Artificial intelligence (AI) and machine learning (ML) may be top of mind for leaders in every industry but understanding the different types of capabilities and how they can help your business isn’t always easy. But never fear – we aim to clear it up in this geek guide to the types of AI and ML.
Scope and maturity classifications
If you search for “types of AI,” you’ll quickly see two major classifications show up in the results as illustrated in the diagram below. These categorizations focus more on the AI scope and maturity level as opposed to the type of technology or application.
By capability – from narrow to super AI
Narrow or weak AI: Artificial narrow intelligence (ANI) are AI systems that are built to focus on one narrow task such as predicting the weather, recommendation engines or home voice assistants. Most AI in use today falls into this category.
ANI examples: Amazon Alexa and Google Home voice assistants for the home (Source: UX Collective[ii])
Strong AI: Artificial general intelligence (AGI) is assumed to be “on par” with human ability in terms of understanding, learning and adapting to solve problems. While supercomputers are bringing us closer to strong AI, the human brain is still faster at an estimated one billion calculations per second (cps).i
Super AI: Artificial superintelligence (ASI) will exceed human ability in multi-faceted understanding, learning, analysis and decision-making. This is the basis for singularity – a hypothetical point in time when super-intelligent machines have capabilities that cannot be predicted by humans.
By functionality – from reactive to self-aware
Reactive machine: This ANI can only perform specific tasks based on present data such as IBM’s Deep Blue looking at the chessboard to predict the next moves. It has no memory-based functionality and does not have the ability to learn.
Limited memory: This is also an ANI but it is capable of learning from past data to make decisions. Most enterprise applications today, such as self-driving vehicles or chatbots, rely on limited memory AI.
Theory of mind: This AGI will be capable of understanding people and things within an environment and adapting its response accordingly. It will likely be a convergence of a few emerging areas of research such as AI with emotional and social intelligence, contextual awareness, etc.
Self-awareness: This ASI is still theoretical but will be capable of not only understanding its environment but also itself as an independent, self-aware entity and act accordingly.
Artificial Intelligence: From the Public Cloud to the Device Edge
In this report we describe how Equinix and NVIDIA are jointly helping with distributed AI applications by providing AI infrastructure that spans across the continuum from edge data centers all the way to the core clouds.Read More
The maturity categories above can also be helpful in understanding the different types of AI technologies as illustrated in the diagram below. For example, image recognition is a limited memory type of ANI that can “learn” to recognize new images based on past images it was trained on. More complex, strong AI systems in the future will likely be based on a combination of these technologies:
Machine learning: ML trains a machine to recognize patterns in the future based on an initial model that can improve over time without being explicitly programmed to do so. The article “A Practical Guide to Artificial Intelligence for the Data Center” describes how this process works. Email spam filtering and image recognition are some examples.
Natural language processing (NLP): NLP enables machines to understand, analyze and manipulate natural language data. Applications include chatbots, concept mining, information extraction, auto-classification of content, question and answering, auto-translation and more.
Expert Systems: Expert systems are designed to mimic the decision-making ability of a human expert to solve complex problems and provide advice. As depicted below, they are comprised of a knowledge base that has the facts and rules, provided by the human expert, and an inference engine that applies those rules to user queries to deduce new facts. Examples include decision support systems, rapid prototyping, interpretation of sensor data, medical diagnosis, asset monitoring, etc.
Vision: Computer vision trains machines to “see” – that is to capture, analyze and understand the visual world and react accordingly. It is usually paired with machine learning to improve recognition accuracy over time. Example applications include facial recognition, visual content moderation, autonomous vehicles, remote sensing, surveillance, etc.
Speech: This includes automated speech recognition (ASR) in which a computer recognizes spoken words and converts them to text (speech to text), as well as artificial production of human speech (text to speech). Voice assistants and automated transcription services typically use a combination of ASR and NLP in their products.
Planning: AI planning leverages autonomous techniques to solve complex planning and scheduling problems based on inputs that consist of a description of the starting state, desired goal/outcome and set of possible actions that can be taken to achieve the goal. The output is the sequence of actions with the highest likelihood to achieve the goal successfully. If/then analyses, conditional and contingency planning with incomplete information are examples of how AI planning might be used.
Robotics: Robots are intelligent machines that can perform tasks in the physical world. As discussed in my previous article, “Autonomous Things vs Hyperautomation,” robots are often deployed in environments that are dangerous to humans, such as clinical care of COVID-19 patients, working with hazardous materials or extreme temperatures.
Getting to strong AI depends on interconnection
Moving from limited AI to strong, integrated AI will depend on No one company can do it alone, which is why vendor-neutral interconnection solutions such as those found on Platform Equinix® will be essential for joining compute resources to the digital ecosystems of clouds, networks and partners working on integrating and improving AI systems.
To learn more, read the white paper, “Artificial Intelligence: From the Public Cloud to the Device Edge.”