Good morning, everyone.
I’ve been thinking for a while now about the role that certain institutions, such as yours, play today within the global music scene (in the field of experimental music, of course). It came naturally to me to compare IRCAM’s role today with the one it played, for example, in the 1990s, when it was a beacon of the avant-garde, particularly due to its contribution to the development of software such as Open Music. We all know the world has changed, but I wonder – now that AI has finally found fertile ground and has become a mature technology – what is the role of an institution like IRCAM, and why is it no longer able to offer not only any innovation but any contribution whatsoever within the possibilities defined by the new AI models (consider, for example, the potential of CLAUDE CODE alone)?
Thanks to anyone willing to help provide an answer
Hi,
I have yet to see any artwork of notice much less merit coming out of the current hype cycle. I also find it odd to demand of IRCAM to follow the pied piper, to whistle along the whims of venture capital and asset managers.
I see no reason for IRCAM’s approach to machine learning to change. Best to avoid resource sinks, no matter the buzz or how flashy.
All the best,
António
Thanks, Antonio—interesting thoughts.
I find it strange that people are talking about “hype” surrounding an institution that recently hosted techno concerts and Kali Malone to boost its image. And even LispWorks, back in the day, certainly wasn’t created by a mutual aid company (I recall that one of its main uses was pure financial speculation).
It simply seems to me that opportunities are being missed which—aside from SUNO (which is what people tend to think of when they talk about AI and music these days, but I’m thinking, for example, of the MPC skills for Claude, such as Topos of Music, which implements Guerino Mazzola’s theoretical musical categories)—could yield interesting results for the composer community from this institution.
I do understand the “sour grapes” mentality to some extent, but it’s striking—even from your response—that there’s a lot of politics at play and little genuine interest in creativity.
best Fabio!
Hi,
You’d be better off asking ChatGPT the meaning of hype cycle.
Mazzola already implemented in software his work himself decades ago.
Solo uando haya dedo se cerrara el anillo. — Roberto Juarroz.
All the best,
António
Hi Fabio and António,
This is a fascinating debate because it points to a real tension in experimental composition today.
Fabio, I agree that the historical importance of environments such as OpenMusic lay not simply in their technological novelty, but in the kind of access they gave composers to musical structures, representations, and processes. The relevant question, therefore, may not be whether IRCAM should adopt every new form of generative AI, but whether it can help transform emerging computational possibilities into genuinely compositional environments.
At the same time, António’s skepticism is understandable. Much of the current AI discourse is driven by commercial systems designed for rapid content generation rather than detailed musical thought. Following that model uncritically would not necessarily serve experimental composers.
There may, however, be a third path between commercial generative platforms and dependence on large institutions. The contemporary Python ecosystem makes it increasingly possible for composers and small research communities to construct hybrid environments combining signal analysis, resynthesis, machine learning, and algorithmic organization. These systems can remain offline, transparent, and comparatively deterministic, while giving composers direct access to parameters and acoustic data.
Praat AudioTools is one modest example of this direction. It connects detailed audio and voice analysis with compositional processes through a hybrid Praat–Python architecture. I would not present it as a replacement for OpenMusic or for IRCAM’s institutional role, but as evidence that experimental tool development is becoming more distributed.
Perhaps this is also where institutions such as IRCAM could remain important: not by competing with commercial AI companies in general-purpose generation, but by developing critical, technically rigorous environments in which new computational models become musically intelligible and compositionally controllable.
All the best,
Shai
Hi,
As a matter of principle, I have no qualms with artists using generative AI in their work. What I’ve expressed is what more and more outside the Bay Area are also figuring out for themselves, these models spam fields and industries much more than they extend capabilities. Buyer’s remorse is becoming more and more pervasive.
The alternative approach you mentioned has been IRCAM’s own approach to machine learning (and intelligence) for years and years. Any other approach would only add to the flood of enshittification digital content is facing, not to mention sink its and artists’ budgets into the ground.
All the best,
António