AI, IT Architecture, and What to Expect in 2023-2030
The global AI market is predicted to seven-fold in 7 years, 2021-28, and the US National Security Commission on AI expects it to reorganize the future life of the world. It’s an easy guess that reshaping the field of AI and IT architecture itself is no exception.
The vulnerability of long HW-supply chains, laid bare by Covid and Putin’s War, is mitigated by insourcing (scroll down to “3.” here) and friendshoring ranging from Arizona and Texas to Saxony and Anhalt or Sweden and Czechia.
This transition takes time nevertheless. Meanwhile, HW bottlenecks amplify the demand for simple robust algorithms already needed in green IT, tiny ML, small data, and edge computing including edge ML.
Google Cloud VP Andrew Moore points out that much of AI is now converging behind intuitive single points of entry for users, e. g. conversational AI, and performs increasingly sophisticated tasks, while the core (or business-logic layer) is becoming a set of intelligent problem-solver engines foundational to our ways of doing business. Moore lists five key areas in corporate AI for the coming years:
- Use Case-Driven AI, integrating R&D and product-focused enterprise-grade services
In my opinion, this shall be top-of-the-agenda in your ITA and EA.
- Data harmony and interoperability to clean and unify potential training data
IMO, good data and cross-connecting disparate data is essential, be big data or small or tiny.
- Responsible and explainable AI as key part of design/training and refinement/retraining.
- Democratization of corporate culture, tools, and interfaces to abstract the complexity
IMO, these three boost each other, although any of them can pave the way for the rest.
- Multicloud from day one to stay scalable and forward compatible
IMO, the Vendor-lockin antipattern has been, is, and will be a serious risk. Remember that tools are available for managing and monitoring multiplatform environments, and some of those cover Edge devices as well.
The SW Engineering Institute of Carnegie Mellon (we use some of their terminology in our Agile Architecture courses, to prevent long hot “same thing twice over” discussions between attendee jargons) emphasizes the incorporation of ML systems and training data into SW architecture.
Here’s a brief summary of their six key points:
Include the ML-model lifecycle and architectural dependencies of AI components, stress architecturally significant requirements in ML (both functional and quality attributes) , use monitorability mechanisms to even handle uncertainty of ML systems, have your data pipeline cover different rates of change in data and ML-retraining/redeployment, co-architect versioning, ML components, and the system at large.
Summing up, the two-way multiple-lane street between AI and architecture is getting even busier than in the recent years, just like architects do. Unlike this sign, tradeoffs between architecturally significant requirements are not just three-dimensional; rather, they’re n-dimensional. : )
Milan and Informator collaborate since 1996 on architecture, AI, modeling, UML, rules, requirements, and design. You can meet him this year at these courses in English or Swedish (remote participation is encouraged, and classroom participation offered too – as for November 2022) :
(on demand: Modular Product Line Architecture )Nyckelord: Prediction, SWA, AI use cases, Models, MLops, GreenIT, business-driven, data pipeline, efficient algorithm, AI, Agile, architecture, UML, it architecture, ITA, EA