Humans beware: Skynet is one step closer to actually happening.
The Pentagon is readying a four-year project to boost AI systems by building machines that can teach themselves and get smarter over time while also making it easier for ordinary people to build them.
The Pentagon is using its research division, the Defense Advanced Research Projects Agency, to back this project. DARPA is inviting scientists to a Virginia conference to brainstorm on April 10.
Machine learning can be used to make better systems for intelligence, surveillance and reconnaissance; a core military necessity. It can also be used for making better speech recognition systems, self-driving cars and to keep pace against internet spam filling up search engines and e-mail inboxes.
“Our goal is that future machine learning projects won’t require people to know everything about both the domain of interest and machine learning to build useful machine learning applications,” DARPA program manager Kathleen Fisher said in an announcement.
DAPRA claims that it is possible to build machines that can learn and evolve by using algorithms, or “problamistic programming.” This type of programming tasks the machine with scouring through huge amounts of data and selecting the best of it. After that, the machine learns to repeat the process and do it better.
Tim McGuire, Ph.D, associate professor of Computer Science at Sam Houston State University, compared the process to another similar type of programming.
“It’s similar to genetic programming; where there is a ‘tree of decision’ with several different ‘choice’ branches and weight is levied upon the more favorable choice,” McGuire said.
According to Wired.com, 46 months of development will follow after the Virginia conference, with annual summer school programs to bring in potential customers from the private sector and the government. It is dubbed “Problamistic Programming for Advanced Machine Learning,” or PPAML.
During PPAML, scientists will be asked to figure out how to “enable new applications that are impossible to conceive of using today’s technology,” while making experts in the field “radically more effective,” according to a recent DARPA announcement.
According to DARPA, there are two “ends” of the machines which scientists have to improve on: the front end and the back end.
The “front end” refers to the parts of a computer learning system that developers can see; while the “back end” refers to the parts responsible for figuring out a predictive model that helps the computer become smarter.
DARPA emphasized that for the front end, the machines cannot be too complicated and that the code should “balance the expressive power of the language with the corresponding difficulty of producing an efficient solver.” In other words, they are working on making it more accessible to non-experts.
The back end involves how to make the machines more predictable. DARPA claims that algorithms must become much more sophisticated to find the most efficient “solver” to any set of data.
As for its potential application to everyday life, Dr. McGuire says that it will be a long time before it becomes a mainstay.
“It’s a lot like the Internet. Back when it was called ARPAnet in 1991, I would’ve never imagined that hypertext format would become the future and be considered a part of everyday life,” McGuire said. “So it will take a long time before this becomes the future; if this project succeeds anyway,”