From Deep Learning''s Birth to the Divergence of Ideas
Four People Who Made the Same Revolution, Now Speaking of Different Futures
Geoffrey Hinton — "the man who created deep learning, now the one warning most strongly." Born 1947 in the UK; started from psychological interest in how the human brain learns; in the 1980-90s when rule-based AI was mainstream and neural networks were considered failures, Hinton never abandoned the path — developing backpropagation-based learning, Boltzmann machines, and distributed representations. 2012: AlexNet (with his students) won the ImageNet competition by an unprecedented margin, proving deep learning could transform AI. Left Google stating "to freely speak about AI''s dangers." Now warns about AI''s uncontrollability, misinformation generation, and possibility of surpassing human intelligence. Nobel Prize in Physics 2024. Yoshua Bengio — "the researcher who is most actively calling for caution." Focused on natural language understanding and sequence modeling; key contributions to recurrent neural networks, attention mechanisms (precursors to Transformer architecture). After ChatGPT emerged, became the most vocal advocate for AI safety regulation among the "Godfathers." Signed the 2023 "Pause Giant AI Experiments" open letter; testified before the Canadian parliament on AI risk; argues that AI development is accelerating faster than our ability to understand and control it. Yann LeCun — "the skeptic who believes AGI is farther than we think." Pioneered Convolutional Neural Networks (CNNs) enabling modern computer vision; Chief AI Scientist at Meta. Unlike Hinton and Bengio, does NOT believe current LLM approaches will lead to AGI — argues LLMs fundamentally lack understanding of physical reality, causality, and common sense; advocates for "world models" as the path to true AI intelligence. Publicly debates Hinton and others on AI risk — believes today''s AI systems are not as dangerous as feared. Fei-Fei Li — "the person who created ImageNet and is now asking about AI''s social responsibility." Created ImageNet (2009) — the massive labeled image dataset that made modern computer vision possible; this dataset enabled AlexNet''s breakthrough and indirectly launched the deep learning revolution. Founded the Stanford Human-Centered AI Institute (HAI); now focuses on ensuring AI benefits all of humanity rather than being captured by narrow technical or commercial interests; advocates for diversity in AI development teams as a prerequisite for AI that works for everyone. The four divergences: Hinton and Bengio see near-term AGI risk as real and urgent; LeCun sees AGI as fundamentally unsolved; all four agree on the need for AI safety research but disagree on the timeline and nature of the threat.
