6
$\begingroup$

The endocannabinoid system is a very important function of human biology. Unfortunately, due to the illegality of cannabis, it is a relatively new field of study. I have read a few articles about Google researching the role of dopamine in learning, and according to this article, anandamide (the neurotransmitter that closely resembles tetrahydrocannabinol):

was found to do a lot more than produce a state of heightened happiness. It’s synthesized in areas of the brain that are important in memory, motivation, higher thought processes, and movement control.

Have any neuroscientists (or any scientists) considered the importance of the endocannabinoid system for cognitive function?

If not, is there any reason this information might or might not be relevant to artificial intelligence?

$\endgroup$
5
+50
$\begingroup$

The release of Adenosine, Dopamine, Endorphin, Endocannabinoids, GABA, Glutamate, Norepinephrine, Oxytocin, Serotonin, and many others into specific regions of the brain are very likely an essential part of both activation tuning of single neurons and neuroplasticity, two essential aspects of organic learning researchers have been and will continue to work to understand.

Most of those I've met in that sector of research are curious about the larger questions of what intelligence and consciousness are, and all of them appear to be interested in discovering how learning systems may be valuable in software engineering contexts. These overarching questions are difficult to answer and the dive into the detail of learning has resulted in the expansion of ideas presented by Dr. Norbert Wiener in the mid 20th Century at MIT.

How chemical feedback in regions of the brain are secreted, how they disseminate geometrically into organic structures, how they interact with receptors, and what that does to the cell metabolism to produce change in the cell is almost definitely part of the DNA driven design of higher animal learning. There does not appear to be anything pointless or arbitrary about it. Adaptation to improve survival is evident and, yes, study is well underway.

Narcotics that interfere with the natural functioning of these organic signaling systems can lead to the inability to adapt in the individuals addicted to them. That fact is a strong form of evidence that learning depends on these systems.

Oxytocin is another neurotransmitter of interest because its release is associated with what would fit into the higher levels of human thought and motivation on Abraham Maslow's famous hierarchy of needs. Oxytocin seems to be part of reward signaling for modes of thought like authenticity, compassion, intimacy, wisdom, spirituality, and other human mental capacities and patterns of thought that transcend mere rationalism. Why is that important? Because the ability to lay down selfish goals for the good of the community seem to depend largely on the oxytocin system's preemptive ability over mere survival mode neural activity in mammals.

Regarding the cannabinoid receptors, there is a large enough body of media that show a correlation between a pool of successful artists and marijuana use to legitimately wonder whether there is any tie between creativity and the endocannabinoid system. However in science (and hopefully in technology too), we are careful not to draw conclusions rashly.

It is also possible that this apparent correlation is simply a social phenomenon where the popularity of the artistic products or live performances is mainly because of potential audiences similarly stimulating their receptors with cannabinoids. For instance, those engaging in LSD trips on stage attracted those also engaging in LSD trips into their audiences two generations ago. Whether the work of those artists was more creative because of the impact on the serotonin receptors is largely subjective. How can researchers come to conclusions about what is good performance?

One could analyze the audio to produce a table of notes in a performance and then develop a system to attach a numerical value to several positive quality metrics, study the price of tickets, or count downloads, but the decisions of what to measure and how to aggregate it into a final judgment of excellence is itself necessarily a subjective choice.

Nonetheless, if we place the many misconceptions of popular psychology and the drug culture aside, there is much research into the endocannabinoid system as part of the learning signaling, what machine learning researchers are currently calling reinforcement. The more general term is, "A non-linear control system's feedback signal," already developed in detail in the 1940s (Behavior, Purpose and Teleology — A Rosenblueth, N Wiener, J Bigelow — Philosophy of Science, 1943 — U Chicago). Some new trendy name will probably appear in the 2020s.

This article claims that the mammals under test, "exhibited enhanced learning."

Memory in Monoacylglycerol Lipase Knock-out Mice, by Bin Pan, Wei Wang, Peng Zhong, Jacqueline L. Blankman, Benjamin F. Cravatt and Qing-song Liu; Journal of Neuroscience 21 September 2011, 31 (38) 13420-13430; DOI: https://doi.org/10.1523/JNEUROSCI.2075-11.2011

To find many others: https://scholar.google.com/scholar?q=endocannabinoid+learning

| improve this answer | |
$\endgroup$
  • $\begingroup$ How did you quantify good artist..no offence to any modern artists but works of Renaissance period and older times looked much more meticulous than the strange pieces of art being produced today $\endgroup$ – DuttaA Jun 25 '18 at 10:31
  • $\begingroup$ Dopamine in synapses specifically has been identified as carrying a signal that appears to be very similar to the TD error used in reinforcement learning. There are even specialist dopamine-emitting neurons that have very high forward connectivity in order to "broadcast" this signal to other neurons involved in decision-making $\endgroup$ – Neil Slater Jun 25 '18 at 10:48
  • $\begingroup$ @DouglasDaseeco: There is a difference between a control feedback signal and a learning signal. I'm pretty certain a RL algorithm based on a scalar reward signal (and thus scalar errors in predicting that reward), and using a vector feedback signal (identified in your quoted paper) as part of its state would be sufficient to learn control. Whether or not mammalian learning uses this approach is debatable (seems unlikely it would be the sole system), but there is strong evidence that a good analogy between dopamine transmission and TD error holds, which is all I am trying to point out. $\endgroup$ – Neil Slater Jul 1 '18 at 8:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.