23 Jan 2017
Will AI replace humans in the development of AI software?
While the development of machine-learning and artificial intelligence technology, as of now, is done mostly by humans, it could soon cease to be so.
MIT, in a recent report, looked at various projects which are employing AI programs to develop machine-learning software, and the results are promising.
This shift, if it happens, will come with its own set of advantages and disadvantages.
Software-developed machine learning software showing promising results
Google Brain recently used AI software to develop a machine-learning program - in the benchmark test, the program surpassed results produced by its human-developed counterpart.
Other independent groups such as Elon Musk co-founded Open AI, University of Berkeley, MIT, Google DeepMind also reported considerable progress is software-developed machine-learning software.
All of the above points to an upcoming revolution in AI software development.
The advantages of software-designed machine learning programs
There is a serious dearth of human expertise in the sphere of AI software development.
As a result, companies pay massive amounts to sign such experts.
However, if self-starting AI programs become a reality, not only will the resource burden go down, but the pace at which machine-learning software is implemented in the economy will see a major jump.
The disadvantages of software-designed machine learning programs
Despite the promise in self-starting AI programs, it poses two major problems.
Firstly, as of now, software which beats its human counterparts at making machine-learning programs requires a massive amount of computational power, and hence is a costly endeavour.
Secondly, such a shift to software-designed software would enforce fears of AI replacing an increasing number of human jobs in various fields.
Eliminating the need for expertise in making AI software
"Currently the way you solve problems is you have expertise and data and computation. Can we eliminate the need for a lot of machine-learning expertise?," asked Jeff Dean, the leader of the Google Brain research group.
Using AI for mundane AI training could benefit human scientists
"Easing the burden [of training AI] on the data scientist is a big payoff. It could make you more productive, make you better models, and make you free to explore higher-level ideas," said Otkrist Gupta, a researcher at MIT Media Labs.