Juergen Schmidhuber, a co-inventor of the LSTM and a long-standing figure in deep learning, has for years argued publicly that the field’s most celebrated researchers do not properly credit the pioneers who came before them. His best-known critique, published on his own IDSIA web page in June 2015, responds to an influential Nature review of deep learning by Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Schmidhuber contends the review violates scientific norms of attribution.
His annotated history makes a series of specific claims: that Alexey Ivakhnenko published working deep networks in the 1960s and is “the father of deep learning”; that backpropagation was developed by researchers including Kelley, Bryson, Linnainmaa, and Werbos before the names usually associated with it; that GPU-accelerated neural nets, max-pooling, attention, and unsupervised pretraining all have earlier originators than commonly cited. He frames the dispute as one of accurate historical record rather than personal grievance, and updates the page as new disputes arise.
The episode is a window into the sociology of a fast-moving field, where the question of who invented what carries real stakes, for prizes, reputations, and the historical narrative itself. Others have pushed back on Schmidhuber’s framing, and the disputes remain genuinely contested rather than settled.
Why business readers should care: the credit fights are a reminder that the tidy origin stories told in keynotes and press releases often flatten a messier, more distributed history, and that “who invented AI” rarely has a single clean answer.