Artificial intelligence (AI) is taking a fascinating and confusing new step, moving further and further away from human thought. Traditionally, AIs have been trained based on human knowledge, replicating our logic and mistakes. But a new generation of AI, like DeepMind’s AlphaGo model, is beginning to take a more experimental approach, promising to propel machine learning far beyond human capabilities.
GTA 6 reveals its loading times: what you will discover will shock you
The rise of self-taught AI
Remember AlphaGo from DeepMind? This AI marked a major breakthrough because it did not receive any human instructions or read any rules. She built her own understanding of the game of Go through a technique called self-playing reinforcement learning.
Instead of following a human model, AlphaGo experimented by trial and error, playing millions of virtual games, to learn from it. In just two years, she beat the European Go champion, and in 2017, she defeated the world’s number one player.
The limits of human thought
AlphaGo shook up our understanding of games by challenging masters like Lee Sedol with unexpected moves. DeepMind then applied similar models to other games, demonstrating that human emulation was not necessary to achieve impressive results.
Xbox One X TV Spot: Microsoft promises the most powerful console on the market
AlphaZero, heir to AlphaGo, outperformed AI models based on human thought like Deep Blue. These AIs do not know the game like a human could but develop their own logic which allows them to dominate their opponents.
The beginnings of a new era with OpenAI
OpenAI, with its new o1 model, is starting to move away from human thinking. Language models like ChatGPT have been trained on the body of human written knowledge, but struggle to provide precise facts, often subject to “hallucinations”.
Two men fall off a cliff while playing Pokémon Go
The o1 model, however, incorporates principles from AlphaGo, taking a trial-and-error approach to perfecting its responses. This could mark the beginning of AI that not only imitates, but redefines thinking and learning.
A journey to the ultimate truth
Embodied AIs, freed from the confines of human thought, could approach reality in an entirely new way. By exploring and interacting with the physical world, they could discover knowledge that humans would never have considered.
These AIs will not just follow the human scientific method, but will develop their own theories and understandings. It’s a fascinating but slightly worrying prospect, because they might cross intellectual boundaries that we can’t even conceive of.
| ???? Autonomy | AIs adopt learning strategies independent of humans. |
| ???? Evolution | AIs like o1 begin to develop their own logic and understanding. |
| ???? Future | Embodied AIs could transform our understanding of reality. |
The future of artificial intelligence
As these models advance, we must prepare for an era where AIs surpass not only our capabilities, but also our understanding of the world. OpenAI’s o1 AI, although it seems modest, represents a giant step towards this reality.
It offers us a glimpse of how these machines, which we created, could one day surpass us in everything. What will happen when they start applying their knowledge on a large scale?
The question remains open: how will our society react to this progress? Are we ready to accept a world where artificial intelligence, freed from our constraints, will redefine truth and knowledge?
