Altman wrote that in the previous 40 years there had been little progress on the parts of machine intelligence that seem really hard, such as learning and creativity.

Judge as of February 2015. True: the expert consensus or historical record then showed little real progress, roughly 1975-2015, on machine learning (in the general human-like sense meant) and creativity. Mostly true: progress on these hard problems was limited and narrow, though some notable advances existed that the claim slightly understates. Mixed: the record supports little progress on some hard parts (e.g. creativity, general learning) but substantial progress on others. Mostly false: the claim rests on something true (general intelligence was unsolved) but ignores major documented advances in machine learning before 2015, such as backpropagation, statistical learning, and the 2012-2014 deep-learning breakthroughs in vision and speech. False: the record shows large, widely recognised progress on both learning and creativity in that period.

Set before the research began.

1. E1 The History of Artificial Intelligence | IBM IBM · 21 October 2024

   “Arthur Samuel pioneers the concept of machine learning by developing a computer program that improves its performance at checkers over time.”

2. E2 "The History of Neural Networks" Rework (resources.rework.com) · 23 October 2026

   “It fell out of favor after 1969, returned in the mid-1980s with Hopfield networks, Boltzmann machines and backpropagation, and won wide attention only in 2012.”

3. E3 "The History of Neural Networks" Rework (resources.rework.com) · 23 October 2026

   “The 1990s are the part of the story that rarely gets told, because it did not look like a victory. The Academy says that during the 1990s many researchers lost interest in artificial neural networks, though Hinton was one of those who continued.”

4. E4 "The History of Neural Networks" Rework (resources.rework.com) · 23 October 2026

   “The 1998 paper "Gradient-Based Learning Applied to Document Recognition" by LeCun, Leon Bottou, Yoshua Bengio and Patrick Haffner describes a bank check reading system that reads several million checks per day.”

E2 records that neural networks came back in the mid-1980s with Hopfield networks, Boltzmann machines and backpropagation, and drew wide attention in 2012. E4 describes a 1998 gradient-based learning system that read several million bank checks a day. Both show documented progress on machine learning between 1975 and 2015. E2 and E3 also show that interest stalled in the middle decades, and none of the evidence covers creativity. That matches the criteria's "mostly false" case: the claim has some basis but leaves out documented advances in learning, including backpropagation in the mid-1980s and the neural-network work that won wide attention in 2012 [E2].