ai as a service
TRANSCRIPT
AI as a ServiceNew Major Technology Trend
Asher [email protected]
HUstart MeetingMay 17, 2017, Jerusalem
About Myself● Close to 40 years of experience in the software technology field● VP Technology, Chief Technology Advisor Group, NDS (2010 - 2012)● Distinguished Engineer, Office of CTO, Cisco Engineering (2012 - 2016)● Today CTO at IRKI● Focusing on software product line strategy
○ Lean Startup○ Wardley Maps○ (Strategic) Domain-Driven Design○ Cynefin○ Promise Theory○ Serverless Architecture○ ...
Why Technology Matters?
EquipmentToolsFrameworksOSMiddlewareServicesTechniques
EquipmentToolsFrameworksOSMiddlewareServicesTechniques
This is NOT your core!
Using Technology Radar
Thoughtworks Technology Radar
The ThoughtWorks Technology Advisory Board, a group of senior technology leaders in ThoughtWorks, creates the radar. They meet regularly to discuss the global technology strategy for ThoughtWorks and the technology trends that significantly impact our industry.
Themes For the Last Edition● Conversational UI and natural language processing● Intelligence as a service● Developer experience as the new differentiator● The rise of platforms● Pervasive Python
Mentioned in this Edition● Techniques
○ Conversationally aware APIs - ASSESS■ Amazon Alexa■ Google Voice
○ Back-end for Front-end - TRIAL● Tools
○ Apache Airflow - TRIAL○ Scikit-learn - TRIAL○ Amazon Recognition - ASSESS
● Platforms○ api.ai - ASSESS○ wit.ai - ASSESS○ Cloud-based Image Comprehension - ASSESS
■ Amazon Recognition■ Microsoft Computer Vision API■ Google Cloud Vision API
○ Nuance Mix - ASSESS○ Voice Platforms - ASSESS
■ Amazon Alexa■ Google Home
● Languages and Frameworks○ Caffe - ASSESS○ DeepLearning.scala - ASSESS○ Keras - ASSESS○ Knet.jl - ASSESS
Azure AI
Google AI
IBM Watson
AWS Open Source AI @ OSCON (Adrian Cockroft)
What’s Your AI Strategy?
ignore AI
Irrelevance in a couple of years
embrace AI
A lot of $$ paid to people who speak language you cannot comprehend and produce models they do not understand themselves
Your Product or Service
Value Chain + Evolution (Wardley Maps)
Your AI secret sauce
AI as a Service
G C P U
Cloud Platform
AI scientific breakthrough
Your Customer Needs
Infrastructure Optimization
Your Product or Service
Value Chain + Evolution + Movement
Your AI secret sauce
AI as a Service
G C P U
Cloud Platform
AI science breakthrough
Your Customer Needs
Infrastructure Optimization
Your Product or Service
Value Chain + Evolution + Movement
Your AI secret sauce
AI as a Service
G C P U
Cloud Platform
AI science breakthrough
Your Customer Needs
Infrastructure Optimization
Your Product or Service
Value Chain + Evolution + Movement
Your AI secret sauce
AI as a Service
G C P U
Cloud Platform
AI science breakthrough
Your Customer Needs
Infrastructure Optimization
Innovate-Leverage-Commoditize
Strategic Domain-Driven AI Design
Generic AIOptimization
Pre-trained Models
Secret Sauce
DDD AI Architecture
Pereferrial Adapters Application Services
Dynamic State Machines
Conversations
Application-level Stimulo
Domain Services
Command/Query Requests EventsRaw A/V/T Inputs
Dynamic Classifications
Anomaly Detection
Event Processors
Feedback and Training
Historical Data
ML/DL In Computational Context
Computation: Input Data X Policy X Current State → Result X New State
value range
# of
pat
hs
exponential growth of complexity
ML/DL In Computational Context: Pure Function
Input Data → Result Simple Function (if-then-else)Pattern MatchingComplex AlgorithmStatistical AlgorithmHeuristicsMachine Trained FunctionNeural NetworkDeep Neural NetworkConvolutional Neural NetworkGenerative Adversarial Networks
ML/DL In Computational Context: Functor
Input Data X Policy → Result Higher-Order FunctionFunctor (map)Monoid (fold, reduce)Monad (flatMap)Applicative (apply)Machine Trained Higher-Order Function?… ?
ML/DL In Computational Context: State Machine
Input Data X Policy X Current Sate → Result X New State State Monad (FP)Concrete/Abstract Class (OOP)Final State MachineHarel/UML StatechartRecurrent Neural NetworkLong Short Term Memory Network
Beware of “Black Swan” and “The Turkey Problem”
volum
evelocity
varie
ty
Variety is the most time sensitive axis of the V3 Cube
edge computing
simulation?
Complex Adaptive Systems are not Casual
Disorder
“Don’t trust us (humans), because we will disappoint you”
AI Trinity
Domain ExpertData Scientist Software Architect
Domain ExpertData Scientist Software Architect
Domain Expert
Data Scientist
Software Architect
“Consensus is poisonous for innovation”
D. Snowden
The last slide