Yes, it’s not a name that sticks easily, but it’s one you shouldn’t forget.
Jürgen Schmidhuber is a German computer scientist, widely recognized as one of the pioneers in the field of deep learning and artificial intelligence. His contributions have been fundamental to the development of artificial neural networks, and many of today’s AI advancements—including technologies like ChatGPT, machine translation, and speech recognition—trace their roots back to his research in the 1990s.
Just a few days ago, Jürgen gave a couple of similar talks—one at TED Talks and another at BBVA (the latter being a bit more comprehensive)—where he discusses the past, present, and future of AI. It’s a masterpiece. English / Spanish
The Vision of Jürgen Schmidhuber: Past, Present, and Future of Artificial Intelligence
In this session, he not only reviewed the history and advancements of AI but also discussed the crucial role it will play in the future, covering topics ranging from the evolution of machine learning to the social and ethical implications of AI.
The Evolution of a Dream
Since the 1970s, Schmidhuber was inspired by Einstein and physics, but he soon realized that his true ambition was to build an "artificial scientist," an AI capable of solving problems beyond human capabilities. Despite skepticism at the time, he devoted himself to researching artificial neural networks, and by the 1990s, he had laid the foundation for what we now recognize as modern AI.
Curiously, at that time, very few people were interested in his discoveries. However, the same algorithms he developed back then now power applications in our smartphones, such as voice recognition and automatic translation. The key to this shift has been the evolution of hardware, allowing AI to be a million times more efficient than in the 1990s. Every five years, computational capacity becomes ten times cheaper, explaining the current AI boom.
AI That Learns on Its Own: Beyond Imitation
One of the most fascinating aspects of his lecture was his explanation of AI systems that do not merely imitate humans but set their own objectives, much like babies exploring the world. In his lab, they have developed AI models that experiment and discover patterns independently, potentially accelerating technological advancements even further.
According to Schmidhuber, many AI models today simply replicate the data they are trained on. However, the future lies in autonomous AI systems that pose their own questions, generate experiments, and learn through discovery, much like human scientists.
AI and Its Impact on Civilization
The influence of AI goes beyond the tools we use daily. Schmidhuber predicts that, in the long term, AI will not only transform our civilization but will also expand into space, colonizing the universe with autonomous systems that will create even more advanced AI. Over time, physical resources will be used to develop self-replicating infrastructures, fundamentally changing our perception of civilization and progress.
AI and the Future of Jobs: What Should We Study?
One of the most interesting moments of the session was when a 12-year-old child asked Schmidhuber what he should study to prepare for the future. His response was clear: while AI can automate many tasks, there are still skills unique to humans that are difficult to replicate in machines. He pointed out that although AI already surpasses humans in digital tasks, such as playing chess or generating text, it remains clumsy in the physical world.
Schmidhuber recommended focusing on skills that AI has yet to master: manual labor, creativity, physical object manipulation, and social interaction. Professions such as electricians, mechanics, and other skilled trades will become increasingly valuable since no robot can currently perform these tasks with the same precision and flexibility as a human. He also emphasized the importance of education in mathematics and physics, as the world is fundamentally built on these principles.
Additionally, he highlighted that humans will continue to invent new professions centered on human interaction. Just as 40 years ago, no one imagined people making fortunes from video blogging on YouTube, new job opportunities will emerge in the future that we cannot yet foresee.
Current Applications of AI: Healthcare, Sustainability, and More
Beyond theoretical research, Schmidhuber highlighted AI’s practical advancements in various fields:
Healthcare: Since 2012, neural networks have outperformed human doctors in detecting cancer and cardiovascular diseases. Applications in diagnosis and treatment are expanding, and if global access to hospital data were available, we could identify correlations that would further improve medicine.
Climate Change: AI is already being used to predict droughts, enhance carbon capture, analyze air quality, and optimize energy use. Advanced models help detect wildfires or predict crop yields using satellite imagery.
Language Barriers: Real-time automatic translators now enable seamless communication between different languages without requiring people to learn them.
Education and Employment: While many office tasks will be automated, manual labor and roles requiring human interaction will remain valuable. Education should focus on areas where AI lacks advantages, such as physical work, creativity, and social engagement.
Risks and Ethics of AI
Schmidhuber also addressed the debate on AI risks. He believes that the biggest concern is not that AI will become conscious and rebel against humanity but rather how humans choose to use it. Today, AI is already being utilized in armed conflicts, with drones identifying targets with high precision. However, according to him, the most significant danger remains nuclear weapons, which have existed for decades and do not require AI to be devastating.
Another major issue is bias in AI. If a model is trained with biased data, it will learn and perpetuate those biases, as has already occurred in facial recognition systems and medical diagnostics.
The Future of AI: Meta-Learning and Self-Improvement
Looking ahead, Schmidhuber advocates for meta-learning, an approach where neural networks not only learn from data but also improve their own learning algorithms. This advancement could lead to a scenario where AI becomes exponentially smarter without human intervention.
Final Reflections
At the end of his lecture, Schmidhuber reflected on his legacy. While he is often called the "father of AI," he sees himself as just another piece in the evolution of civilization. AI will not merely be a tool but an autonomous form of intelligence that will extend its influence far beyond what we can imagine today.
Thirty years from now, we will look back and be amazed at how naive we were to think that ChatGPT and other current AI models represented the pinnacle of technology. Are we ready for what comes next?
Comments