What is Artificial General Intelligence?
Artificial General Intelligence (AGI for short) is the artificial general intelligence that enables a computer program to understand or learn any intellectual task that a human can perform. AGI can also be referred to as Strong AI or Full AI. Thus, computer programmes can have sentience, self-awareness or even consciousness. Expert systems, on the other hand, only deal with solutions to predetermined tasks.
Even in the early days of artificial intelligence (AI), scientists assume that it is possible for machines to think like humans. There is no known greater cognitive power than that of the human brain. So in many ways it makes sense to replicate it and thus achieve a high level of cognitive processing.
There is a debate about which is the best path to true Artificial General Intelligence. Great strides have been made in Deep Learning, but there are other avenues opening up. Artificial General Intelligence is an ultimate goal in the development of artificial intelligence.
AGI should be able to learn, reason, plan, also understand natural human language and of course show common sense. With AGI, the machine is able to think and learn similarly to a human. In this way, it should be possible to understand situational contexts and also to apply what was learned during the completion of a task to further tasks.
However, this has not really been achieved by machine intelligence so far. Even small children are able to apply things they learn in one environment in a way to completely new tasks. Unfortunately, even the most complex AI-controlled machines have not yet managed to do this. Researchers are, however, working on this problem. Many different approaches are being pursued and the main focus is on the Deep Learning with the aim of reproducing intelligence.
Neural networks are a recognised state of the art technique for learning correlations in training datasets. With Reinforcement Learning the machine can be taught to easily figure out on its own how to complete tasks with given rules. Generative Adversarial Networks (GAN) can enable computers to take more creative approaches to problem solving. Challenges are the Image recognition and natural language understanding. Knowledge graphs can contextualise data.
What is Artificial General Intelligence capable of?
Artificial general intelligence is able to use knowledge-based presentations, natural language communication and the use of strategies, and to assess and sense irregularities and act. This enables AGI systems to recognise and respond to threats. They should support cognitive research and exhibit mathematics-oriented intelligence and implement decision-making.
What are the application examples?
There is the Robotics as an application area with drones, cars and submarines that are supposed to learn in real time and can adapt to new situations. AI systems can detect money laundering, terrorist financing and fraud in the financial industry. AI systems that can play chess, go or poker at world-class level are also conceivable. AI systems that can detect accidents, wrong-way drivers and pedestrians are also useful for real-time road monitoring. AI systems that can understand written documents, answer emails and pay bills would also be important. AI systems that can make forecasts about possible burglaries, planned attacks and general crime are also necessary.
Which are known systems with AGI?
The following systems are relatively new and of great importance for a digital and sovereign Europe:
- Aleph Alpha
- OpenAI
- Deepmind
