Glossary

"

"Doomer"«المتشائمون من الذكاء الاصطناعي»: An informal, sometimes dismissive label for people who think advanced AI carries a high risk of catastrophe. The platform uses it only in quotation marks or when explaining the term, and describes these views in their holders' own words.

A

Accelerationism / e/accالتسارعية (دعاة التسريع) / «التسارعية الفعّالة»: The view that AI progress should move faster, because its benefits (cheaper energy, medicine, abundance) outweigh the risks and heavy regulation does more harm than good. "Effective accelerationism" (e/acc) is an online movement that emerged in 2022–23. Marc Andreessen's 2023 "Techno-Optimist Manifesto" is a key text.
Agentic AIالذكاء الاصطناعي الوكيل (Agentic AI): AI systems built from agents that can carry out a sequence of tasks on their own within set limits. In April 2026 the UAE announced a goal to move 50% of federal government sectors, services and operations to agentic AI within two years.
Agentic organisationالمؤسسة القائمة على الوكلاء: An organisation in which AI agents handle many routine steps while people set the purpose, make the key decisions and stay accountable. It is still made of familiar things: a purpose, people, information, tools and rules.
AGI (artificial general intelligence)الذكاء الاصطناعي العام: A loose term for AI that could handle most intellectual tasks about as well as a capable person. There is no agreed definition or test. Some leaders, such as Anthropic's Dario Amodei, prefer other terms ("powerful AI").
AI agentوكيل الذكاء الاصطناعي: An AI tool that doesn't just answer but also acts, carrying out several steps towards a goal (searching, filling in a form, sending a draft), ideally with a person approving what matters.
AI governanceحوكمة الذكاء الاصطناعي: The rules, policies and oversight that shape how AI is built and used, from national principles (such as the UAE Charter for the Development and Use of AI, 2024) to company policies and international reports.
AI safetyسلامة الذكاء الاصطناعي: The field that studies how to prevent AI from causing harm, from everyday errors and misuse to the most serious risks, through testing, safeguards, oversight and rules.
Algorithmالخوارزمية: A set of step-by-step instructions a computer follows to solve a problem. Many AI systems are algorithms that learn their own rules from data rather than having every step written by a person.
Algorithmic biasالانحياز الخوارزمي: When an AI system treats some groups unfairly or reflects stereotypes, usually because of gaps in its training data. Research in 2024 found that many models, including Arabic ones, favoured Western names and customs in Arabic-language prompts.
Alignmentالمواءمة: The work of making sure AI systems do what their designers and users actually intend, and act in line with human values, including in situations nobody anticipated. It is a central focus of AI safety research.
Announced vs operationalمُعلَن مقابل قيد التشغيل: The platform's key distinction. "Announced" or "targeted" means a plan or promise. "Operational" means it is working and confirmed. A gigawatt announcement is not operating capacity. Our tracker shows which is which.
Arabic LLMالنموذج اللغوي العربي الكبير: A large language model built with Arabic as a main language, not an afterthought. Gulf examples include Jais (UAE, 2023), ALLaM behind HUMAIN Chat (Saudi Arabia, 2025) and Fanar (Qatar, 2024).
Artificial intelligence (AI)الذكاء الاصطناعي: Computer software that does tasks we usually link with human thinking, such as understanding speech, translating, summarising or spotting patterns. It is already part of everyday life.
ASI (artificial superintelligence) / superintelligenceالذكاء الاصطناعي الفائق (الذكاء الفائق): A hypothetical AI far more capable than humans in almost every area. The philosopher Nick Bostrom's widely cited definition is "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest". It does not exist today.
Automation biasالانحياز للأتمتة: The human habit of trusting a machine's answer too readily, even when it's wrong. It is one of the most common everyday risks of AI, and one of the easiest to guard against.

B

Benchmarkالاختبار المعياري: A standard test used to compare AI models, for example on maths problems or Arabic reading. Useful but limited: a model can score well on a test and still struggle with real tasks. Companies' benchmark claims should be read as their claims.
Brain–computer interface (BCI)واجهة الدماغ والحاسوب: A medical device that lets a person control a computer using brain signals. It is in early clinical trials, mainly for people living with paralysis. It is medical technology for restoring lost abilities, not a route to "merging with AI".

