We are currently seeing impressive advances in AI using deep learning(Edwards 2015). This has dramatically improved the state-of-the-art in speech recognition, computer vision, natural language processing and a number of other domains. These improvements have come largely from using larger data sets, and deeper neural networks:

“Before, neural networks were not breaking records for recognizing continuous speech; they were not big enough.” Yann LeCun, quoted in (Edwards 2015)

Of coures[sic], more data and bigger neural networks means we need more processing power. As a result, GPUs are now frequently used to provide this processing power. However, being better able to recognize speech or identify objects has not lead to an improvement in deep learning itself. The deep learning algorithms have not improved themselves. Any improvements to the deep learning algorithms have been hard won by applying our own intelligence to their design.