Google's AutoML project was set up to teach artificial intelligence (AI) programs to create other AIs and replicate itself. Google has now announced that the clone created by AutoML is actually performing better than the human-made program.
The new AI created by AutoML is faster, more efficient, and more powerful than even the best human-designed systems, reports Business Insider. The announcement was made after certain tasks were handed to both the human-programmed AI as well as AutoML and comparisons were made on how they performed.
AutoML achieved 82% accuracy while sorting photographs based on their content. If that seems simplistic, a task that requires the program to identify the placement of objects within the photograph was also given to the AIs. In this test, the human-built program scored 39%, while AutoML scored 43%.
The difference is significant because even a tech giant like Google does not have the required expertise to build the next generation AI. "Today these (AI programs) are handcrafted by machine learning scientists and literally only a few thousands of scientists around the world can do this," said Sundar Pichai as quoted by BI. "We want to enable hundreds of thousands of developers to be able to do it."
It is easier, notes the report, to pick up an existing AI and modify it to meet new demands and train it to do specific tasks than to build a new one. AutoML was built from the ground up and it is reported that building new complex systems will only become easier for systems. Humans' role in this scenario could be relegated to a gate-keeping one, says the report.
With the grunt work of programming systems handed over to AI, humans will have to refine and fine-tune it. The report points out AI could inadvertently make biased connections based on race, colour and gender identities. This is something that the human engineers will have to oversee.
It is not clear if Google will task AutoML to create its own AI and if it does, it raises the question of whether it will be smarter than its creator.