AI is limited and narrow

AI is specialized

But every piece of AI software, every neural network, still has to be programmed by humans. In supervised learning, we have to train the software with a mass of data. A spam filter is strong at recognizing spam, but it cannot drive a self-driving car. The software of such a car can stop in time for a deer crossing the road, but it cannot use you for a translation task. In fact, each piece of AI software has its own specialty. A lot of AI software is very "narrow AI" for that reason.

But AI is on an unstoppable march. Amazon, Netflix and Spotify give their users recommendations ("recommendations") based on the behavior of other visitors and on similarities between certain products. Self-driving cars predict the driving behavior of cars in their neighborhood to avoid accidents. Weather forecasts based on AI are more accurate than ever before. Digital assistants, such as Google Home and Amazon Alexa, understand (some) common questions and try to answer them meaningfully. Google Translate translates a little better every day.

Why is AI limited?

AI is limited and achieving Artificial General Intelligence (AGI) has not yet been realized for several reasons:

• Narrow AI (Limited AI):Current AI systems are mainly based on narrow AI, meaning they are designed to perform very specific tasks. These systems are optimized for one specific domain and have limited skills beyond that domain. They lack the general intelligence that allows humans to perform different tasks with flexibility and adaptability.
• Lack of contextual understanding:AI systems struggle to develop contextual understanding and world knowledge as human intelligence does. They can analyze data and recognize patterns, but often lack the deeper understanding of the meaning and broader context behind the data. This limits their ability to interpret situations and respond to complex scenarios.
  Limited learning capacity: While AI systems learn from large amounts of data, they are limited to what they have seen in their training data. They have difficulty applying knowledge outside their training domain or generalizing to new and unfamiliar situations. This limits their ability to learn and adapt to changing circumstances.
• Problem of causal reasoning:AI systems have difficulty with causal reasoning, that is, understanding cause-and-effect relationships. They can identify correlations in the data, but have difficulty understanding the actual causes behind a phenomenon. This limits their ability to reason, make predictions and make logical connections as humans can.
• Ethics and values:Achieving AGI also brings ethical and philosophical challenges. It is important to develop AI systems that are consistent with human values and moral principles. Defining these values and ensuring the responsible deployment of AGI is a complex task.
  Although significant progress has been made in AI research and applications, achieving AGI, where systems can reach the level of human general intelligence, remains a challenge that requires further research and development.
  

An ai that can perform multiple tasks: Gato, the first step towards AGI?

Gato was developed by Google Deepmind. It appears to be the first AI capable of learning and performing multiple tasks. Gato is a so-called deep neural network for a range of complex tasks. It can perform tasks such as engaging in dialogue, playing video games, controlling a robotic arm to stack blocks, etc. 

According to The Independent, it is a "generalist agent" that can perform a huge range of complex tasks, from stacking blocks to writing poems."  The technology is described as "general purpose artificial intelligence" and a "step toward" artificial general intelligence.