How well a dataset fits the specific question or purpose at hand. Highly relevant data answers the question directly; less relevant data may need to be combined with other sources or filtered down.
The dataset's relevance to the pricing question was high—every record had the variables they needed.
The cost of acquiring and verifying the original data—either gathered in-house or sourced from a third party. Distinct from buying an already-collected dataset (cost) or analyzing it (analysis cost).
The research cost of running their own survey was steep, but it gave them data no competitor had.
The design, development, and deployment of AI systems in ways that prioritize ethical considerations, fairness, transparency, accountability, privacy, and inclusivity. It ensures that AI technologies benefit society and mitigate potential harms, such as bias, misuse, or unequal access.
"A healthcare company implemented responsible AI practices by refining its chatbot to prevent hallucinations, ensuring it provided accurate medical advice and redirected complex queries to licensed professionals."
A strategy for 'grounding' AI tools in accurate facts by combining generative and retrieval models. RAG uses a pre-trained model to retrieve relevant information from vetted and verified documents (a knowledge graph) and then uses a generative model to output content based in fact.
"Our chatbot was confidently giving incorrect answers to our patients. Since we're in medicine, that's not acceptable, even for casual use, so we used a RAG strategy so that the chatbot retrieve facts from medical texts but then makes them easier to read for lay users."
Key sources of income for a business model.
"The company's revenue structure included subscriptions, in-app payments and professional service fees."
Using machines to mimic repetitive human processes performed on computers (such as changing file formats or sending out meeting invites). RPA is ideally applied to free up human capacity for more high-value opportunities—and sometimes is used to reduce jobs. RPA does not require machine learning, artificial intelligence, or other advanced technologies and thus is often a first step in automation and augmentation efforts.
“Our company is doing a ‘build your own bot’ RPA project, where IT is helping people automate mundane computer work.”
The highly specialized machinery used to make computer chips. Controlling access to this equipment (tooling, servicing and maintenance) is another way to limit a rival's ability to build advanced chips independently.
Even if China ramps up its own chip factories, without access to the right SME, those factories can't produce chips advanced enough to compete.
A type of machine learning that combines a small amount of labeled data with a large amount of unlabeled data, allowing the model to improve performance without requiring full labeling.
"We used semi-supervised machine learning with a small set of labeled medical scans and many unlabeled ones to improve detection accuracy."
A technology layer that enables developers to configure, monitor, and manage interactions between various services or microservices designed with interoperability in mind. Users can often choose various options from different providers to create a fully-customizable 'service mesh' that has all the functionality they require.
"The service mesh of the online store included a no-code website builder, inventory management, and payment microservices, each from different providers but working seamlessly via APIs."
IT that is not managed by the IT organization but used by employees to do their work (or otherwise touches company data).
"I copied all of that into a Google Doc so we could work on it together using their new AI tools. It'll be much quicker, just use your personal email account and don't tell anyone we're using shadow IT."
An evolution of monolithic architecture where resources are consolidated into a central organization that serves 'internal clients' in the company.
"Our company's shared services include our facilities, IT, and tax/accounting functions that serve our various operating companies and departments."
In a simple problem, both the problem and solution are known.
"Increasing the visibility of our brand seems like a simple problem—we just need to advertise more or better."
Using AI systems to mimic real-world processes or environments for testing, training, or experimentation.
"We ran a simulation of customer traffic to plan store staffing."
A (usually hypothetical) point in time beyond which predictions of the future are not likely to be accurate. "The Singularity" often refers to a technological singularity, where AI and other advanced tech reach a state so advanced that humans and society are fundamentally and irreversibly changed.
"Some thought leaders say the concept of needing to 'work for a living' will be outdated when The Singularity arrives (since machines will do so much)—while others point out that AI-generated value will not be distributed evenly unless we change our policies."
Common points of discussion or sharing within a community or network that create or reinforce relationships; for example, stories, pictures, tips, or comments.
"The social objects of Facebook include posts, events, and comments."
A set of digital tools and components provided by a company for use by third-party developers who want to make applications that are compatible with a specific platform, operating system or framework.
