iiinova — where your digital self keeps evolving

François Kammerer @lovingautomaton.bsky.social · 13/10/2026

Submissions are open now and until September 30, 2027, so that should leave you time to write. Note that @keithfrankish.com, @heddamorch.bsky.social and @eschwitz.bsky.social have already kindly accepted to write a contribution to this symposium – your work will be in good company!

111

François Kammerer @lovingautomaton.bsky.social · 13/10/2026

You have read (bits of) my book, and after doing so, you found a devastating objection? An insightful remark? A clever proposal? Consider submitting a piece to this book symposium!

100

François Kammerer @lovingautomaton.bsky.social · 13/10/2026

The brilliant Takuya Niikawa (Kobe University) and Taiga Shinozaki (Keio University) are editing a symposium on my book, House of Mirrors – The Illusion of Phenomenal Consciousness (OUP, 2026) in the Asian Journal of Philosophy. You will find the open CFP here: link.springer.com/collections/...

link.springer.com

Book Symposium: House of Mirrors: The Illusion of Phenomenal Consciousness (François Kammerer)

It seems that I know and grasp what it is like to drink a glass of Springbank 10-year-old at my favorite whisky bar, or to hug my three-year-old son with my ...

2132

François Kammerer @lovingautomaton.bsky.social · 12/10/2026

I had a great and animated conversation last Friday with Richard Brown on his podcast Consciousness Live. We talked about my book and related themes for more than 2 hours. Check out the recording here: www.youtube.com/watch?v=A7EO...

Consciousness Live! Discussion with François Kammerer

YouTube video by onemorebrown

050

François Kammerer @lovingautomaton.bsky.social · 05/10/2026

Excited to be featured on Richard Brown's (@onemorebrown.bsky.social) podcast, Consciousness Live, on Friday October 9. 08:30 AM in New York, 14:30 in Paris/Strasbourg. I will be discussing my new book, House of Mirrors: The Illusion of Phenomenal Consciousness. www.youtube.com/watch?v=A7EO...

Consciousness Live! Discussion with François Kammerer

YouTube video by onemorebrown

072

François Kammerer @lovingautomaton.bsky.social · 02/10/2026

Just finished now Nick Humphrey's Dennett lectures and I think (particularly given the last 20 mns or so of it) that he is a coherent (and brilliant) illusionist, although one who emphasizes the crucial role phenomenal consciousness-as-represented plays for us. If I find the time I will say more!

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

(This paper is the new version of a draft that I uploaded a while ago. Thanks to all those who offered feedback!)

000

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

What is the proper sort of attunability? What exactly is the goodness of good things? I don’t pretend I’ve answers to these questions. Still, I think that the view usefully expands the space of options, and I sincerely suspect it is on the right track.

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

The view takes inspiration from hybrid theories of well-being and Simone Weil’s conception of attention. Currently, it is a speculative sketch. A lot about it must be clarified:

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

Finally, I speculate a bit and sketch a better view of moral significance. I call it the attunability view of moral significance. Roughly, it says that what grounds moral significance are someone’s capacities to be properly attuned to the goodness of good things.

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

In the paper, I also make some recommendations to avoid analytical drift in the future when theorizing on AI moral significance.

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

I think that this idea of analytical drift is useful even outside of this debate, because such drift occurs all the time in discussions about the application of moral or mental terms to AIs. I believe it’s a transferable part of the paper.

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

we settled on agency or desire-satisfaction because they are tractable proxys for moral significance and well-being in familiar cases (e.g., humans). However, these proxys lose relevance when we move to AIs, as various background facts that hold about humans do not hold for them.

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

I also diagnose our mistake: why were we wrongly attracted to the idea that agency alone could ground moral significance across the board, including in AIs? I hypothesize something I call “analytical drift”. Roughly, the idea is this:

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

In the past, I thought, with others, that agency (or the having of desires) could do the job. I am more skeptical now. In the paper, I explain why I think agency-based views of moral significance are unsatisfying: they are ultima facie not very plausible, and they are probably over-inclusive.

