AI, Architecture and Machine Learning

Introduction: Abstraction levels, AI, ML’s past and present

• Abstraction levels: architect versus devs
• AI apps overview: pushing up the bar for intelligent systems
• The evolution of ML apps:  from slow and auditable, to fast smart and opaque. Despite ML’s roots (rule induction and mining) in the successful decryption of unbreakable ciphers, its near future seems encoded as weight values in deep neural networks. Whereas for example black-box flight recorders clarified the chain of events & decisions in past emergencies, ML has now, paradoxically, landed in black boxes that hide opaque logic.
• A short exercise
• Predictions for the next 10-18 months: AI investment, and tech challenges to cope with.
• A short exercise

Consequences in architecture, analytics, HMI, RT

• NFR and Quality Attributes, businesses where “better on average” is no longer “good enough” as it was in recommender engines or marketing to layman consumers
• Three categories of ML
• Deep Learning
• The comeback of Data Architecture in Software Architecture
• The increasing overlap  RT – DB apps, and edge ML & computing
• The two-way street: Architecture for AI, and AI for Architecture.
• A short exercise

The role of architects

• What is, and is not, in the job description of an architect
• Perspective (broad, integrated) compared to others (dev, researcher, prime mover)
• The systems portfolio and digital twins (“avatars”) cross-connecting perspectives
• Architectural tactics, thinking, abstraction level, and background
• A short exercise
• Agile architecture, variability, and edge ML.

Tackling the leveled-up tradeoff

• Top-down: a framework of corporate values and procedures to boost V&V and transparency
• Bottom-up: tech and practices in IT to build in transparency mechanisms upfront (visualization, explanation, traceability, etc.)
• Explainable AI (XAI) “Big bang,” via research into self-explanation capabilities for deep neural networks
• Corporate “think big, start small,” XAI via hybrid AI.