"Apple offers several different SDKs for developers wanting to create apps for iPhone."
The strategy and process for creating or updating computer programs, taking into account every aspect of the software's use, including planning, deployment, and eventual retirement.
"Originally, our developers' software development life cycle was focused on long-term, well-defined functions; since it didn't always account for how quickly we have to change things in today's market, we've also added in lean development and other approaches."
A tool used by futurists to explore the question "what if?" by creating other-than-real-world scenarios, both utopian—and dystopian. Speculative fiction helps futurists explore and explain how key technologies or other trends could change everyday life.
"To explore AI's impact on our organization, we wrote speculative fiction to imagine how users would interact with digital 'agents' in addition to our existing employees."
A focused, accelerated work period for a team, intended to complete a substantial software development goal.
"Once we finish this sprint on the new user interface, we'll move to the next core part of the app."
A step after a sprint is completed, where all involved share feedback and learnings to improve the experience and efficiency of the next sprint.
"Always keep a note of all the issues and challenges you faced, as we can use them during the sprint retro to learn how to make the process better next time."
Being able to fine-tune, adjust, correct, or otherwise guide an AI system to operate more in line with the expectations and ethics of its owner or user. Traditional analytical algorithms are generally easier to steer—and explain—than machine learning and generative AIs.
"While we've had success steering AI for content recommendations to our users when it was based on tags, we're concerned about adding generative AI to that because it will be harder to adjust."
Information that is already categorized and organized into clearly defined fields, such as rows and columns in a database, making it straightforward for computers to search and analyze.
"Our sales transactions are stored in a database with dates and amounts—they're structured data."
A type of machine learning where the model is trained on labeled data (inputs paired with correct answers) so it can learn to predict outcomes for new inputs.
"We used supervised machine learning on thousands of labeled emails to train a spam detector."
The creation of plausible, factually-grounded data for training of machine models rather than, or in addition to, importing real-world data. Synthetic data use is intended to reduce bias, quickly train models, and improve accuracy. For example, synthesizing demographically-accurate data about the population of a university might be preferable to risking leaks of individuals' real addresses, grades, or other private information.
"The credit-scoring firm introduced synthetic data that corrected for inherited privilege, to counteract societal biases against women."
A backlog of necessary tech updates, caused by postponing updates for fear of disruption or cost.
"With larger, older companies having decades of tech debt, it's no wonder that newer organizations can move and adapt more quickly to changing circumstances."
The broader network of companies, platforms, tools, and communities that make up a technology industry or sector.
Our financial services startup chose to work in New York to benefit from being in a strong tech ecosystem that's also familiar with banking, where we can access talent, investment, and partnerships more easily.
Highly specific and rigorous reference and support materials for technologists and other experts
"We try to make sure our technical documentation is updated as often as our user docs are—it makes sure we can bring in new developers quickly. We're able to do that more easily now with generative AI."
a professional with both marketing expertise and technical skills, allowing them to leverage technology and data in marketing strategies and campaigns.
"We hired a technical marketer to turn implement our brand strategy into in-app messages, chat, and marketing automation tools."
The use of advanced technology — like surveillance cameras, facial recognition, and AI—by a government to monitor and control its own population.
The contactor we're looking at makes tools that can be uses for security, but which could also lead to techno-authoritarianism to surveil and repress dissident citizens—so we've decided our company can't do business with them.
Job loss due to automation, augmentation, or other efficiency gains made by new tools. First noted during industrialization, technological unemployment remains both a fear and a reality for workers. While new jobs are often created by tech, workers cannot always transition to them quickly, and it is not certain that digitally-displaced jobs will be replaced with new opportunities.
"I am experiencing technological unemployment—my editing job was replaced by text synthesis tools like ChatGPT and robotic process automation."
The tools and infrastructure needed to use a dataset—storage, processing power, specialized software, or AI models. Large or unstructured datasets may demand expensive setups before they're usable.
The technology needed to process the video data—GPUs, cloud storage, custom AI models—doubled the project's budget.
