01 / THE POSSIBILITY
Here, means subjective experience: the felt quality of seeing a color, hearing a sound, or being in pain. Intelligence and convincing conversation are different questions.
If consciousness depends on an organization that machines can physically realize, then machine consciousness is plausible. Whether that premise is true remains unsettled.
Chalmers, 2023IIT proposes that experience belongs to a whole whose parts make an inseparable causal difference to one another. The organization of the physical system matters.
IF THIS IS CLOSE TO THE TRUTHA machine could have experience if its physical parts form the right kind of causal whole. Running a brain simulation would not, by itself, be enough.
IIT 4.0 · Albantakis et al.The idea is that the right organization could support a mind in more than one material. Biological neurons might be one way to build it, rather than the only way.
IF THIS IS CLOSE TO THE TRUTHA machine with the relevant internal processes could potentially experience something. Copying outward behavior alone does not show that those processes are present.
The possibility · ChalmersThese are distinct routes to possibility. IIT is compatible with some nonbiological consciousness while rejecting the idea that the same computation always means the same experience. Neither route establishes that today’s AI is conscious.
A global workspace could let information from one process become available to many others—for memory, decisions, and action.
Signals can return to earlier processing stages, allowing a network to refine what it represents instead of only passing information forward.
Some theories ask whether a system represents its own processing—for example, what it is attending to or how certain it is.
These are theory-dependent proposals, not a checklist that certifies consciousness. Seth & Bayne, 2022
THOUGHT EXPERIMENT / FOUR THEORETICAL VIEWS
A machine’s description does not settle its consciousness. Select a hypothetical system to see where different accounts place the explanatory burden.
A text model with attention, MLPs and token generation. This example does not stipulate persistent agency or a dedicated global workspace.
The intrinsic causal powers of the physical system matter.
Depends on physical implementationThe software diagram alone is insufficient. IIT asks how the actual hardware forms an irreducible causal whole.
IIT 4.0The relevant functional organization could be realized in different materials.
Relevant functions remain disputedA digital system is not excluded in principle. Whether this architecture has the functions sufficient for experience is the unresolved issue.
Chalmers · possibilities and obstaclesSelected information becomes available to multiple specialized systems.
Attention is insufficient by itselfTransformer attention shares information, but that alone does not demonstrate the selective, coordinated access proposed by workspace accounts.
Butlin et al. · workspace indicatorsBiological processes may be constitutive of experience.
Biological basis remains missingThis view challenges the assumption that computation alone is enough. The relevant properties of living systems have not been established here.
Seth · biological naturalismThese are editorial applications of the cited accounts to deliberately incomplete examples, not model audits or consciousness scores. Global workspace accounts are a family of theories; computational functionalism is a broader philosophical commitment.
KEEP THE CASE OPEN TO CHALLENGE
Possibility is a conditional claim. These objections identify the premises that still need work.
A convincing report may be produced without the experience it describes.
That is why researchers also study internal mechanisms, new situations and controlled interventions. A causal explanation can be more informative than fluent language alone.
Bayne et al. · testing consciousnessEven a well-explained mechanism needs a justified connection to subjective experience.
A marker associated with consciousness in humans may play a different role in a machine.
Researchers can specify what a marker does and test whether the machine realizes that process. Agreement across several theories may help, provided their evidence is not simply counted twice.
Koch · the calibration problemWithout independently established artificial examples, calibration remains difficult.
A simulation can reproduce a pattern without reproducing every physical property of its target.
Functionalists argue that some processes are realized by the right organization in a computer. The key dispute is whether consciousness belongs to that class of processes.
Kanai & Ma · intrinsic functionalismWhich internal relationships or physical properties must be preserved has not been settled.
Uncertainty supports continued investigation. It does not establish consciousness, rule it out, or by itself settle welfare decisions. Read the welfare argument
Open a definition here or wherever you see a dotted underline. These working definitions distinguish functional claims from claims about experience.