Skip to content
yourinductionsucks.fyi

Variables influencing conditioning-evoked hallucinations

You can find the original paper here. Paper by Benjamin R. Fry, Dominic Roberts, Katharine N. Thakkar and Alexander W. Johnson

Terms

  • unconditioned stimulus (US)
  • conditioned stimulus (CS)
  • conditioned Response (CR)
  • conditioned taste aversion (CTA)
  • representation mediated taste aversion (RMTA)

Hallucinations and associative learning:

What drives the CR?

  • Early ideas:
    • automatic activation of a motor response triggered by a particular stimulus (stimulus-response S-R representations)
    • CS evokes response via associatively-activated representations between CS and US (stimulus-stimulus, S-S representations)
  • More modern:
    • structure of learning and computational rules for linking events together
    • In modern literature, this research confirms learning mechanisms influence our perceptual experiences

An example of stimuli influencing perceptual processing:

  • smell of almonds - they smell sweet, but there are no indications of sweetness. This is through learning (through olfactory and gustatory experiences.) This is repeatable with other smells.

Both humans and animals respond to stimulus-stimulus conditioning:

  • Holland starts with CTA in animals. (Causing gastric malaise for sucrose, for example.)
  • Later - they stacked an auditory tone on top. Using this tone created a similar, but weaker aversion to the food. This was called representation-mediated taste aversion (RMTA.)
  • The CS (tone) representation over time becomes less realistic and more distinguishable from the reward.

Similar to RMTA - we can condition rich perceptual experiences in humans. However, there is more variance in humans than animals for conditioning response.

Factors influencing the strength of conditioned hallucinations

  • Similar to how animals are more easily conditioned in dire conditions (starvation,) experimenters had better luck conditioning humans when they were told that their continued employment depended on their ability to discern the color of a screen - again using a tone. They are uncertain if this shift in motivation simply changed their reports or it actually induced hallucination.
  • Supporting this - studies reporting weak or null results did not use techniques like this.

…the studies reporting more robust conditioned hallucinations required participants to make perceptual decisions about the second stimulus… These perceptual decisions during training may have led to stronger stimulus associations (e.g. by enhancing attention to the stimuli, resulting in stronger conditioned hallucinations) and invite the suggestion that motivation and attention to contingencies may mediate the strength of the controlled hallucinations.

  • Conditioned hallucination effectiveness may depend on the modalities of the hallucination making sense. (EG - sweetness of the smell of almonds, vs a sweet taste with the color of a room.)
  • Some senses may be more ‘reliable’ than others. EG - visual is more reliable than speech processing, so speech perception is modified by lip movement. In addition, hallucinations may be more effectively conditioned when S1 is more reliable than that of S2 - S1 would override S2.
  • Hypnosis appears to strengthen conditioned hallucinations. Summary: Strength may depend on motivation, attention, and ability to concentrate.

Computational approaches to understanding conditioned hallucinations

  • Prediction errors can be viewed from the equation λ-V, where λ is the maximum learning, and V is the associative strength, leaving the prediction error.
  • Dopamine transients (applying to both humans and animals) point towards an elevated amount of importance/weight to an event. If these fluctuate inappropriately, this could be a contributing factor to psychosis. Dopamine neurons (in recent work) encode facets of our experience including ‘sensory prediction errors,’ and this role may contribute to conditioned hallucinations.
  • Bayesian Inference models uses weighted priors for learning, as well as varying weights to each new input. (The new prior is influenced by a pile of previous priors.)
  • Bayesian Inference could be done through Bayesian predictive coding, where prediction error is processed up through a hierarchy of priors.
  • Hallucinations could be due to (incorrectly) precise priors. There could be more weight on the hallucinated or incorrect prior than the sensory input.
  • Weak sensory input could also create prediction errors - where we keep our (incorrect) priors, like Charles Bonnet Syndrome - where people lose their vision and experience hallucinations.
  • Circular inference - where we look at (in our case weak) sensory input repeatedly, then attribute it to an incorrect prior, can increase confidence in hallucinations, especially if those priors are pushed ‘up’ the system.
  • So, in conclusion of this section…
    • Hallucinations could be caused by strong priors.
    • Hallucinations could be caused by imprecise sensory input.

Neurobiology of impaired reality testing and conditioned hallucinations

  • Rats with neonatal ventral hippocampal lesions (NHVL, also brain damage) had stronger RMTA than the control group - meaning they wanted less to do with the non-poisoned pellets than the healthy rats.
  • In rats, prolonged THC use made rats more vulnerable to RMTA.
  • Similar to how they did RMTA (by playing a tone alongside a stimulus,) researchers fed the mice sugar, and then still with the tone, swapped it out for water. The schizophrenic mice took longer to uh - realize they’d been duped. Ketamine exposed mice also experienced similar effects.
  • In a PET study - a conditioned auditory signal could activate the “occipital cortex at similar levels as that of a visual stimulus.”