A combination of technologies and systems used to create value in a company or app.
"Our tech stack is mostly made up Microsoft products like databases, servers and development tools, but we have AI models from about twenty different companies use."
A setting that controls how random or predictable a model’s responses are: low values make outputs more focused and repetitive, high values make them more creative and varied.
"When we set the temperature to 0.2, the chatbot stuck to safe, generic answers; at 0.9, it offered more imaginative suggestions."
How well a dataset's update cadence matches the timeline of an investment strategy. Annual reports work for long-term value investing but not for high-frequency trading. Mismatch can make even good data useless.
Their time frequency of investment strategy ruled out monthly data—algorithmic trading needs millisecond refreshes.
A set of tools used together, sometimes in a specific sequence, in order to perform a complex action or process.
"Creating updates for our software products is so much faster and more reliable now that we use a consistent DevOps toolchain to develop and test new versions before release, make sure they are secure, and monitor their performance.”
A modern AI design that looks at all the words (or parts of an input) at once and figures out which parts should pay attention to each other. This makes it possible to build powerful language and vision models like GPT.
"Modern chatbots like ChatGPT are built on transformer models that can understand and generate text at scale."
A modern AI design that looks at all the words (or parts of an input) at once and figures out which parts should 'pay attention' to or affect each other. This makes it possible to build powerful language and vision models like GPT.
"Modern chatbots like ChatGPT are built on transformer models that can understand and generate text at scale."
How well a dataset performed in a pilot or test before full commitment—accuracy, usefulness, integration ease, and whether it delivered on its promised value.
The trial quality looked promising on paper, but the dataset failed when integrated with their production pipeline.
A shorthand for describing the effort and/or complexity of a task in a larger project.
"The t-shirt size of the bug fix for capitalized text is an extra-small, but integrating 'sign in with google' is a size large."
The mental discomfort or revulsion some humans experience when viewing machine-based experiences or content that realistic, but 'not quite real enough,' such as some AI avatars, generative AI chatbots, or virtual reality.
"This synthesized video is creepy—I wish they had just avoided the uncanny valley and paid a real person to present."
When a model is too simple to capture the underlying patterns in the data, leading to poor performance on both the training data and new, unseen data.
"Early AI face generators made blurry, indistinct images because they were underfit—they hadn’t learned enough detail and parameters for faces."
A situation where certain categories of facts, groups of people, languages, experiences or other information are missing or too scarce in the data used to train AI, leading to biased or less accurate results for them.
"Voice assistants often struggled with accents and dialects because of underrepresentation in their training data."
The practice of identifying existing mental models, mindsets, and biases and then selectively 'letting go' of them to make room for new thinking.
"We're unlearning old ways of thinking about our relationship with customers as they move from being consumers of what we make to becoming co-creators of digital content and data."
Information without a fixed format, like text, images, or audio, which requires extra processing for computers to interpret.
"Customer emails and social media posts we discover are unstructured data; we use AI tools to analyze their sentiment."
A type of machine learning where the model finds patterns or groupings in unlabeled data without being given the correct answers—sometimes identifying new insights.
"We applied unsupervised machine learning to customer data and discovered natural clusters of buying behavior."
A growth-at-all-costs strategy focused on maximizing user engagement/usage metrics (e.g., daily active users). In AI and social platforms, it can include relaxing guardrails or adding highly engaging features to boost usage, often criticized as prioritizing engagement and revenue over safety or well-being.
"'If you want your ChatGPT to respond in a very human-like way, or use a ton of emoji, or act like a friend, ChatGPT should do it (but only if you want it, not because we are usage-maxxing),' said Sam Altman."
A purpose for applying a technology to a real problem or opportunity. In business, a 'use case' is a way to ground advanced technology discussions in practical value for the organization or its end users.
"Sure, ChatGPT sounds cool, but right now, there's not a practical use case for it in our regulated environment. We expect to see one soon, but our investors demand clear use cases to justify allocating labor and cash to it."
A final phase of a software development process wherein users determine whether the software performs as intended and expected.
