CF Debate

Chinese Room Argument

IntentionalitySubstrateEntity predictions

Overview

A sufficiently large look-up table could replace any interaction with the target algorithm, while remaining input/output identical. However, it is a priori implausible that a look-up table, with its simple mechanics, could be conscious, no matter how vast it is.

The canonical example addresses 'understanding' rather than 'consciousness': A non-Chinese speaker inside a closed room uses a look-up table to identify fluent, convincing responses to Chinese sentences provided by an external Chinese speaker. The external person might think that the person inside 'understands' Chinese, but they do not. A common response is that understanding exists not in the look-up table or the operator's mind but somehow in the 'room as a system', e.g. the look-up table + mechanical sensor/operator. However, this response does not help the 'consciousness' critique, because all those elements remain as simple as the look-up table. Note that the original framing can be read as an argument against the Turing Test as a behavioural test of consciousness rather than an argument directly against CF, hence the change in emphasis in this presentation.

Responses

  1. CF could be restricted to require a particular causal structure for the algorithm, rather than allowing any variants of the algorithm that merely have an identical input/output mapping.

  2. Deny the intuition that large look-up tables would be 'too simple' to be conscious.

    BUT: Need to motivate a threshold (even if gradual in nature) by which small non-conscious look-up tables would transition into large conscious look-up tables.

  3. To accommodate all possible (infinite) language responses, the 'look-up table' must in fact be a different system, likely applying compression, generation, extrapolation, and other functions. Such advanced functions might plausibly compound into generating consciousness even if a look-up table would not.

    BUT: This additional function needs motivating, because relatively simple combination rules may already be sufficient alongside a very large table to generate adequate linguistic complexity for any realistic conversation.

  4. The computation did in fact happen and did generate conscious experiences, but only when the look-up table was created (i.e. a significantly more complex act than reading off the look-up table).

    BUT: That means that there was only one conscious experience, contradicting the intuition that conscious experiences should be taking place during the conversations in question. Now we can just re-use the look-up table as often as we like but without generating any new moments of experience.

Further reading