Hacker News Reader

Comment by cwmoore | Hacker News Reader

It seems to me that AI is more plant than animal, and the specific biological response mechanisms that plants use have explanatory power toward understanding and communicating about AI in a more productive way than the popular conception of GAI as sentient.

View on HN

As our testing procedures improve, we find out more and more about the cognitive abilities of animals, and have since found self-awareness in chimpanzees, elephants, dolphins, magpies, and recently, through sniff tests, dogs. I find it not far fetched that at some point, we start to recognize (some) plants as sentinent beings, maybe on a different timescale than us. Would be interesting to know why you think they are so different from animals that they are better models for AI?

Primarily because their mechanisms of action are easier to study in vivo. Sentience is a red herring.

I addressed this in https://news.ycombinator.com/item?id=18000854

Dogs are not self aware, otherwise I would love a source. That just sounds like some feelgood popsci.

Web sources seem to agree with justtopost:

https://www.psychologytoday.com/us/blog/the-human-beast/2017...

Dogs fail the mirror test.

They are certainly quite intelligent, but technically don't meet the criteria for self-awareness.

And the scent thing where they spend less time on their own scent, or don't mark over their own scent, is dissmissed as a reflexive action.

I'm referring to the "Sniff test of self-recognition" (STSR), which has been proposed by Gatti in [Gat16] and run on a very small sample of dogs. Horowitz has proposed another testing method, derived from STSR, in [Hor17]. [Hor17] has been criticized by Gallup et al. in [Gal18] for lacking evidence according to their framework of self-awareness (dogs need to sniff on themselves after exposure to their own scent). But the authors agree that the "olfactory mirror" test is a valid extension of the "mirror test", a self-awareness test based on visual clues. Gatti has shared some of the criticism recently [Gat18], but attributes the lack of evidence to formal problems, namely [Hor17] failing to cite important aspects of his original work in [Gat16]. And indeed, while [Gal18] cites [Hor17], and [Hor17] cites [Gat16], [Gal18] makes no mention of results by Gatti in [Gat16], which would support their framework: "[...] when released together inside the enclosure and left free to move and interact with each other and with the five samples, the four dogs repetitively sniffed the excretory organs of the others and the containers, sometimes stopping to sniff themselves."

I'm no expert, so I don't want to make any strong assertions here. But it seems that only recently, alternative self-awareness testing procedures (to the "mirror test") have become acceptable by researchers. Now new tests will be designed, and there is a body of evidence to be gathered for each of those tests. I personally believe that, given that the defense of human exclusivity for cognitive features has been a running fight, self-awareness will eventually suffer a similar fate.

[Gat16] Cazzolla Gatti, Roberto. Self-consciousness: beyond the looking-glass and what dogs found there. Ethology Ecology & Evolution, 2016

[Hor17] Alexandra Horowitz. Smelling themselves: Dogs investigate their own odours longer when modified in an "olfactory mirror" test. Behavioural Processes, 2017

[Gal18] Gordon G. Gallup Jr. & James R. Anderson. The "olfactory mirror" and other recent attempts to demonstrate self-recognition in non-primate species. Behavioural Processes, 2018

[Gat18] https://robertocazzollagatti.com/2018/06/07/self-awareness-i... - 2018

Interesting... Care to elaborate?

I am not prepared to properly address this at the moment, but at root it has to do with the great complexity of plant's adaptation without the advantages of locomotion. Certainly interested in your thoughts and others'.

Really? Considering AI has (or will soon have) the best "brain" when it comes to adapting its default heuristics based on new information, plants don't seem to fit the right explanatory model.

I would like to know in what way you think they are a better model than animals.

Similarly to the fallacies of anthropomorphism, the amygdala-hijacking that often occurs when ascribing sentience undermines the effectiveness of the animal model for AI. The central-nervous system ('best "brain"') analogy fallaciously assumes advanced executive function in a natural (ie. not rule-based) environment that confuses the role of currently tenable AI applications.

Robotics and agent-based systems may provide some counterarguments. My original comment was more about explanatory reasoning for machine learning systems than about development paradigms or goals.

Reply on news.ycombinator.com