This February 2021 archive item used a hierarchy of needs to explain why an AI project depends on work that happens before model development. Its foundation was reliable data collection and infrastructure.
The sequence moved from collecting data to storing and transporting it, then cleaning and analysing it. Features, labels and simple models came next. Those early models provide a baseline against which a more complex approach can be judged.
The original article also highlighted bias and the need to communicate uncertainty to stakeholders. A sophisticated model cannot compensate for unrepresentative inputs or an unclear objective. For product teams, the framework is a way to identify the next missing capability: sometimes the most valuable work is fixing collection or making data accessible, rather than training another model.
A March 2021 archive item revisited reporting by The Wall Street Journal about Engineer.ai and the extent of automation in its app-building service. The report questioned how much work was performed by software and how much depended on developers.
The February 2021 archive included a Tangled AI article on preparing a small business for AI adoption. Its main theme was to begin with a business problem and readiness, rather than with enthusiasm for a particular platform.
The original November 2022 explainer introduced three terms often grouped together under artificial intelligence: narrow AI, general AI and superintelligence. They describe different levels of capability, rather than three interchangeable labels for the same technology.
Tell us about your project — we'll tell you honestly whether we can help.