Personal tools

Narrow AI (ANI)

University of Toronto_050922A
[University of Toronto]


- Overview

Artificial Narrow Intelligence (ANI), also known as "weak" artificial intelligence, is the artificial intelligence that exists in the world today. Narrow AI is AI that is programmed to perform a single task. It is a type of machine intelligence limited to a specific or narrow area. 

Weak AI simulates human cognition and benefits humans by automating time-consuming tasks and analyzing data in ways humans sometimes cannot. Many existing systems that claim to use "artificial intelligence" today may be operating as a weak AI, focusing on a narrowly defined problem. A better definition of weak AI is "augmented intelligence," which makes it clear that the technology is here to complement the human mind, not replace it.

 

- Narrow AI Systems and Examples

Narrow (or weak) AI systems do not have general intelligence; they have specific intelligence. An AI that's good at telling you how to drive from point A to point B usually won't be able to challenge your game of chess. An AI that can pretend to speak Chinese to you probably won't sweep your floor. Weak AI lacks human consciousness, although it may be able to simulate it. However, weak AI can automate the tedious, repetitive parts of most jobs, leaving humans to handle the parts that require human care and attention.

Weak AI is both the most limited and the most common of the three types of AI. The idea behind weak AI is not to imitate or replicate human intelligence. Instead, it's meant to simulate human behavior. So it's far from human intelligence, and it doesn't try to do so. A common misconception about weak AI is that it has little intelligence — more like artificial stupidity than AI. But even the smartest looking AI today is only weak AI. In reality, then, a narrow or weak AI is more of an intelligent expert. It does the specific task it was programmed to do very smartly.

Narrowly defined AI systems are good at performing a single task or a limited range of tasks. In many cases, they even outperformed humans in certain domains. But once they encounter a situation that exceeds their problem space, they fail. Nor are they able to transfer their knowledge from one field to another.

While narrow AI has failed at tasks requiring human intelligence, it has proven its usefulness and found its way in many applications. Some common and well-known examples of Narrow AI includ Facebook's news feed, Amazon's suggested purchases, and Apple's Siri, an iPhone technology that answers users' verbal questions. Email spam filters are another example of weak AI, where computers use algorithms to learn which messages are likely to be spam and then redirect them from the inbox to the spam folder. 

 

- Key Features of Narrow AI

Weak AI helps turn big data into usable information by detecting patterns and making predictions. The erratic speed of data evolution creates prerequisites for narrow AI. Combine that with the recent avalanche of growth in user-generated content, and it's clear that no organization (or human being) can handle it without the help of Narrow AI.  

  • Narrow AI systems are programmed to perform specific tasks based on certain conditions and parameters. For example, an AI-based recommendation engine can recommend products based on your previous purchase history.
  • Machine learning and deep AI have grown into two major subsets of Narrow AI. They are deployed in different systems to learn from human behavior and input to provide relevant insights.
  • Natural Language Processing (NLP) is another narrow AI technique used to help machines conduct conversations by understanding human communication in natural language through chatbots and voice assistants.
  • Narrow AI can react instantly to a situation or context or user input based on pre-programmed logic. This is called reactive AI, and it's pretty basic. In contrast, Limited Memory AI can learn from human behavioral data over time and has advanced responsiveness.

 

- Types of Narrow AI Technologies

The narrow AI technologies we have today basically fall into two categories: symbolic AI and machine learning.

  • Symbolic AI, also known as old-fashioned artificial intelligence (GOFAI), has been a major research area for most of the history of artificial intelligence. Symbolic AI requires programmers to carefully define the rules that specify the behavior of intelligent systems. Symbolic AI is suitable for applications with predictable environments and well-defined rules. Although symbolic AI has fallen out of favor over the past few years, most of the applications we use today are rule-based systems.
  • Machine learning is another branch of narrow AI that develops intelligent systems by example. The developer of a machine learning system creates a model and then "trains" it by providing many examples. Machine learning algorithms process examples and create mathematical representations of the data that can perform prediction and classification tasks.

For example, a machine learning algorithm trained on thousands of banking transactions and their outcomes (legitimate or fraudulent) will be able to predict whether new banking transactions are fraudulent.

 

 

[More to come ...]


Document Actions