There are two so-called AI winters, in which the enthusiasm for AI and support for AI declined because it did not live up to expectations.
AI Winter One - 1974-1980
Early AI research floundered. There was limited computing power for AI, and the personal computer revolution was starting to take hold. Government support was declining.
AI Winter Two - 1987-2000
Expert systems were inflexible and did not live up to expectations and commercial support was declining. The Internet and personal computer revolution were in full bloom.
How Did It Change?
AI began to emerge in the mid-2000s after NVIDIA developed the CUDA language that enabled its GPUs to perform parallel processing on any kind of data, not just 3D graphics. In addition, as the Internet was exploding, Google and other search engines were collecting huge amounts of data, all of which would later be used to train AI models. See
CUDA,
AlexNet and
AI in a nutshell.