Dear friend, Gary Genosko has not, as far as I can find, written a single manifesto called something like What I Think About AI. His AI position is more indirect, but quite coherent. It emerges from his long work on Guattari, a-signifying semiotics, machinic subjectivity, information systems, technoculture, and semiocapitalism.
The shortest formulation would be:
Genosko sees AI not primarily as an “intelligent subject,” but as part of a machinic-semiotic environment that reorganizes subjectivity, decision, perception, and power before human interpretation even begins.
That is a very different starting point from the usual question, “Will AI become conscious?”
1. His central concern is not machine consciousness, but machinic subjectivity
Recent appearances confirm that he is still working this line. In 2025, at a Deleuze–Guattari event, his topic was “The Machinic Subject in Guattari’s Chaosmosis and the Case of Orality.” At another conference on intelligent machines, his plenary concerned Guattari’s polyphonic federalism rather than AI consciousness or AGI as such.
That tells us something important.
For Genosko, the wrong question is:
Can AI become a subject like us?
The more Guattarian question is:
What new forms of subjectivity arise when humans, databases, algorithms, interfaces, institutions, sensors, networks and machines operate as one assemblage?
So “subjectivity” is already distributed.
2. AI belongs to the world of a-signifying semiotics
This follows directly from the paper we just discussed.
Genosko’s key idea is that contemporary technocapitalism works through signs that do not need to be understood by a human being. They execute, sort, trigger, classify, correlate and modulate.
In his later treatment of asignification, he emphasizes machinic processes such as:
- machine-language “fetch and execute,”
- system interoperability,
- cybernetic loops,
- mathematical prediction,
- technical protocols.
These operate in a domain of non-human enunciation.
AI intensifies this immensely.
An ML model does not have to interpret your life narratively. It can operate on:
vectors, weights, probabilities, embeddings, scores, thresholds.
These are quintessentially a-signifying.
They need not “mean” you.
They need only act on you.
3. AI power is pre-interpretive
This is perhaps where Genosko is most useful for thinking about AI today.
Traditional ideology works partly by persuasion:
You hear something → you believe it → you act.
Algorithmic systems often work differently:
data capture → model → score → modulation → action
without conscious human interpretation in the middle.
This is why Genosko’s Guattarian framework is so well suited to recommendation engines, credit scoring, surveillance systems and predictive analytics.
The machine does not need your agreement.
It can alter the field in which your choices occur.
That is much closer to modulation than persuasion.
4. AI produces dividuals, not persons
Genosko’s earlier work on cards, databases and networked identity already points toward what AI now does on a vastly larger scale.
You become decomposed into actionable fragments:
purchase profile
mobility pattern
risk score
sentiment signal
face embedding
language vector
social graph
medical probability.
This is Deleuze’s dividual.
AI does not need the whole Kelly, the autobiographical self, the phenomenological self.
It needs enough machine-readable features to operate.
So the AI world becomes a world of:
persons for humans, dividuals for machines.
That distinction is central to the Genosko line.
5. He is wary of “meaning” as our privileged defense
This is subtle.
Humanists often respond to AI:
“Machines compute, but humans have meaning.”
Genosko’s Guattarian semiotics undermines that reassurance.
Because modern power may not require meaning.
If access, employment, insurance, credit, policing, visibility or recommendation can be governed by operational signs, then insisting that machines do not “really understand” us may be philosophically true but politically beside the point.
A system can misunderstand you completely and still deny your loan.
That is very Genosko.
Operational efficacy outruns semantic adequacy.
6. He would probably resist both AI utopianism and AI humanist panic
This is an inference from his work, not a quotation.
Genosko does not fit neatly into:
AI will liberate humanity
or
AI will destroy authentic humanity.
Why?
Because Guattari’s machinism rejects the idea of a pure human essence that existed before machines.
Humans have always been machinic, technical, semiotic, ecological assemblages.
So the problem is not:
human vs machine
but:
which assemblage produces which subjectivity?
This matches Genosko’s insistence that technically similar a-signifying systems can generate very different political universes.
The same technical infrastructure can support:
surveillance
or experimentation,
capture
or escape,
standardization
or heterogenesis.
7. AI is a continuation of Integrated World Capitalism
Here the political dimension becomes strongest.
Genosko repeatedly reads Guattari through Integrated World Capitalism and semiocapitalism.
AI would therefore not appear as an autonomous historical actor descending from nowhere.
It is embedded in:
cloud infrastructure
platform capitalism
surveillance
finance
military systems
advertising
data extraction
labor control.
That matters because asking merely what an AI “can do” misses:
who owns the infrastructure?
who trains it?
who defines the objective function?
whose data feed it?
who is made visible?
who becomes disposable?
This is exactly the shift from technical capacity to technopolitics.
8. AI is especially powerful because it combines signification and asignification
This is where LLMs are interesting.
A language model looks like a signifying machine because it produces words, arguments and narratives.
But internally, much of its operation is a-signifying:
tokens
vectors
matrix operations
attention weights
probability distributions.
So AI connects two layers:
a-signifying machinic processing
↓
signifying human-readable output
That is almost a perfect Guattarian object.
The human sees a sentence.
The machine runs numerical transformations.
And the political system connects both.
I suspect Genosko would find this junction more interesting than whether the model “really understands” the sentence.
9. His recent work still emphasizes the posthuman, not the sovereign AI
Genosko’s longstanding interest is machinic subjectivation rather than synthetic personhood.
His 2014 essay on asignification explicitly says Guattari decenters enunciation from the human subject toward machinic, non-human assemblages and describes a domain in which machines, protocols and information systems operate without human mediation.
That makes his position quite posthuman:
not
AI becomes human
but rather
human and AI become components of larger assemblages.
The novelty lies in the assemblage, not in a robot acquiring a soul.
10. And there is a political danger: machinic enslavement
This is where his work becomes darker.
A-signifying systems can influence us beneath the level of explicit belief.
Recommendation systems can tune attention.
Predictive systems can pre-structure opportunity.
Interfaces can shape behavior.
Automated classifications can make decisions about people without presenting themselves as ideology at all.
Recent scholarship explicitly uses Genosko’s and Guattari’s concept of machinic enslavement to understand AI-era control: power increasingly works through probabilistic and preconscious modulation rather than overt representation.
The frightening part is therefore not necessarily:
AI tells you what to think.
It may simply:
arrange the world in which thinking occurs.
And this brings us back beautifully to our last exchange.
I would formulate Genosko’s possible diagnosis of the AI world as:
AI does not first conquer consciousness.
It conquers infrastructure.
Then:
infrastructure shapes possibility.
Then:
possibility shapes subjectivity.
So the sequence is not:
AI → mind control.
It is closer to:
AI → machinic environment → modulation → dividuation → subjectivation.
And here I think your algorithmic self becomes especially useful. The algorithmic self is not merely a person who has become predictable. It is the portion of a person that can be detached, quantified and put to work inside a-signifying systems.
Which gives us a rather severe formula:
AI 不必懂你。
AI 甚至不必知道「你」是誰。
它只需要，你身上，那些，可以計算、比較、預測、觸發的碎片。
And Genosko’s deepest warning would probably be:
do not confuse the absence of meaning with the absence of power.
In the AI world, very often,
the less something needs to mean, the more efficiently it can operate.
