Slides from my talk at aginext 2019.
In this session we’ll examine the AI capabilities available today in simple layman’s terms and explore how these will be used to augment and shape the agile world of tomorrow.
Artificial Intelligence (AI) has catapulted us into a brave new world of self-driving cars, delivery drones and talking robots. A combination of AI technologies including advanced machine learning, deep learning and natural language processing are now set to change the way we build and deliver products enabling us to build smarter software faster.
Imagine a world where product backlog prioritisation and feature discovery are aided by unbiased data analytics from AI systems. Self-learning products and adaptive user interfaces will automatically respond and adapt based on data driven analysis of real time user behaviour. Trouble shooting and recovery from production outages is accelerated and assisted by AI operations bots. This convergence of AI systems with the agile world will offer teams unprecedented visibility into their work and their products.
33. Alpha Zero
2017
Alpha Go
2016
Deep Blue
1997
Watson
2011
custom algorithms
tree Search
natural language
processing
deep QA
supervised
machine learning
unsupervised
machine learning
monte carlo tree
search
36. AI powered expert tooling
NLP will provide
● enriched requirement models
● better code generation
● improved test automation
ML will provide
● more predictable timelines
● test coverage recommendations
42. Self Healing Systems
machine learning analysis of
• operation logs => root cause and fixing outages
• network traffic
• the OT => neutralise cyber attack vectors
43. evolution of the user interface
Command
Line
Interface
Graphical
User
Interface
Natural
User
interface
44. Natural user interface design
Intuitive systems that respond to speech and gestures
Immersive interfaces
46. Self learning systems
Convergence of machine learning with web analytics will provide a deeper
understanding of our users
● AI assisted personalization
● Automatic persona discovery
● Persona refinement
● Intelligent feature usage reporting
● cohort analysis
48. software developers from coding fixed business logic to training machine learning
models
AI powered tooling for code generation
software testers Automatic test case generation
AI assisted test scenario planning
UX designers Natural User Interface Design
AI assisted persona mapping
operations ChatOps
Self healing systems
product owners AI assisted backlog prioritisation
Feature / persona usage insights
agile teams AI assisted planning through expert systems
49. What will a product team
look like in the future?
63. proposing the AI manifesto
“We recognise that artificial intelligence will have a significant impact on our
society. We pledge to only build AI systems that will contribute to the greater
good.”
❖ AI systems must treat all people equally, they must not create
or reinforce unfair bias
❖ AI systems must not be weaponized
❖ AI systems must be socially beneficial
❖ AI systems must respect privacy
Hinweis der Redaktion
pace of digital transformation is rapidly accelerating
Gordon Moore, founder Intel in 1965
unprecedented
apply metric to fuel consumption - 4 litres drive round earth
40% AI Start Ups
AI is the ability of software and computer-controlled robots to perform tasks that require HUMAN LEVELS Intelligence
LEARNING
REASONING
SELF-CORRECTION
INDEPENDENT DECISION MAKING
https://coggle.it/
look ahead procedure creates a branching tree of possible game futures rooted in the current state of the board
200 million moves per second
it can only play chess , nothing more
look ahead procedure creates a branching tree of possible game futures rooted in the current state of the board
200 million moves per second
it can only play chess , nothing more
long standing American TV Game Show
MASSIVELY PARALLEL HYPOTHESIS GENERATION AND EVALUATION TASK
WIKIPEDIA, BOOKS,
GENERATES A WIDE RANGE OF POSSIBLE ANSWERS
GATHERS EVIDENCE
DEVELOPS A LEVEL OF CONFIDENCE ABOUT EACH ONE
20 MS chunks and compares recordings , uses statistical models to determine each word
TENSE
VERBS
SYNOMYNS
GRAMMER RULES - LEXICON TO DETERME THE MEANING
WATSON IS A GENERAL PURPOSE PLATFORM FOR ANALYZING UNSTRUCTURED DATA LIKE TEXT
3000 years old
number of configurations of the board is near infinite
masters can't articulate how they play, its a feeling
Alpha Go - trained with thousands games using supervised machine learning....
learned to recognise winning patterns an
THE GOAL OF MACHINE LEARNING IS TO ENABLE COMPUTERS TO LEARN FROM HISTORICAL DATA ON THEIR OWN ......... ............
.... AND TO PREDICT FUTURE OUTCOMES WITHOUT HAVING TO BE PROGRAMMED TO DO THIS
SET OF TRAINING DATA IS FED IN
MANY DIFFERENT ALGORITHMS THAT BREAK IT INTO DIFFERENT FEATURES AND LAYERS - called FEATURE DETECTION
THE MORE LABELLED DATASETS THE MORE ACCURATE IT GETS
CONVULUTION NETWORK INSPIRED BY THE ARCHITECTURE OF THE VISUAL CORTEX IN HUMANS
Neurons only see a tiny window of pixels , lots of neurons looking at the same window .... then shifted slightly
Strength of the connections between the neurons is trained
with BACK PROPOGATION
A TRULY GENERAL PURPOSE GAME PLAYING AI PROGRAM
Trained against itself only ... no human bias
Unsupervised learning is very much how children learn lanugages ... from listening and observing in an unsupervised fashion
CONWAYS LAW
HISTORICAL TEAM ESTIMATES
HISTORICAL ACTUALS
data feeds
requirements, releases, metrics,
Alert teams for potential issues and possible fixes
advised testing strategies
guided agile cermonies
overall reported improvements in test case prioritisation and coverage .... BUT ... HUMAN FEAR OF MACHINES
CHAT OPS - unifying communications DEV , TEST ,through to DEPLOY and Resolving OUTAGES
2013 Stadford researcher Michael Kosinski released a white paper that demonstrated how AI can predict human personalities and traits by analysing FB likes
Cambridge Analyticia used this
Algorithm is not difficult, data is the key
87 million FB users ... through personal friend data which was illegal
US Election, Brexit - tone of language, personality style, political views
Mark Zukerberg apologised before US Congress for how 3rd party apps were able to gather this data
REINFORCED DISCRIMINATION
2017 Virginia computer science professor Vicente Ordonez -
Machine Learning software trained on datasets don't just MIRROR those biases they AMPLIFY them
Amazon internal recruitment tool... score candidate CV's
10 years of resumes for current employees
favoured masculine language
penalised female schools