Robots won't be able to wrest control of the planet from us silly humans until they learn how to collaborate. Sure, they can mow the lawn or mix a drink, but only when you give 'em explicit instructions. Luckily for our future robot overlords, The EU's JAST project is studying the ways that humans work together, in the hope that it can someday teach robots to anticipate the actions and intentions of a human partner. "In our experiments the robot is not observing to learn a task," explains Wolfram Erlhagen from the University of Minho. "The JAST robots already know the task, but they observe behavior, map it against the task, and quickly learn to anticipate [partner actions] or spot errors when the partner does not follow the correct or expected procedure." This bad boy has a neural architecture that mimics what happens when two people interact, and the video below shows the rather melancholy automaton trying to convince his human partner to pick up the right pieces to complete a simple task. Watch it in action after the break.

Researchers at Aberystwyth University in Wales have developed a robot that is being heralded as the first machine to have discovered new scientific knowledge independently of a human operator. Named Adam, the device has already identified the role of several genes in yeast cells, and has the ability to plan further experiments to test its own hypotheses. Ross King, from the university's computer science department, remarked that the robot is meant to take care of the tedious aspects of the scientific method, freeing up human scientists for "more advanced experiments." Across the pond at Cornell, researchers have developed a computer that can find established laws in the natural world -- without any prior scientific knowledge. According to PhysOrg, they've tested the AI on "simple mechanical systems" and plan on applying it to more complex problems in areas such as biology to cosmology where there are mountains of data to be poured through. It sure is nice to hear about robots doing something helpful for a change.

[Thanks, bo3of]

Read: Robo-scientist's first findings
Read: Being Isaac Newton: Computer derives natural laws from raw data

You know, when armchair futurists (and jive talkin' bloggists) make note of some of the scary new tech making the rounds in defense circles these days it's one thing, but when the Doomsday Scenarios come from official channels, that's when we start to get nervous. According to a report published by the California State Polytechnic University (with data made available by the U.S. Navy's Office of Naval Research) the sheer scope of the military's various AI projects is so vast that it is impossible for anyone to fully understand exactly what's going on. "With hundreds of programmers working on millions of lines of code for a single war robot," says Patrick Lin, the chief compiler of the report, "no one has a clear understanding of what's going on, at a small scale, across the entire code base." And what we don't understand can eventually hunt us down and kill us. This isn't idle talk, either -- a software malfunction just last year caused US. Army robots to aim at friendly targets (fortunately, no shots were fired). The solution, Dr. Lin continues, is to teach robots "battlefield ethics... a warrior code." Of course, the government has had absolutely no problems with ethics over the years -- so programming its killer robots with some rudimentary values should prove relatively simple.

Researchers have long been trying to model actual brains in order to build a better computer "brain," and it looks like IBM is now getting a helping hand from none other than DARPA in its attempt to create one that it hopes will one day have the intelligence level of a cat. To that that somewhat unnerving end, DARPA is pouring $4.9 million into a project that'll include five universities and scientists of all stripes, who will work together to create an artificial brain that behaves like a real one right down to the neuron level. As the BBC reports, the researchers are describing this latest initiative as a "180 degree shift in perspective" from previous efforts, as they're now seeking an algorithm first and problems second, as opposed to starting with an objective and devising an algorithm to achieve it. As for DARPA's ultimate goal, well, that's still a bit of a mystery, though let's just say we won't be surprised if future robots start to become very easily distracted.

[Via Daily Tech, image courtesy Mack J, Truth and Beauty Bombs]

While some folks are considering taking the Turing test one step further and applying it to military robots, a group of researchers in the UK led by none other than would-be cyborg Kevin Warwick are doing their best to keep things as Turing intended and simply trying to fool some humans into thinking that the robot they're taking to is actually a person. Fortunately for us on the human side of the equation, they weren't quite successful, though one "robot" known as Elbot did get relatively close to the goal, fooling 25% of its human interrogators, which is just 5% off the mark set by Alan Turing. Each of the four other "artificial conversational entities" also managed to fool at least one of their questioners, though they eventually showed their true colors with random answers like "soup" when pressed as to what their job was.

Cognition Technologies' new Semantic Map lets computers -- and, conceivably, evil robots -- "understand" the English language in much the same way humans do, based on word tenses and context in a sentence. With this technology, a computer or search engine can understand virtually every word in the English language -- for a vocabulary about ten times that of a typical American college graduate. The system is already being employed in search engines, allowing people to ask questions in human-phrasing instead of unnatural, machine formatted word strings. Researchers say the ability to understand language is an important building block of the nascent Semantic Web, and will make the Replicants of the future extremely difficult to detect.