Carina Zona
Algorithms have consequences on people
MarIO
Target:
If we wanted to figure out if someone was pregnant, even it they didn't want us to know, could you do that?
As long as we don't spook her it works
Shutterfly thinks you had a baby.
Inadvertent Algorithmic cruelty
Works most of the time, but not all the time
We cannot intuit inner state, emotions, private subjectivity
Accidental algo run-ins
- Out of the user's control
Tools used to make film, the science of it, are not racially neutral
- optimized for white skin with Shirley cards
If a machine is expected to be infallible, it cannot also be intelligent - Alan Turing
Using a black box for decisions is regressive, to fairness and equality.
- Project the likelihood
- minimize negative consequences
- Not unlike the hippocratic oath
- Build in recourse for a wrong conclusion
- disclosure of limitations
- track bias in your own black box
Do not build without understanding consequences to others
@searls
- Tests make big objects harder to manage
- test combo of all params
Test can do 3 things:
- sets up (Given)
- invokes (When)
- verifies (Then)
rspec-given
Sandi Metz's Squint test
- Unfocused tests
- Too realistic tests
- slower
- take time to write/change
- so many moving parts
- Redundant code coverage
- Careless mocking
- Treats symptoms of test pain, not cause
- Application Frameworks
- encourage only integration tests
- Useless messages
- Bad messages provide wasted time
- judge message quality of tools
- Slow loops
- limits the tasks you can do in a day
- Painful Data
- Test Durations
- False negatives
- Unfinished tests
- erode confidence in tests
- redundent coverage, slow tests cause false negatives
TL;DR Fewer integration tests
@testdouble
- Don't hard code labels
- Support extra types of characters
- Not cheap/easy, but beats the alternative
- Authentication is not Authorization is not security
- Leave room for encryption
- Be able to delete users
- Delete data after 7 years or so
- needs to be more useful than stack overflow
- They are not state secrets, they are subtle self promotion
- Good for dev on-boarding
- Good for build procedures
- what your software is trying to get people to do
- Software that creates good behavior w/ minimal user effort
- You're never going to fix it later
- Find actual users
- Don't tell them how to do it, see how they understand it themselves
- if you can't hire an expert, become a student
- Not everyone uses the retina screens we do
- We're only temporarily able-bodied, so do it for yourself
- Style guides
- brown bag lunches
- pair programming best practices
- add tests for accessibility and usability
- cultivate diversity and representation on your teams
- ask questions
Section 508: no money if not ADA compliant
Advice for winning: make it about money
Mostly live coding an interpreter to understand types, which is here:
github.com/jcinnamond/trump
Bad estimates are not unique to software
- Customers reasonably expect estimates
- Communicates the investment
- Communicates progress
- Communicates scope
- Bad estimates communicate bad things
- unrealistic expectations
- way too much
- allows for micromanaging
- Like guessing time to the airport if you don't know where it is or if you even need to fly
- Nobody cares about feature estimates
- Only useful in aggregate
- Only care about when users can use stuff
- Estimates are probabalistic
- Bad assumptions
- scope creep
- Unknown complexity
- Rework
- sometimes there's pressure
Developers are optimists
A lower estimate doesn't make it less complex
Don't chase false precision
What goes into estimate:
- Complexity of the task
- We're bad at estimating time on a task
- Skill of dev
- De-moralizing if personal
- Consistent
- time on task
- Devs are pretty good at estimating this
Estimate complexity and let time sort itself out == point
- Everything else being consistent, the number of points per iteration becomes the most useful
Key point: decompose problem into smaller stories
The above doesn't help a one-off estimates
Fake your velocity in the beginning.
Don't chase precision here.
What you'll get is a range
Round point estimates upwards
Clearly describe the limits to your estimate
Worst Ideas Evar -- Aaron Patterson
http://github.com/thagomizer/stupidideas thagomizer.com
Augmented reality!!!
Lunar distance used to calc longitutude was not simple way back when
Keeping track of the point of sunset was ok, if you could keep time in different places.
- John Harrsion built a sea clock and asked to has it tested, tests it on fleet to Lisbon
- Builds a second sea clock, then 17 years building a 3rd
- Figures out how to bring it down to the size of a pocketwatch
- SMASHING success on trans-atlantic voyage
- Design was stolen by new Royal Astronomer
- King George forces them to pay the prize for longitude
- they pay most of the prize money, but dissolve the prize entirely
- John Harrison worked as a recluse
- could've worked together for better results
- both these methods now guide spacecraft
Most people only saw early flight in stunt shows, which doesn't convince them to try it in any way.
