Glossary of AI Terms

 

Artificial Intelligence (AI) – A specific field of computing that focuses on creating systems capable of performing tasks which normally require human intelligence.

 

API An Application Programming Interface (API), is a set of functions and procedures for building software. It makes it easier to develop a computer program by providing all the building blocks, which are then put together by the developer.

 

Autonomous – Autonomous refers to the ability to act independently without any outside control. An AI system is autonomous if it doesn’t need help from people. Driverless cars utilise autonomy; the cars do not require humans to navigate, steer, brake or accelerate.

 

Black box – A black box is a system which can be viewed in terms of its inputs and outputs, but the observer has no knowledge of what’s happening on the inside. A black box AI system could be mistaken for human intellect. Our CTO Gareth wrote about black boxes here.

 

Channel – Any of the various communication platforms where a chatbot can go live such as Facebook Messenger, Viber, WhatsApp, Skype etc.

 

Chatbot – A chatbot is a software system designed to engage in conversation with human users by communicating through text chats, voice commands, or both. Chatbots range from basic bots to advanced conversational AI that use natural language processing like Siri. 

 

Conversational AI – A branch of artificial intelligence that focuses on interpreting human language and communicating with humans. It includes advanced features of AI such as NLP, machine learning, and neural networks.

 

Conversational UI – If you have used any messaging platform like Facebook Messenger or Viber, you will have interacted with a conversational user interface. The user interface determines how the experience looks and feels for users in terms of tone, humour, language, and graphical elements such as buttons, menus and images.

 

Deep learning – Deep learning is a type of machine learning in artificial intelligence. It utilizes multiple levels of artificial neural networks to solve problems. The artificial neural networks are built like the human brain, with nodes connected together like a web. While programs analyse data linearly, deep learning enables machines to process data nonlinearly by passing on data from one node to the next, which, after each pass improves the machine’s accuracy. Deep learning is generally used for high-scale, complex problems.

 

Intent – Represents a mapping between what a user says and what action should be taken by a chatbot. These are the core blocks that make up building a conversation. For example, if a user wants to book a flight on a chatbot, that is his or her intent. Most main intents would be defined upfront, and then refining or defining of intents would take place from customer interactions.

 

Machine learning – A cornerstone of AI is machine learning, where a computer learns independently through experience without being explicitly programmed. AI systems that utilise machine learning get better as more data comes through. Machine learning includes many subsets or methods such as deep learning, transfer learning, and natural language processing.

 

Moore’s Law – Named after the co-founder of Intel, Moore predicted in 1965 that the number of transistors that can be placed on an integrated circuit doubles every two years. This trend has been continuing since 1965 with no signs of any slowdown yet. It can be applied in general to a range of technology areas that are growing at an accelerating rate. It has remarkable application in AI that is currently allowing systems to approach human capabilities in a few specialist fields.

 

Natural language processing (NLP) – Natural language processing technology allows AI to interpret human communication. It takes an advanced neural network to parse human language. Chatbots and conversational artificial intelligence show the potential for NLP.

 

Neural network – A neural network is a framework used for deep learning and an approach to machine learning. It developed out of attempts to mimic processes of intelligent thought in the human brain. It gives AI the ability to solve complex problems by breaking them down into levels of data (neurons).

 

Node A node is where different points of a conversation flow intersect. Ambit’s platform has multiple nodes so users can control the human-to-bot conversation.

 

Supervised learning – When you train an AI model using a supervised learning method you teach the machine the correct answer either ahead of time or after it makes mistakes or has low ‘confidence level’ in its guess. In supervised learning, the AI knows the question and answer. This is the most common method of training because it yields the most data – it explains cause and effect between the input and output.

 

Transfer learning – A subset of machine learning is transfer learning, where machines stores knowledge gained from one task and apply it to a related problem. Take the example of image identification, the knowledge gained from learning to recognise cars could be applied to recognising trucks.

 

Turing Test – a test proposed (1950) by the English mathematician Alan M. Turing to determine whether a computer can “think.” Thought the test was originally conceived as a way of determining if a human could be fooled by a conversation over text, between a human and an artificial intelligence ‘black box’, it has since become shorthand for any AI that can fool a person into believing they’re seeing or interacting with a real person.

 

Utterance – An input, either written or spoken, made by a human user to a chatbot or other conversational AI instance. The chatbot has to understand the utterance’s intent to determine its next appropriate action. For instance, “I want to apply for a mortgage” is an utterance, and the chatbot will match that utterance to the intent of applying for mortgages and respond appropriately.

 

Unsupervised learning – AI Machines are capable of learning by themselves, and they’re using a lot of data and processing power to do so. Rather than feeding answers into the machine like, “why people choose one car over another”, we simply feed a bunch of data so that it can find whatever patterns it is able to. Although extremely powerful, this type of learning is very domain specific and is currently not very common anywhere.