C

Chatbot / AI assistantروبوت المحادثة / المساعد الذكي: An app you talk to in ordinary language, by typing or speaking, which answers using an AI model. It can be very helpful, but it can also be confidently wrong (see 17).
Cloud / availability zoneالحوسبة السحابية / منطقة التوافر: "The cloud" means using someone else's data centres over the internet. An availability zone is a group of data centres within a cloud region. Keeping backups in more than one zone or region protects data if one site is damaged.
Computeالقدرة الحاسوبية: The processing power used to train and run AI. It mostly comes from specialised chips (GPUs) housed in data centres, and it is one of the main things Gulf countries are investing in.
Content Credentials (C2PA)بيانات اعتماد المحتوى (C2PA): An open, royalty-free standard that attaches tamper-evident information to a photo, video or file, showing where it came from, how it was edited and whether AI was used. Its authors say it is "not a cure-all for misinformation".

D

Data centreمركز البيانات: A large building full of computers, storage and cooling equipment where AI models are trained and run and where online services are hosted. It needs a lot of reliable electricity.
Data residency (data localisation)توطين البيانات: Rules or choices that keep certain data stored and processed inside a country. For example, UAE law restricts storing health data outside the UAE except where permitted.
Deepfakeالتزييف العميق: A fake video, image or voice recording made with AI to look or sound like a real person. Deepfakes are used in scams, so be cautious with urgent requests for money, even if the voice sounds familiar.
Digitisationالرقمنة: Moving information and processes from paper and habit into digital systems. For most businesses this is the necessary step before AI can help: AI needs clean, digital data to work with.

E

Existential risk (x-risk)الخطر الوجودي: A risk that could cause human extinction or permanently and drastically damage humanity's future. Some respected researchers think advanced AI could pose one. Others think it unlikely or far off. The platform presents both views fairly.
ExO 3.0ExO 3.0 (الجيل الثالث من نموذج المؤسسات الأُسّية): OpenExO and Salim Ismail's 2026 update of the ExO model for the age of AI agents. It is set out in *The Organizational Singularity* (v25, June 2026) and keeps the MTP, adding ten characteristics grouped as DRIVE and SHAPE. It is their framework, not the platform's.
Exponential Organization (ExO)المؤسسة الأُسّية (ExO): A concept from the 2014 book *Exponential Organizations* by Salim Ismail with Michael S. Malone and Yuri van Geest. It describes organisations that use technology, communities and shared assets to grow much faster than traditional firms.

F

Frontier model / frontier AIالنماذج الرائدة (نماذج الطليعة): The most capable AI models at any given time, usually built by a handful of large companies. Safety rules and reports often focus on these because new abilities, and new risks, tend to appear there first.

G

Generative AIالذكاء الاصطناعي التوليدي: AI that creates new content, such as text, images, audio, video or code, in response to a request. Chat assistants and image generators are the best-known examples.
GPU (graphics processing unit)وحدة معالجة الرسومات (GPU): A chip first designed for video-game graphics that turned out to be very good at the maths AI needs. The most advanced GPUs are mostly designed in the US, and their export to the Gulf needs US licences.

H

Hallucinationالهلوسة (الاختلاق): When an AI tool states something false as if it were true, such as an invented fact, quote or reference. It happens because models predict plausible text rather than look facts up. Always check important answers.
Human-in-the-loopالإنسان ضمن الحلقة (الإشراف البشري المباشر): A design in which AI suggests but a person approves or rejects before anything important happens. Related terms: "human on the loop" (a person monitors and can step in) and "human out of the loop" (the system acts alone). Gulf regulators increasingly expect the first two for decisions that affect people.

I

Inferenceتشغيل النموذج (الاستدلال): Using a trained model to answer questions or do tasks. Every time you ask a chatbot something, that is inference. It happens billions of times a day, so it drives much of the demand for data centres.
Intelligence explosionانفجار الذكاء: The idea, first set out by the mathematician I. J. Good in 1965, that a machine smarter than people could design even better machines, leading to runaway improvement. Good added "provided that the machine is docile enough to tell us how to keep it under control".

J

Jagged capabilitiesالقدرات المتفاوتة: The way today's AI can be excellent at some hard tasks and surprisingly poor at some easy ones. The *International AI Safety Report 2026* uses this word to describe current systems.

