facing each other in ranks.
Michael Cook at Imperial College, London has developed an AI system ‘Angelina’ that can build complete computer games from scratch (minus the graphics and sound). It separates the components within the game and designs each in isolation, bringing them together to deliver a finished product. While not up to the standards of today’s immersive 3D games, the results are reminiscent of early PC games from the 80s. Read an article here: http://www.newscientist.com/article/mg21328554.900-ai-designs-its-own-video-game.html Try one of the games here: http://www.gamesbyangelina.org/
Slow at first and then with increasing speed the robot sprints across the harsh terrain. Its razor-like feet digging into the rugged surface propelling it forward. It knows how long it has to reach the waypoint and the consequences if it fails to get there in time. It assigns even more processing power to its pattern recognition algorithms, keen to identify any obstacles before they hinder its progress.
Everybody’s favourite funding source DARPA has release a video of a ‘cheetah’ robot that has just set the record for the fastest legged robot. It can run at an impressive 18mph, which would allow it to complete the 100m in under 13 seconds. I will keep a look out for it when the olympics kick off in London later this year. See a video and press release here: http://www.darpa.mil/NewsEvents/Releases/2012/03/05.aspx
Synthetic brain implants hold out the promise of augmenting the capabilities of the human mind as well as capturing and exporting that state of the mind in sci-fi tech such as the neural lace.
This possiblity is a step closer as shown by Professor Matti Mintz and team at Tel Aviv University. They have recently presented reults to show that a synthetic neural implant can be grafted into the brain of a rat to restore functions the rat had lost. Once implanted the rat proceeded to use the implant to learn new behaviour.
Although the most simple neural functions were targeted, it is a significant step forward and demonstrates that the science is a possibility.
Read the abstract of the paper here: http://www.sens.org/node/2210
Read coverage in International Business Times here: http://uk.ibtimes.com/articles/225407/20111005/robot-rat-robotic-brain-implant-restores-rat-s-mobility.htm
There have been numerous press articles about the conversation between two AI chatbots that has been set up by Igor Labutov at Cornell Creative Machines Lab. It connects two instances of the online chatbot Cleverbot.
The experiment of putting two chatbots up against each other is interesting, but I am hugely disappointed by the resultant conversation and the underlying algorithms that are revealed by the conversation. If this is AI – then it is using a very weak interpretation of the word artificial. While it may cause some amusement, I can’t help but feel that claiming these types of algorithms to be AI does an injustice to the ideals of machine intelligence.
You can see the press release and video here: http://creativemachines.cornell.edu/AI-vs-AI
Or browse some other interesting work done at Cornell Creative Machines Lab: http://creativemachines.cornell.edu/
Artificial intelligence is a field as wide as the endeavours of humankind itself. One of the best ways to handle such a complex subject is to find niche areas that can be investigated in depth. Then, at a later point, to abstract away from these areas to find patterns of common solution.
Speech recognition is one such area. Microsoft Research has recently announced progress in building a general-purpose speech recognition capability – which does not require the hours of training typically needed to adapt the software to your way of speaking.
Their approach, presented during Interspeech 2011, breaks down the problem to use the finer-grained yet more numerous senones, rather than the traditional phonemes. This has led to an improvement in word-recognition of more than 30%.
Read the article at Microsoft Research here: Speech Recognition Leaps Forward
IBM have made a lot of noise in the AI space, with their recent remarkable Watson victory – using a massive compute power and sophisticated algorithms to beat humans at natural language knowledge tasks.
Another of their projects has just received a $21m injection from everyone’s favourite future-tech funding source DARPA. The money supports the SyNAPSE project which links IBM staffers with a host of researchers at top universities.
I’ve dabbled in research in this area myself and fully appreciate the challenge and complexity of even trying to frame the problem correctly. I look forward to seeing the outcome of the SyNAPSE project – if not today, then at some point in the future.
If you want to check out the SyNAPSE project, then visit the IBM page:
SyNAPSE: A cognative computing project from IBM research
A novel use of the Internet to try to bring education about a classic science fiction plot construct – artificial intelligence. I’m sure we are a paradigm shift or two away from getting this right in reality, but the more people that have a solid gounding in the current techniques, the more likely that a breakthrough will come.
The class and registration link here: