Glossary: (new) technological terms
In this article, I explain the most important and current terms that are increasingly encountered in the technological world. These concepts are relatively new and are now being used in policy documents, political decisions, and strategic organizational plans. They also appear more frequently in news reports, opinion pieces, and professional literature.
Because these terms sometimes carry a different nuance or meaning than you might be used to, I will briefly go through them with you. This way, you can quickly understand what they mean and better follow what is happening at the intersection of technology and societal developments.
Large Language Models (LLMs)
Large Language Models are advanced algorithms trained on vast amounts of text data to understand and generate human language. They analyze the context of your input and predict the most likely subsequent sentences (output). Examples include GPT-3 and GPT-4 by OpenAI, which are used in chatbots, text generation, translations, and more.
Traditional chatbots, such as those from KPN or DHL, often use simpler models that respond with fixed answers; they are less flexible and may not always interpret your question correctly. For instance, if you ask, “How can I change my subscription?” such a chatbot often provides a standard response without asking about your specific situation. LLMs, on the other hand, can answer complex and varied questions by understanding context, although they remain limited to the data they were trained on.
Artificial Intelligence (AI)
Artificial Intelligence encompasses not just software or algorithms but the entire ecosystem of hardware, networks, databases, sensors, and cloud platforms required to enable machines to perform tasks that typically require human intelligence.
AI systems learn to recognize patterns in data through machine learning and improve themselves as they process more data. Examples of AI applications include smart personal assistants like Siri and Alexa, which recognize speech and execute commands, and self-driving cars that analyze traffic situations in real-time and respond accordingly.
Humanoid robots, such as Sophia from Hanson Robotics, also fall under this category; they can communicate, recognize emotions, and act independently in certain situations. The difference with LLMs is that AI systems often use hardware and sensors to operate in the physical world and make autonomous decisions.
Generative AI
Generative AI refers to AI systems that create new content based on patterns they have learned from their training data. In addition to text, they can also generate images, music, and videos. Examples include DALL·E (image generation), Jukedeck (music composition), and GPT-4 (text creation).
This technology is used in creative industries such as marketing, art, and entertainment, where it can automatically produce social media content, advertisements, or music. Generative AI leverages neural networks that recognize complex relationships and create original, surprising combinations.
A practical application is the automatic generation of personalized newsletters, where the content is tailored to the interests of each recipient.
AI Agents and Agentic AI
AI agents
AI agents are software programs designed to perform specific, often limited tasks. They respond to commands they receive and execute them step by step. These agents are typically task-oriented and operate within a predefined scope.
Example: An AI agent can help you schedule an appointment in your calendar, answer a standard question in a customer service chat, or automatically sort and respond to emails. They follow a fixed set of rules or algorithms and only perform what is explicitly requested.
While AI agents operate autonomously within their task, they lack the ability to set goals, plan, or perform tasks outside their assignment. They cannot flexibly switch between different types of tasks without human adjustment or new input.
Agentic AI (Autonomous AI Agents)
Agentic AI is much more advanced and can independently plan and execute complex goals. It is not limited to a single task but can combine multiple different tasks and flexibly switch between steps within a process.
This AI operates autonomously without continuous human intervention. It can decide what information is needed, conduct independent research, analyze data, draw conclusions, and take actions that contribute to achieving the goal.
Example: Imagine asking an Agentic AI to create a comprehensive market analysis on the future of healthcare. This AI will independently search for relevant reports and current news articles, summarize them, create graphs and presentations, and then automatically produce a video that is published on various social media platforms, each styled to suit the platform (e.g., professional on LinkedIn, more casual on TikTok). All of this happens without you needing to provide new instructions or manually manage processes.
Agentic AI is already being used in sectors such as financial services, where it automates advanced workflows like customer onboarding, risk assessment, and fraud investigation. This AI can independently make decisions, combine various steps, and adapt the process to changing circumstances.
Key differences between AI Agents and Agentic AI are as follows:
- Task Scope: AI Agents usually perform one specific task within a clearly defined framework. Agentic AI can combine multiple, diverse tasks and determine which steps are necessary to achieve a goal.
- Autonomy: AI Agents wait for instructions and execute them without initiative. Agentic AI works proactively, sets its own goals, creates plans, and adjusts them as needed.
- Complexity: AI Agents are often simple, task-oriented programs. Agentic AI is more complex, can flexibly switch between tasks, learns from new input, and adjusts processes independently.
Examples:
- AI Agent: A chatbot that answers customer questions using fixed scripts.
- Agentic AI: A system that independently conducts a market analysis and automatically distributes the results as a video and report across various channels.
Artificial General Intelligence (AGI)
AGI refers to AI that fully matches and surpasses human cognitive abilities. Unlike current AI, which is specialized in specific tasks (e.g., language processing, image recognition, or playing games), AGI can learn, reason, and solve problems across diverse domains without prior training on those specific tasks.
This means an AGI system could, for example, write a scientific article, run a business, and develop a strategic plan. AGI could replace much human labor and accelerate innovation. It is expected that AGI could become available within 2 to 10 years, although there is significant debate about its technical feasibility and ethical implications. Researchers are focusing on “meta-learning” and “transfer learning” as key building blocks.
Artificial Superintelligence (ASI)
ASI is a future vision of AI that not only surpasses all human intelligence but also develops entirely new ways of thinking and capabilities that humans cannot yet imagine.
ASI could independently develop science and technology, solve complex societal problems, and potentially possess self-awareness. This raises important philosophical and ethical questions, such as how we can maintain control over a superintelligence.
Some experts predict that ASI could emerge between 2028 and 2040, while others believe it will take much longer or may even be unattainable. Examples of ASI currently only exist in fiction, such as in movies like Her or Ex Machina, but the topic is a high priority for AI researchers and policymakers worldwide.
AI Hallucinations
AI hallucinations occur when an AI model generates incorrect or fabricated information that appears convincing but is factually inaccurate. This is a known issue with large language models like ChatGPT and Gemini.
The risk of hallucinations is particularly concerning in applications where accuracy is critical, such as journalism, healthcare, and legal advice.
Multimodal AI
Multimodal AI systems can process and understand multiple types of data at the same time such as text, images, audio, and video. This enables richer, more context-aware analysis and interactions. Applications include advanced search engines and interactive assistants that respond effectively to various input formats.
Generative Engine Optimization (GEO)
GEO is a new optimization strategy focused on improving how digital content is found and presented by generative AI systems like ChatGPT and Google Gemini. Unlike traditional SEO, GEO uses AI-specific metadata and structured content to increase the chance of being featured in AI-generated answers.
Diffusion Models
Diffusion models generate realistic images, audio, or other data by gradually adding noise to data and learning how to reverse this process to create coherent outputs. Examples include Stable Diffusion and DALL·E, which produce high-quality images from text prompts.
Deepfakes
Deepfakes are AI-generated synthetic media that manipulate someone’s face, voice, or appearance to create realistic but fake content. Originally known for face-swapping videos, deepfakes now include fake audio and images, raising concerns about misinformation and prompting the development of detection and regulation tools.
Blockchain
Blockchain is a decentralized, immutable digital ledger where transactions are recorded in secure, sequential blocks. It ensures transparency and prevents fraud without intermediaries. For example, all Bitcoin transactions are publicly recorded, allowing anyone to verify that coins aren’t spent twice.
Cloud Computing
Cloud computing means renting computing resources (power, storage, applications) over the internet instead of owning physical servers. Services like Amazon Web Services (AWS) let you run virtual servers and store data flexibly, paying only for what you use.
Internet of Things (IOT)
IoT connects everyday devices (thermostats, fridges, industrial machines) via the internet to collect and share data. For instance, a Nest smart thermostat senses when you’re home and adjusts heating to save energy.
5G
5G is the latest mobile network generation, offering very high download speeds (gigabits per second), low latency, and the capacity to connect millions of devices simultaneously. In smart factories, 5G syncs robots and sensors for efficient production.
Post-Quantum Computing
This field develops new encryption methods that remain secure against powerful quantum computers. Organizations are adopting standards like CRYSTALS-Kyber from NIST to protect data in a future with quantum computing.
Zero Trust
Zero Trust is a security model that assumes no user or device is trusted by default, even inside a network. For example, Microsoft Azure AD Conditional Access requires multi-factor authentication and device security checks every time before access is granted.
Metaverse
The Metaverse is a virtual 3D environment where users interact, collaborate, and attend events using digital avatars. Platforms like Decentraland enable buying virtual land, building on it, and hosting events, often transacted via cryptocurrency and tokens.
Tokens
In AI:
- Tokens are units of text—words, fragments, or punctuation—that AI models like GPT use to learn and predict language patterns. For example, “ChatGPT is smart” might tokenize as [“Chat”, “G”, “PT”, ” is”, ” smart”].
In Blockchain:
- Tokens represent value, rights, or access on a blockchain:
- Utility tokens grant access to services.
- Security tokens represent ownership or shares.
- Governance tokens allow voting on network decisions.
Examples
In ChatGPT, more tokens in input/output increase compute cost.
On blockchain platforms like Uniswap, governance tokens enable voting on updates, while utility tokens pay fees.
