'Godfather of AI’ Geoffrey Hinton quits Google
Warns over the misuse of AI
Geoffrey Hinton, the pioneer of artificial intelligence who helped build the foundations of modern AI, has left Google with a dire warning about the dangers that AI technology poses.
Geoffrey Hinton is widely considered the 'godfather of artificial intelligence' for his foundational contributions to the field of deep learning and neural networks.
Hinton worked at Google for over a decade and developed technologies that led to current AI systems like Google’s Bard, and OpenAI’s ChatGPT. But he decided to leave the company so he could raise the alarm about AI's risks.
He told The New York Times that he regrets the work he contributed to the field. "I console myself with the normal excuse: If I hadn't done it, someone else would have," he said.
In the short term, he fears AI will make it impossible to know what's true, as fake images, videos and text spread everywhere, he warned.
Yet in the future, AI could learn dangerous behaviors on its own, conceivably powering killer robots. He also warned AI could severely disrupt the job market.
"The idea that this stuff could actually get smarter than people - a few people believed that," he said. "But most people thought it was way off. And I thought it was way off. I thought it was 30 to 50 years or even longer away. Obviously, I no longer think that."
He argued regulations are needed to ensure companies like Google and Microsoft don't race toward danger without control. Those companies might already be secretly building perilous systems, he suggested.
"I don't think they should scale this up more until they have understood whether they can control it," he said.
Dr. Hinton joins other AI experts’ warning of the "profound risks to society and humanity" that AI poses. In recent months, two major open letters signed by many who helped build AI sounded the alarm.
Like others, Dr. Hinton became more concerned about AI risks over the past year, as OpenAI's ChatGPT and Google's Bard emerged.
Google’s chief scientist, Jeff Dean, said in a statement that Google appreciated Hinton’s contributions to the company over the past decade.
“I’ve deeply enjoyed our many conversations over the years. I’ll miss him, and I wish him well!
While misinformation is Hinton’s primary immediate worry, he is also concerned about potential long-term impacts of AI. On a longer time horizon, Hinton fears that AI could displace many routine jobs and tasks. And possibly pose a challenge to humanity itself as AI systems become capable of advancing their own capabilities through recursive self-improvement and autonomous development. AI that can sustain its own progress could potentially accelerate beyond human comprehension or control.
Some of Geoffrey Hinton's most important contributions to AI include:
1. Helping popularize backpropagation and backprop-style neural network training algorithms. Backpropagation is the workhorse algorithm that allows us to train multi-layer neural networks, and it's fundamental to deep learning. Hinton helped develop efficient methods for training neural networks with backprop in the 1980s.
2. Co-inventing Boltzmann machines. Boltzmann machines are a type of neural network that helped revitalize interest in neural networks. They demonstrated that neural networks could learn interesting representations and model complex data. Boltzmann machines were a precursor to deeper models like deep belief networks.
3. Helping to launch the deep learning movement. In the mid-2000s, Hinton and his students developed techniques for pre-training and fine-tuning deep neural networks. This allowed much deeper networks to be trained than before. Their results on tasks like image classification and speech recognition helped renew interest in neural networks and drive progress in the field.
4. Pioneering new neural network techniques like dropouts, rectified linear units (ReLUs), and generative adversarial networks (GANs). These techniques have become foundational building blocks for neural networks. Dropouts help prevent overfitting, ReLUs help networks train faster, and GANs have been hugely influential for generative modeling and unsupervised learning.
5. Training many leading researchers in the field. As a professor, Hinton trained many students who went on to become world experts in deep learning and artificial intelligence. His "academic grandchildren" and beyond have built on his work and spread throughout the AI community. This dissemination of knowledge has been a huge driver of progress.
6. Helping to found successful AI companies like DNNresearch and DeepMind. Hinton has helped turn research progress into real-world impact by contributing to startups. DeepMind, in particular, has been hugely successful, creating systems like AlphaGo that have accelerated progress in reinforcement learning and game-playing AI.
So in many ways, Geoffrey Hinton is responsible for helping to lay the foundations of modern AI through his pioneering research, dissemination of knowledge, and real-world startup experience. His impact on the field is unparalleled.