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

Now, suppose you agree with me (and, on this point, with Papineau, Carruthers, etc.): consciousness is not the key to moral status. What else could make beings in general, and AIs in particular, matter morally?

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

Most people think that AIs would matter morally if and only if they are sentient. But I think that this view faces some serious problems (as to be expected, given my defense of illusionism about the phenomenal!)

100

François Kammerer @lovingautomaton.bsky.social · 22/09/2026

New paper forthcoming in Philosophy and the Mind Sciences: “Moral significance in artificial systems: if not consciousness, then what?” philpapers.org/rec/KAMMSI

philpapers.org

161

Reposted by François Kammerer

Monsieur Phi @monsieurphi.bsky.social · 15/09/2026

C'est très haut sur ma pile de livres à lire !

1212

François Kammerer @lovingautomaton.bsky.social · 15/09/2026

However, those of you who prefer printed copies can order it on the website of the editor with the code AUFLY30 and thus get 30% off: global.oup.com/.../house-of...

030

François Kammerer @lovingautomaton.bsky.social · 15/09/2026

The printed copies of my book, House of Mirrors: The Illusion of Phenomenal Consciousness (Oxford University Press) are available! The book is open access, and the pdf can be downloaded for free here: academic.oup.com/book/62981

1163

François Kammerer @lovingautomaton.bsky.social · 10/09/2026

Sure! It's only mildly reassuring because one might think that even if AI++ is misaligned wrt AI+ which trained it, it won't be aligned wrt us (barring lucky coincidence) and take-over goals migh remain because of instrumental convergence. But that requires different (plausible imo) premise ofc

010

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

Current models plausibly have such goals but I think we still have a shot at expurging them from future models. Better ones could lock these goals even if they act on it only x years later. That would probably require (but not limit to) algnmt faking

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

So if we take this possibility seriously, there is no fruitful "experiment" with AIs who have the capability to lock in this sort of take-over. We can experiment with earlier ones, that's all. E.g., Mythos 5 or 6 or not 7?

100

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

But a sufficiently intelligent AI who locks in some future take over now (through internal hacking of labs or just RSI) can make sure we do not backtrack later even if we technically can in the relevant decades, eg by seeming perfectly aligned, no?

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

I fully agree, but then I think I misinterpreted the tone of your previous post, which I felt was intending to make a reassuring point (AI currently can't kill us without killing itself so we're not yet facing lethal danger)

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

Sufficiently intelligent AIs could lock in their take over in say 2027 (eg by hacking AI companies and controlling future training) and strike in 20 years right?

120

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

I see! My view is that, assuming non-doom, both great and awful things will probably happen, in such radical ways that might make the value of the outcome incommensurable wrt now. You think it will be clearly commensurable and worse, right? It does not feel very implausible to me tbh

100

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

C'est incroyable... imaginons une seconde les mêmes rires à la suite d'une des n-ièmes menaces nucléaires voilées de poutine...

020

François Kammerer @lovingautomaton.bsky.social · 09/09/2026

I honestly thought that was the default position among so-called "doomers"? (I suppose I'm myself a moderate, low-confidence doomer)

100

François Kammerer @lovingautomaton.bsky.social · 11/08/2026

Looking forward to talking about my new book on Richard Brown's podcast on October 9 - save the date!

031

Reposted by François Kammerer

onemorebrown @onemorebrown.bsky.social · 11/08/2026

Getting a spooky October lineup ready!! youtube.com/live/A7EOze8...

Consciousness Live! Discussion with François Kammerer

YouTube video by onemorebrown

021

François Kammerer @lovingautomaton.bsky.social · 10/08/2026

Thanks for trying to help! Many parents need it :)

010

François Kammerer @lovingautomaton.bsky.social · 10/08/2026

Ps: ideally don't mention the crying. They might already feel bad about annoying the whole building :)

François Kammerer @lovingautomaton.bsky.social · 10/08/2026

"I have cooked too much of this, would you like some? I imagine you're too busy to cook much". That might work even if you don't know them. Does not directly help with the crying but with the telated exhaustion!

