Teach Robots To Speak Plain System English
Like humans struggling to communicate during international collaborations, a significant language barrier divides humans and computers. The inability to casually converse with computers lowers the efficiency of the computer networks that run more and more of the world, and prevents new collaborations that could drastically increase the power of automated systems. The best way to improve these artificially intelligent devices will be through one common language, a University of Southampton study suggests.
By using a procedural language called sEnglish, or “system English,” computers can express their own “thoughts” more clearly to humans. Humans will then be able to respond accordingly, teaching computers rules, goals, values and other types of functionality that would be near-impossible with a standard programming language.
With sEnglish, robots could extensively describe their progress to humans, allowing us to trust that they know what they’re doing. That trust could in turn allow a larger number of untrained humans to deploy autonomous robots for various jobs.
“Practical applications of [sEnglish] are most important for autonomous machines and robots,” Veres told InnovationNewsDaily. “In traditional programming there are deeply entrenched conventions that are difficult to change. However, in the emerging market of autonomous robots [sEnglish] can enable the development of machines that do conceptual reasoning like us humans in a limited world.”
System English reads like a combination of MATLAB and Standard English, with keywords representing different thoughts or ideas that the computer can parse and absorb. For example, one chunk of sEnglish reads, “An ‘environmental map’ has the following properties: its ‘terrain model’ that is a geometric feature, its ‘objects’ that is a set of environmental objects and its ‘object locations’ that is a cell array.”
Not exactly Shakespeare. But while proper grammar clearly isn’t paramount, the sEnglish language still follows certain explicit guidelines.
“In an ordinary computer language, basic types such as numbers, character strings, structures and objects are used in a sequence of instructions and flow control commands,” Veres said. “In [sEnglish], each instruction is directly linked to a conceptual structure of human thought using a sentence.”
Veres says that he and his team will release a version of system English to the public within the next few weeks.
Published: InnovationNewsDaily.