Send us Fan Mail
The most powerful computers in the world run on open source. The
internet runs on open source. And yet most companies still think
"open source" means "cheap license fee" instead of strategic
infrastructure.
Michael Berns spent seven years as Director AI at PwC Germany,
recognised as a Leader in AI Services by Gartner and Forrester.
Then he left, to become Head of AI at the Eclipse Foundation —
the Brussels-based, vendor-neutral home for over 440 open source
projects, including the IDE most developers have used at some
point without ever thinking about who's behind it.
In the current episode of Masters of Data, BI & AI, he
debunks the biggest myths about open source AI, tells the real
story behind his move, and explains what European digital
sovereignty actually looks like once you get past the slogan.
A few things that stuck with me:
On the three biggest myths about open source AI:
It doesn't mean free — many projects come with SLAs,
quality gates, and years of paid development behind them. It
doesn't mean less safe — an open codebase is inspectable by
anyone, which is a stronger security posture than a black box,
not a weaker one. And it doesn't mean nobody's accountable — when
a project sits inside a real foundation, committers are named
individuals, and Eclipse just introduced voluntary identity
verification for contributors.
On why he actually left PwC:
No corporate spin here — becoming a father, Brexit, and
what he candidly calls "a late midlife crisis" pushed him to
rethink what he wanted his work to mean. He'd spent years
volunteering on the side for causes like the UN World Food
Programme. Eclipse Foundation offered him the chance to make that
his actual job, not a side project.
On PanEval, the story Rocky specifically wanted to hear:
Michael brought Chinese BAAI technology into the European
open source ecosystem to build PanEval — an EU AI Act-aligned,
open source AI evaluation platform. Not a press release. A
running project, and a concrete example of what building
optionality actually looks like instead of just talking about
sovereignty.
On the real data problem, even now:
Despite everything AI can do today, 60-70% of real
implementation work is still spent finding, cleaning, and
governing data. Department heads will tell you their data is
perfect — until someone actually checks. His advice for
consultants and vendors alike: do the full assessment first,
manage expectations early, and get buy-in from the people who
actually hold the data, not just the C-suite.
On where regulation could help or hurt open source AI:
The current OSI definition of open source AI is strict
enough to disqualify roughly 99% of models on the market. Michael
expects the field to move toward a graded system instead —
matching transparency requirements to the sensitivity of the use
case, healthcare needing far more disclosure than a low-stakes
internal tool.
What's your take — does true open source AI require full training
data transparency, or is a graded system the more realistic path
forward?
LinkedIn Profile
LinkedIn Profile of Michael Berns:
https://www.linkedin.com/in/aithoughtleader/
LinkedIn Profile of Rocky Khan:
https://www.linkedin.com/in/khanrocky/
Business Book
The Five Dysfunctions of a Team by Patrick Lencioni — Michael
still returns to it years later. His reasoning: the more senior
you get, the more your real problems are people problems, not
technical ones.
Kommentare (0)
Melde dich an, um einen Kommentar zu schreiben.