Open source software is the best path to
clinical translation
Ga¨
el Varoquaux
Sharing is caring
Slide 2
Slide 2 text
For academic research
As a research enabler
- Sharing code for reproducibility
- To enable sharing of ideas
Not a business plan
Most of it will die, that’s fine
better open source that dies
than closed source that dies
G Varoquaux 1
Slide 3
Slide 3 text
Open Innovation
Disassembling to understand
Reusing parts in surprising ways
Enabling education
G Varoquaux 2
Slide 4
Slide 4 text
Openness for trust & control
Health = Private, sensitive data
Own the technological stack
Health = High-stake decisions
Inspecting the decision-making
G Varoquaux 3
Slide 5
Slide 5 text
The lure of software profits
Software seldom makes money
by itself
Network effects
Platform business
Scale economies
Most software never makes it
G Varoquaux 4
Slide 6
Slide 6 text
Software = capture economy
Once you become
dependent on software
the vendor milks you
The old game
of vendor lock in
G Varoquaux 5
Slide 7
Slide 7 text
The unsustainability
insurance
Research seldom creates products
so let’s be open source
Not disenfranchise the
stakeholders
We are researchers,
closed source kills us long term
G Varoquaux 6
Slide 8
Slide 8 text
The long haul
Long term strategy
for open source
A strategy of commons
Sharing the common cost
- Much software is infrastructure
- It needs to be a common
Sustainable software business
dilute costs via open source
G Varoquaux 7
Slide 9
Slide 9 text
Communities
Two kind of open source
Open as disclosure
& transparency
Community-driven shared objects
Commons
Complementary benefits to the field
G Varoquaux 8
Slide 10
Slide 10 text
Open source is the game that gives us a shared benefit
as opposed to a bet on an individual win
G Varoquaux 9
Slide 11
Slide 11 text
Messy, imperfect
But can be beautiful
Organic
But requires process
open source should not mean amateur
Many failures, granted
True also of closed-source
(winners curse)
G Varoquaux 10
Slide 12
Slide 12 text
Simple is beautiful
But the demands of the real world...
- Feature creep
- Product vision
Not a strength of “design by comittee”
Software is about managing
complexity
G Varoquaux 11
Slide 13
Slide 13 text
Closed does not make safe
“Many eyes make bugs shallow”
Better a known bug than an
unknown one
Battle of narrative
More scrutiny on open
= more criticism
log4j failure:
a testimonial to open source adoption
G Varoquaux 12
Slide 14
Slide 14 text
Maintenance
To fix bug
To adapt to a changing reality
touch screens, GPUs, AI, data integration
Back to process
and the need for scrutiny
Due diligence
Easier with open
liability & competition are drivers
G Varoquaux 13
Slide 15
Slide 15 text
Supply chain
Software does not sit in vaccum
But builds on shoulder of giants
linux, python, pytorch
G Varoquaux 14
Slide 16
Slide 16 text
Open standards
Phasing out obsolete systems
Avoiding lock in
The battle against standards is ugly
Counter narrative: Go fast alone
G Varoquaux 15
Slide 17
Slide 17 text
Data
is the new oil
The era of AI
A trained model is tied legally to
the data it was trained on
The capture game
is moving to a new field
Privacy at tension with openess
G Varoquaux 16
Slide 18
Slide 18 text
People
technology is a social endeavor
Data, software
the endgame is sharing effort
and distributing control
Challenge: aligning individuals
IT manufactory
No scarcity economy
G Varoquaux 17
Slide 19
Slide 19 text
Open source is the long run
imperfect, but enables anyone to improve
G Varoquaux 18