How to speak professionally about artificial intelligence: 20 key terms from this field

Basic concepts

1. Artificial Intelligence (AI) — a branch of computer technology that allows machines and programs to mimic human intelligence: to learn, make decisions, and solve a variety of tasks.

2. Neural network — a technology (mathematical model, machine learning method) that mimics and adapts to the human brain to analyze and process data at different levels. In their work, neural networks use artificial neurons similar to the way the human brain uses a system of neural connections.

3. Prompt — an incoming request or instruction that a human composes and provides to an AI system in order to get a result in the form of a specific response. Accordingly, the process of developing and improving prompts for AI models in order to obtain desired results is called prompt engineering, and specialists who develop and optimize requests or instructions for generative AI are called prompt engineers.

4. Machine learning — a subsection of AI that trains machines to recognize patterns and improve tasks as more data becomes available, without explicitly programming each action.

5. Deep learning — an advanced machine learning method that uses artificial neural networks to analyze huge volumes of unstructured data (images, text, audio). The goal of such learning is to discover underlying patterns and establish complex relationships that are inaccessible to humans.

Technical terms

6. Algorithm — a set of explicit instructions provided by AI that define how the system should process data, learn, and make decisions. Algorithms are used to create AI models and perform specific tasks such as classification, prediction, or optimization.

7. Model — a mathematical or statistical representation that is derived by training an algorithm on data. A model 'learns' patterns and dependencies in the data to make predictions or decisions based on new, unknown information.

8. Perceptron — the basic building block of neural networks that converts incoming signals into outgoing signals. It was invented by Frank Rosenblatt in 1957 as an algorithm for data classification.

9. AI singularity — a concept according to which the level of artificial intelligence in the future will surpass human intelligence in all spheres and its further development will become uncontrollable or unpredictable for humans.

10. Transformer — a type of deep learning model that processes sequential data and determines the relationship of different elements to each other, similar to the way a person pays attention to word order while reading to understand the meaning of an entire sentence. The best-known example is the generative pre-trained Transformer (GPT), on which the ChatGPT chatbot operates.

Concepts that describe the behavior and ethics of AI

11. Responsible AI — a type of AI that takes into account the ethical aspects of people's lives, based on the principles of fairness, transparency, and responsibility. Allows for results that are not biased, discriminatory, or otherwise harmful to humanity.

12. Deepfake — content generated or edited by AI that resembles real people, objects, places, etc., and is misleading as to its authenticity or truthfulness.

13. Hallucination — inaccurate, irrelevant, or meaningless answers given by the AI. For the most part, this happens due to incomplete training data. Depending on which model is used, the frequency of hallucinations can vary: for example, about 3% in ChatGPT as opposed to 27% in Google's systems.

14. Grounding — a method of reducing AI hallucinations based on linking results to data from a reliable database. To minimize inaccuracies, it is worth taking care in advance to collect accurate data and other background information if necessary.

15. Black box — a term that describes a situation when an AI model makes a decision or makes predictions, but the mechanism of this process remains opaque or hard to understand. This is due to the complexity of many modern AI models, in particular neural networks, the multitude of parameters, and non-linear relationships between them.

General concepts

16. AI-agent — a program that works in the digital space using AI algorithms without human intervention. Many of the created AI-agents help to facilitate processes in the financial sphere (for example, they analyze the financial indicators of a company, trade on crypto exchanges, issue NFT-collections, or analyze the purchase/sale of cryptocurrency on certain wallets).

17. Big Data — the name given to a large amount of structured or unstructured information that is collected from various sources and analyzed to discover patterns and trends that businesses can rely on to formulate their strategy. Because of the sheer volume of data, it is analyzed using computer-based, rather than traditional, methods.

18. Strong (general) AI — an AI that is capable of performing the same tasks as humans and often even better. This is possible due to the use of human-like cognitive abilities (adaptability, creativity, reasoning, self-learning).

19. Weak AI — an AI that is capable of handling different tasks but cannot understand them in a human-like way (lacks its own awareness). Weak AIs include chatbots, language models, translators, voice assistants (Siri, Google Assistant).

20. Large language model — a powerful language model that is trained on huge amounts of unlabelled text and is able to generate, translate, analyze text, and provide human-like responses. For example, such models are used in various versions of ChatGPT from OpenAI, as well as in Google's BERT and LaMDA systems.

Familiarizing yourself with the above concepts will open the door to the exciting world of AI, which is currently shaping the future of many areas of our lives. It is likely that you will then want to dive deeper into this topic to learn even more interesting things about artificial intelligence.