Artificial intelligence - all just chat?

The above paragraph is concise, informative, believable and made up by ChatGPT. AIMA does not exist and although there may well be a real Dr E. Collins, they have nothing to do with AIMA. To produce the above, all I needed to do was to ask ChatGPT “Write a story about AI in the style of a BBC news article”. When Artificial Intelligence (AI) can produce such reasonable responses to questions such as this along with many more complicated questions how can we know what is “real” anymore? Perhaps this whole piece has been written by AI – would you know any difference?  

Artificial Intelligence as a concept is not a new thing, indeed in Greek myth the intelligent automata Talos is introduced. It was in 1956 at a workshop at Dartmouth College, New Hampshire, USA that modern AI was born. Since this time there have been some notable peaks and troughs in the development of AI but now research and real-world applications have exploded largely driven by increased computing power, amount of data available and the refinement of machine learning techniques. AI is generally defined as the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with humans.

The most well-known AI at this moment in time (late 2023) is probably ChatGPT. Chat Generative Pre-trained Transformer to give it its full name is a large language model chatbot. Generative here means that it can generate human-like content, pre-trained is because it has been trained on a large set of text and transformer is a type of deep learning architecture based on neural nets. A large language model is an artificial neural network that uses massive amounts of data to learn billions of parameters during training and consumes a large amount of computational resources to be able to understand and generate human-like language.

ChatGPT is a very impressive tool that can answer questions on a huge variety of topics, it can write poetry, stories, computer code, business ideas, and student essays. It is not perfect though; it can present as fact, false or misleading information an effect known as hallucination in in analogy with the process that happens in the human mind. It is due to this requirement for a human mind that some commentators have said that the phenomenon is more accurately described by the word Confabulation which is the generation of plausible-sounding but potentially inaccurate or fabricated information. Whatever you choose to call it the issue is an inevitable by-product of AI producing responses based on limited or incomplete knowledge.

So, what now for AI and its impact on humankind? AI now has the potential to significantly change the way we work; indeed, it could replace people in many jobs. But just as the internet and holiday booking websites have not seen the demise of traditional travel agents, AI does not mean it will take all our jobs. Rather the types of jobs we do will change and AI will become more integrated into everyday jobs, helping productivity. Chatbots are just one type of AI.

AI has applications in many different areas. Already it is being used (really this time) in the medical sector to help diagnosis, in the finance sector to make trading recommendations and in the agriculture sector to help farmers to identify areas that need irrigation to name just a few areas of AI application. One thing that ChatGPT, along with other AI systems, is not, is an artificial general intelligence (AGI). ChatGPT is trained and tuned to be a chatbot it is not meant to be an AI medical diagnostic tool.

AGI on the other hand would be able to perform any task that a human could perform and indeed outperform humans in a lot of cases. So, in the future will we be interacting with humanoid robots powered by an AGI brain that are near indistinguishable from humans? Will we need 3 laws of AI/robotics? It may seem outlandish and just fiction, but all science is fiction until it becomes fact.

AI and Inspec

Inspec provides substantial coverage of AI. The following parts of the classification scheme contain the relevant information:

• C6200 Artificial intelligence software and techniques
• C6210 Knowledge-based systems
• C6220 Knowledge acquisition
• C6220D Data mining
• C6230 Knowledge representation
• C6240 Knowledge verification
• C6250 Reasoning and inference techniques
• C6260 Machine learning (artificial intelligence)
• C6261 Supervised learning
• C6262 Unsupervised learning
• C6263 Reinforcement learning
• C6264 Neural nets
• C6265 Support vector machines
• C6266 Other learning models (inc. Naive Bayes)
• C5190 Neural net devices
• B1295 Neural nets (circuit implementations)
• C1230 Artificial intelligence (theory)
• C1230D Neural nets (theory)
• C1230L Learning in AI (theory)
• C1230R Reasoning and inference in AI (theory)
• C3260N Intelligent actuators
• C3240N Intelligent sensors
• B6210Q Intelligent networks
• B7230S Intelligent sensors

The most important controlled terms from the Thesaurus include new terms being introduced to Inspec in 2024 to cover the AI area:

New controlled terms introduced in 2024

• ensemble learning
• explainable AI
• federated learning
• Q-learning
• transfer learning (artificial intelligence)

Existing controlled terms important to this area of research

• adaptive resonance theory
• affective computing
• AI chips
• ant colony optimisation
• ART neural nets
• artificial bee colony algorithm
• artificial intelligence
• artificial life
• backpropagation
• belief maintenance
• belief networks
• belief propagation
• blackboard architecture
• Boltzmann machines
• case-based reasoning
• cellular neural nets
• cerebellar model arithmetic computers
• cognitive systems
• common-sense reasoning
• convolutional neural nets
• cooperative systems
• data mining
• DATALOG
• deductive databases
• deep learning (artificial intelligence)
• description logic
• diagnostic reasoning
• expert systems
• explanation
• feedforward neural nets
• frame-based representation
• fuzzy cognitive maps
• fuzzy control
• fuzzy neural nets
• fuzzy reasoning
• generalisation (artificial intelligence)
• graph neural networks
• Hebbian learning
• Hopfield neural nets
• inference mechanisms
• intelligent actuators
• intelligent control
• intelligent design assistants
• intelligent manufacturing systems
• intelligent networks
• intelligent robots
• intelligent sensors
• intelligent structures
• intelligent transportation systems
• intelligent tutoring systems
• knowledge acquisition
• knowledge based systems
• knowledge engineering
• knowledge representation
• knowledge representation languages
• knowledge verification
• learning (artificial intelligence)
• learning automata
• learning by example
• learning systems
• logic programming languages
• model-based reasoning
• multi-agent systems
• multi-armed bandit problems
• multilayer perceptrons
• naïve Bayes method
• nearest neighbour methods
• neural chips
• neural net architecture
• neurocontrollers
• neuromorphic engineering
• nonmonotonic reasoning
• ontologies (artificial intelligence)
• optical neural nets
• perceptrons
• planning (artificial intelligence)
• PROLOG
• radial basis function networks
• random forests
• recurrent neural nets
• reinforcement learning
• self-organising feature maps
• semantic networks
• semi-supervised learning (artificial intelligence)
• software agents
• spatial reasoning
• supervised learning
• support vector machines
• swarm intelligence
• temporal reasoning
• truth maintenance
• uncertainty handling
• unsupervised learning
• wavelet neural nets