"The team thought they were done with the 'save' dialog box until they got to UAT and found that it didn't show up on Android mobile devices."
The sum of all interactions a user could have with an app or other digital offering. Also can refer to the discipline of examining and working to improve the way that human users must interact with a digital tool or website to get value from it.
"I stopped using that app because the UX was terrible. I could never figure out how to get back to the menu, and it kept logging me out."
Examples of use cases for a product, usually collected from the real world via a feedback loop; essential to Continuous Improvement and Agile development processes.
"With structured user stories, we have a better idea of how our product is actually being used in the real world, by how many people, and how to improve it for our users."
Value Propositions specify how a business's offerings will solve a particular user or customer need and why they are more attractive than other solutions.
"The new app from AcmeCo has a very compelling value proposition—it's a learning app to help people quickly pick up new concepts and skills using techniques informed by brain science to save time and repetition."
A database optimized for storing and searching embeddings (numerical representations of text or other data that make it easy for an AI system to compare similarity).
"We stored embeddings in a vector database to power semantic search across thousands of documents."
A simplified mathematical fingerprint of information (like a word, image, or sound) that a computer can compare with others to find similarities.
“We converted product descriptions into vectors so the AI could recommend similar items to shoppers.”
In the context of AI/digital factories, velocity is a reflection of a team's ability to communicate efficiently. It refers to the speed at which the software can be adapted to changes in security environments, the needs of users, or business updates.
"One of the best measurements of a digital factory's success is velocity—and we have improved ours ten-fold since switching to an agile methodology."
An informal, quick approach to developing software via AI chat with generative and agentic AI tools, which create and update the code and resources needed to accomplish the user's goals.
"Troy vibe-coded the entire app in a weekend to show the concept, and it looks really cool—but it doesn't really work great and I'm concerned about the security.
Virtual reality is technology which immerses a user into a computer generated 3D world, often accessed through VR headsets like Meta's Oculus or HTC's Vive Pro.
"Meta's Oculus devices can be used for VR apps like video games, or for accessing their version of the metaverse."
A traditional, linear method of software development where each phase of the project cascades into the next, and all movement is in a single direction toward a 'finished' product that (hopefully) requires no updates or revisions.
"For compliance reasons, we're using a waterfall approach for this project, so we need to make sure we have all our features planned out before we start development—we won't be able to go back and change them easily."
Early, subtle indications of progress sought by futurists, investors, and ethicists to identify or validate trends. Seen on their own, weak signals may not appear relevant, but can show patterns and directionality when analyzed through a particular lens. Fundamental research breakthroughs (like initial OpenAI GPT models) and early adoption of apps or technologies (like Facebook's rollout in universities in the early 2000s) are examples of weak signals.
"The venture capital board regularly invited guest speakers and researchers to help them make sense of weak signals in the marketplace so they could invest wisely in the right kinds of startups."
In a wicked problem, the problem is not fully known, nor is a solution.
"Climate change is a 'wicked problem'—it involves numerous causes, stakeholders who don't always agree, and effects that are spread across the globe. Worse, solving one aspect might reveal or create other problems."
Another term for broadly-applicable general AI, or sometimes artificial general intelligence (AGI). Contrasts to narrow AI, which is purpose-specific.
"OpenAI's ChatGPT tool can be thought of as a wide AI because it can handle almost any human language prompt."
A map of a webpage(s) or screen(s) showing the visual structure of its functions, abstracted to the most essential information. Primarily used for planning user interface layout, content placement, and navigation through the site, service or app.
"We wanted to see how all the pages on our site will be connected to each other before we placed the content, so we asked our developer to create a wireframe."
An early neural network model that learned word embeddings by predicting surrounding words in a sentence.
"Word2Vec made links like 'king' minus 'man' plus 'woman' is close to 'queen'—and revealed many biases in our language and training data."
A method that maps words into vectors (numerical representations in AI models) so that words with similar meanings are placed close together.
"Word embeddings let the AI understand that 'dog' and 'puppy' are related concepts."