Charles Lindbergh one of the earliest Air Mail pilots
No one tried in first 5 years of prize for NYC - Paris
No one sold Lindbergh a plane, so he has an inexperienced engineer (Donald Paul) to build one: The Spirit of St. Louis.
Lindbergh super famous when he comes home.
- Donald Paul is not, is left out of celebrations
- Talks only about the plane, gives no realy credit to the engineers who built it.
Put a robot on Mars!!
We explore by physical presence, which is hard to do for the planets.
- Exploring planets through screens is hard, and leads to erroneous models of those planets
- Video games can be useful tools, they're designed to make us feel like we're elsewhere
Virtual Reality can be really, really effective.
Focus on the people you're encountering, and how you might work with new people.
@drjeffnorris
- Powerful
- bad documentation
- no error handling
- no access to source code
Cognitive bias: heuristics the brain uses to make very quick decisions, especially in a pinch.
You'll recognize a person and their emotional state far faster/easier than you'd solve 17 x 24
Uses cognitive bias to approximate results that are ok most of the time
You can't turn it off, and you can't always know when it's happening
search for/interpret info in a way the confirms preconceptions
- might get angry if preconception is challenged
- use devil's advocate if needed
- Undue liking for things merely because we're familiar with them
- Cognitive Ease
- Cognitive Strain
- something we find that we're not familiar with
- seen as a potential threat up front, requires more effort
- fewer errors, much slower
- Objective Criteria can be helpful for evaluation
- Personal Impression, only as support
- Seperate evaluation/execution
Overestimating the degree to which people think like or agree with me.
- We project our own assumptions on other people
- Stronger when we're a part of the same group
- Be explicity to mitigate the effect
- Encourage questions
- Collect opinions first
members try to minimize conflict to reach consensus w/o evaluation, to preserve harmony
- Changing or discarding own opinions to fit this
- limits critical and independant thinking
- Evolutionarily developed
-
highly cohesive
-
structural faults
- isolated
- homogenous members
-
situational context
- perceived threats
- recent failures
-
Have diverse teams in demographics and viewpoints
-
Encourage critical evaluation
-
Don't prime a discussion by stating your own opinions
-
Get outside opinions
-
appoint a devil's advocate
When we all think alike, then no one is thinking - Walter Lippman
Our comforting conviction that the world makes sense rests on a secure foundation: our almost unlimited ability to ignore our ignorance. -Daniel Kahneman
@laura_nobilis
@tehviking
Achievement is a system of control
- It's somebody else's rules
likes tools and techniques asks how focuses on tactics
fascination with people asks who what where when
likes gaps in the world asks why focus on vision and invention
What word do you do when you're stressed?
Invest in yourself and others
- You can opt of the the cycle
- Your preferences are your friends
- Build the platform you want to live with
@jmmastey
People are working so very much, foregoing time off.
When demand exceeds capacity, we make the easy choices.
Not quantity of time, but quality of energy.
- Pay Attention to energy
- Get energy
- Spend it wisely
- Recharge it better
- Get enough sleep, and restorative sleep
- exercise, at least a little
- eat well
- should have short-lived, moderate levels of stress
- willpower does not diminish, unless you believe it does
- multi-tasking is like the cake - a lie
- procrastinate less
Take some damn time off - Joe
Tom Macklin
The Reasonable Expectations of Your CTO, Software Craftsmanship 2012
Assuring is not good, but evident assurances let other people decide for themselves
The right controls in the right places with the right assurances.
- abstract OS "hooks" with FFI
- stub OS controls for development
- don't give your app system priveleges
tmacklin
@eileencodes
benchmarking: how slow it is profiling: what is slow
github.com/eileencodes/integration_performance_test
time is not a good baseline to use
gem benchmark-ips
gem ruby-prof
- lots of options
- CallStackPrinter visualizes call stack
gem StackProf
- more focused profiling
Always Be Benchmarking
Turns out OSX is awful at measuring CPU time
use wall time
RubyVM.stat
AllocationTracer tool
Know your tools, use multiple tools
speakerdeck.com/eileencodes
@darinwilson
Sonic-pi.net
https://gist.github.com/darinwilson/137764d4b06072de626c
60 == middle C, can use :C4
each integer represents a half-step note
sleep takes a number of beats to rest as an arg