Smart AI

This Smart AI Can Now Steal Code to Write Its Own Pro­grams We’re not wor­ried, honest.

New links added!

Our arti­fi­cial intel­li­gence sys­tems are con­tin­u­ing to get smarter, with sci­en­tists demon­strat­ing that a sys­tem called Deep­Coder is now clever enough to bor­row bits of code from other pro­grams to solve basic prob­lems. But the team behind the tool don’t want to put human pro­gram­mers out of a job – they want to make it eas­ier for peo­ple to build pro­grams with­out any cod­ing knowl­edge. Deep­Coder is a project run by Microsoft and the Uni­ver­sity of Cam­bridge, using deep learn­ing tech­niques to mimic the neural net­work of a brain, where vast amounts of data are processed and eval­u­ated to make deci­sions. That means AI can think for itself with­out those choices being hard-​wired into the code to begin with, not unlike Google’s human-​beating Go sys­tem. “All of a sud­den peo­ple could be so much more pro­duc­tive,” MIT’s Armando Solar-​Lezama, who wasn’t involved in the work, told New Sci­en­tist. “They could build sys­tems that it [would be] impos­si­ble to build before.” Given a list of inputs and out­puts, Deep­Coder can fig­ure out which pieces of code will bring about the desired result, and its devel­op­ers say the sys­tem could poten­tially enable peo­ple to sim­ply describe an idea for a pro­gram to cre­ate. Human coders will already be famil­iar with bor­row­ing bits of other pub­licly avail­able pro­grams, but a sys­tem like Deep­Coder can search much faster and think much more quickly that pre­vi­ous AI sys­tems. It could also com­bine bits of code in ways that humans haven’t even thought of yet. And rather than run­ning each bit of code through a trial and error process – as ear­lier ver­sions of the sys­tem did – Deep­Coder can now pre­dict in advance which frag­ments are going to be use­ful and which aren’t. The sys­tem is even able to get smarter as it goes, learn­ing from its choices. That means we could soon be pro­duc­ing pro­grams that spot bugs in soft­ware or iden­tify objects in pho­tos much faster – and all with­out human coders hav­ing to lift a fin­ger. Want a pro­gram that moves all your pho­tos from one com­puter to another, resiz­ing them along the way? One day, you might just have to ask. We should point out that the tech­nol­ogy still needs to be tested by inde­pen­dent researchers, as it’s yet to be pub­lished in a peer-​reviewed jour­nal. But the team is being con­ser­v­a­tive about its claims regard­ing its cur­rent capa­bil­i­ties, so this looks like the real deal. For now, Deep­Coder can only solve sim­ple pro­gram­ming prob­lems that require around five lines of code, but the researchers are hop­ing to scale up the sys­tem and make it eas­ier to oper­ate for non-​coders in the future. “Gen­er­at­ing a really big piece of code in one shot is hard, and poten­tially unre­al­is­tic,” says Solar-​Lezama. “But really big pieces of code are built by putting together lots of lit­tle pieces of code.” A paper on the sys­tem is cur­rently under review for pre­sen­ta­tion at the 2017 Inter­na­tional Con­fer­ence on Learn­ing Representations.

You may also want to view this!

And this

And this, if you really want to know a lot more

Add com­ment

User Guide is located at here. There you can also find tags for adding media to your comment


Security code
Refresh

joomla tem­platesfree joomla tem­platestem­plate joomla
2017 Genet­icMem­ory glob­bers joomla tem­plate