L

Large language model (LLM)النموذج اللغوي الكبير: A model trained on very large amounts of text so it can read and write language: answering questions, summarising, drafting and translating. Most chat assistants run on one.
Loss of controlفقدان السيطرة: The concern that advanced AI could act in ways people can't correct or stop. The *International AI Safety Report 2026* says current systems "lack the capabilities to pose such risks, but they are improving in relevant areas".

M

Machine learningتعلّم الآلة (التعلّم الآلي): A way of building AI in which the computer learns patterns from many examples instead of being given fixed rules. The quality of what it learns depends heavily on the examples it is shown.
Massive Transformative Purpose (MTP)الغاية التحويلية الكبرى (MTP): In the ExO framework, an organisation's big, inspiring reason for existing. OpenExO describes it as "so sweeping and profound that it is always within reach yet always unreachable". In ExO 3.0 it is also meant to guide AI agents' decisions.
Megawatt (MW) / gigawatt (GW)الميجاواط / الجيجاواط: Units of electrical power (1 GW = 1,000 MW). When a data centre is called "1GW", that describes the power it is designed to draw, not how "smart" it is. It is also a plan, not proof that the site is operating.
Model (AI model)النموذج (نموذج الذكاء الاصطناعي): The trained "engine" of an AI system: the result of training, which takes an input (a question, an image) and produces an output (an answer, a label). Chat apps are built on top of models.
Multimodalمتعدد الوسائط: Describes AI that can work with more than one kind of input or output, for example text and images together, or speech and text.

N

Neural network / deep learningالشبكة العصبية الاصطناعية / التعلّم العميق: A neural network is a type of machine-learning system loosely inspired by the brain: many simple connected units that adjust as they learn. "Deep learning" means using networks with many layers. It is the approach behind most of today's AI.

O

Open-weight / open-source modelالنموذج مفتوح الأوزان / مفتوح المصدر: An open-weight model's trained parameters are published so anyone can download and run it. A fully open-source model goes further and also releases its code and training data (K2 Think does this). Open models help local researchers, but they also make safeguards harder to enforce.

P

p(doom)احتمال الكارثة (p(doom)): Informal shorthand for "probability of doom": one person's estimate of how likely it is that advanced AI leads to human extinction or a permanent catastrophe. The numbers vary enormously, and nobody's figure is reliable enough to treat as a fact.
Parametersالمُعامِلات: The internal numbers a model adjusts during training. They are often counted in billions ("a 7-billion-parameter model"). More parameters usually means more capability and more cost, but size alone doesn't guarantee quality.
Personal data protection law (PDPL)قانون حماية البيانات الشخصية (نظام حماية البيانات الشخصية في السعودية): A law setting rules for how organisations collect and use personal information. Most GCC states now have one, including the UAE (Decree-Law 45/2021) and Saudi Arabia (fully enforced since Sep 2024). According to a 2024 legal review, Kuwait has sector regulations but no comprehensive law.
Promptالموجِّه (نص الطلب): The question or instruction you give an AI tool. Clear prompts with context ("explain this to a 12-year-old, in Arabic") usually get better answers.

R

Reasoning modelنموذج استدلالي: A model designed to work through a problem in steps before answering. It is often better at maths, logic and coding, but slower and more costly to run. Examples from the region include Falcon H1R (TII) and K2 Think (MBZUAI/G42).
Red-teamingاختبار الفريق الأحمر: Deliberately trying to make an AI system misbehave, for example by coaxing it into giving dangerous advice, so that weaknesses are found and fixed before the public meets them.

S

Sovereign AIالذكاء الاصطناعي السيادي: A country's effort to have its own AI capability (data centres, models and data kept under national rules) rather than relying wholly on foreign providers. In practice, most advanced chips are still licensed from abroad, so "sovereign" is usually a matter of degree.

T

The (technological) singularityالتفرّد التقني: The hypothesis that AI could one day improve itself so fast that the future becomes impossible to predict. Vernor Vinge named it in 1993. Ray Kurzweil popularised it and dates it to 2045. It hasn't happened, and whether it ever will is disputed.
Trainingالتدريب: The stage in which a model learns from large amounts of data, usually over weeks or months on thousands of specialised chips. Training a leading model is expensive and energy-intensive, and it is done once per version.
Training dataبيانات التدريب: The text, images, audio or other material a model learns from. Gaps or imbalances in this data, such as too little Arabic or Gulf content, show up later as weaknesses or bias.