Artificial Intelligence (AI) isn’t just a buzzword anymore. It has disrupted technology with a widening range of applications in everyday life. AI was acknowledged as a possible technology by John McCarthy in the year 1955. He believed machines could be made to think, behave and perform as humans. Renowned as the Father of AI, McCarthy created Lisp computer language in 1958. It is the programming language used even today to create standard AI applications.
AI is a multifaceted technology with a variety of types and categories. In this article, I have simplified the types of AI by dividing them into two broad categories:
Types of AI
The realm of AI includes machine learning, big data, cognitive intelligence, deep learning, natural language processing (NLP), and much more. Here is a broader categorization to help you easily navigate through the types of AI and analyze which fits your requirement.
Type 1- AI can be categorized into three subsets based on current utilization and future viability.
1. Narrow- Narrow AI is the simplest form of AI that is efficient at performing a single task. Weather forecast applications and speech-to-text converters are typical examples of Narrow AI. Though sophisticated, Narrow AI is referred to as ‘Weak AI’ considering its capabilities are limited. Narrow AI-powered devices cannot perform functions beyond their scope. Autonomous cars, computer vision, Google Translate, and other NLP-enabled apps, which we use frequently, have reached only up to the stage of Narrow AI. It doesn’t perform on the standards of human intelligence but is considered adequate for routine jobs. Narrow AI can cut down human intervention from repetitive and monotonous jobs.
2. General- General AI is a more complex version of AI as it is predicted to work like the human brain. This human-level AI will be capable of understanding, reasoning, and reacting to its surroundings. It is known as ‘Strong AI’, and we are coming across several challenges as we strive to achieve it. Computers can process data and pull up relevant information from their memory. Unlike humans, it is difficult for them to react to unprecedented situations or innovate to create something that never existed. Teaching computer skills like planning, perception, and decision-making based on unrelated information is a difficult task. Though it isn’t unachievable, General AI is still distant from real-life applications.
3. Super- Artificial Super Intelligence (ASI) is a concept given by Nick Bostrom, a University of Oxford scholar and AI expert. He intends to create an AI better than the combined power of best human brains from every creative and scientific field. ASI would be a more capable form of General AI and is unachievable at the moment. It is predicted to be a fully capable form of AI. Scientists like Stephen Hawking believe it to be devastating for the human race, while others claim it will enable humans to do more.
Type 2- Another classification of types of AI can be done based on its cognitive capabilities and functionality. The four subsets of type 2 AI are:
1. Reactive- Reactive AI systems are the oldest form of artificial intelligence enabled machines with extremely limited capability. They do not have a memory and emulate the human mind’s ability to respond in the present situation only. They are programmed to perform a set of tasks in response to a stimulus. Reactive AI machines do not have the ability to ‘learn’ as they do not store experiences or refer to them when performing automatic actions. These machines can effectively respond to a complex combination of instructions. IBM’s Deep Blue is a reactive AI-based machine. It is renowned for being the only computer to beat a human at chess. The machine won chess matches against Grandmaster Garry Kasparov in 1997.
2. Limited Memory- Limited memory machines have the capability to learn from historical data and make quick decisions. It derives information from stored data fed to the system earlier. Besides, it builds experiential knowledge by observing actions from its environment. The system uses large volumes of training data for deep learning. It creates reference models and uses them along with information stored in its memory for solving problems. All present-day AI systems fall under this subset. These include virtual assistants, chatbots, and autonomous vehicles. The process to train them includes image recognition AI and a wide variety of related images to act as training data. It scans thousands of pictures to identify and store them for future use. When the machine is exposed to a real environment, it makes predictions based on its ‘learning experience’ and stores them. The learning process continues to reach higher levels of accuracy. For instance, an autonomous vehicle is trained using images of routes, landmarks, and other vehicles. It stores the data and when on the road, it reconciles the information to make a decision. The vehicle also detects patterns and driving behavior of human-driven cars. It makes real-time adjustments based on these changes.
3. Theory of Mind- Theory of Mind is a proposed form of AI aimed at creating machines with the emotional capabilities of a human. These systems are complex and go beyond the memory or decision-making capabilities. They are focused on exhibiting human-like emotions to drive conversations. We can consider them the advanced versions of our current voice assistants or chatbots. The concept is aimed at building a thought process of the machine to derive a behavioral output. These machines are being trained to identify, understand, remember, and respond to emotional inputs. The applicability of the ‘Theory of Mind’ has been noticed in robots Kismet and Sophia. Kismet was created in 2000 and could imitate human emotions through facial expressions. Sophia, launched in 2016, is a humanoid robot with human-like physical features. This bot has the capability to respond to human interaction with thoughtful responses and appropriate facial expressions.
4. Self-awareness- Self-awareness AI is the most advanced stage of AI development. It is a theoretical concept that perceives AI to reach the self explanatorily stage. An evolved form of AI that may possess human consciousness. Creating this type of AI is challenging, but all AI research by scientists aims to achieve this stage. AI systems based on self-awareness will possess emotions, understand human needs, and realize their own feelings. The system will be an advanced stage of Theory of Mind enabled machines, as it will not replicate human beliefs and feelings. It will be self-regulated with the capability to think, make inferences, and react. According to experts, this type of AI can be potentially dangerous for our existence or may boost our progress as a species. The dream is farfetched, and we cannot yet comment upon the outcome of implementing self-awareness AI-powered systems.
The rapid developments in the field have empowered companies to come up with innovative solutions backed by AI. Among the emerging types of AI and associated technologies are Generative AI, Federated Learning, and Neural Network Compression. Generative AI is a breakthrough development in the AI industry. It aids in content development by enabling computers to create 3D and 2D images using text, audio, and pictures. Federated Learning is an emerging machine learning technique that assists in training and improving AI-based tools. Neural Network Compression is another upcoming technology to address Big Data storage issues and facilitating access to Deep Neural Networks (DNN). It compresses data for easy storage and deployment on the cloud.
As we progress towards developing more competent AI systems, we inch closer to the most advanced tier of both Type 1 and 2 of AI. The future of our race with cutting-edge AI technology is uncertain, but if appropriately implemented, these cogent systems will bring the dawn of progress for us.