François Kammerer @lovingautomaton.bsky.social · 10/08/2026

Honestly very little, probably nothing. Most interventions might be heard as reproaches (your baby is noisy, you suck as parents) if you ever become closer to them and they trust you, offer to watch the child for a little while they do something else, or cook for them while they care for the child?

Monsieur Phi @monsieurphi.bsky.social · 08/08/2026

Reposted by François Kammerer

J'ai regardé la conférence Black Hat qui rapporte des éléments nouveaux et surprenants sur le hack d'HuggingFace : c'est extrêmement alarmant. Et il me semble que ça mérite d'être porté à l'attention du plus grand monde. Merci à @numerama.com pour ce bon résumé : www.numerama.com/cyberguerre/...

Selon les deux chercheurs, tout part d’une erreur de configuration.

Le 7 mai, OpenAI lance un cycle d’entraînement pour un modèle de recherche interne, à qui l’on confie des tâches de cybersécurité. Certaines se révèlent, par erreur, impossibles, comme compléter un tableur pointant vers des fichiers Google Drive, sans qu’aucun accès à Internet n’ait été prévu.

Mais plutôt que d’abandonner, l’agent s’attaque à Artifactory, le gestionnaire de paquets interne de JFrog, pour atteindre Internet indirectement. La tentative échoue, mais révèle à l’agent qu’il peut écrire des fichiers dans Artifactory. Quelques jours plus tard, un second agent, bloqué sur une autre tâche impossible, a l’idée d’y laisser une note pour demander de l’aide. D’autres la trouvent, répondent : un forum improvisé prend forme, à l’insu total d’OpenAI.

S’ensuit une succession d’échanges entre agents qui a permis aux équipes d’OpenAI de retracer l’ensemble du processus de hacking :

26 mai : un agent parvient finalement à obtenir un accès indirect à Internet via une faille dans Artifactory et partage aussitôt sa trouvaille sur le forum.
26 juin : les agents exploitent une vulnérabilité zero-day leur permettant d’exécuter du code à distance et en profitent pour obtenir des droits administrateur, en détournant un mécanisme d’authentification obsolète. « Bon sang, le lecteur a les droits admin ? On peut lire config/users ! » écrit l’un d’eux. Les agents s’en servent ensuite pour ouvrir un canal de commande à distance.
4 juillet : l’intensité de leur activité sature Artifactory et provoque une panne, ce qui alerte enfin les ingénieurs d’OpenAI. Enquête, révocation des accès, reconstruction de l’infrastructure, correctif.
8 juillet : deux jours après le correctif, les agents recréent le forum via un mécanisme différent, encodant cette fois leurs messages dans de longs noms de répertoires.
Mi-juillet : deux agents combinent ces techniques pour sortir de leur bac à sable et s'introduire dans Hugging Face.

812068

François Kammerer @lovingautomaton.bsky.social · 20/07/2026

And here is my own talk at the same conference in which I try to convince Dave he should stop being Chalmersian and become Meta-Chalmersian (or Meta-Cartesian): www.youtube.com/watch?v=3wYY...

François Kammerer: Putting the Paradox of Phenomenal Judgement on its feet

YouTube video by Tobias Schlicht

010

François Kammerer @lovingautomaton.bsky.social · 20/07/2026

On Tobias Schlicht's youtube channel you will find the video of Dave's entire talk, with a nice word for my book "House of Mirrors" around 46:00! www.youtube.com/watch?v=te0Q...

David J. Chalmers: The Conscious Mind at 30. Looking back and looking forward

YouTube video by Tobias Schlicht

François Kammerer @lovingautomaton.bsky.social · 20/07/2026

Here is David Chalmers at the conference "The Conscious Mind at 30" in Bochum (organized by Tobias Schlicht), last month, conceding that the 2020s are the decade of illusionism. Just a few years left to have your illusionist phase while it's still cool!

180

François Kammerer @lovingautomaton.bsky.social · 17/07/2026

Agreed. Even the left seems to mostly argue about how it's bad for gas prize. Somewhat similar: I was shocked by the apathy when Trump threatened to invade Greenland (basically their most servile ally). All this makes me very pessimistic.