It seems to me our current experience of the subjective consciousness of a human is currently based on the one of the following:
- reporting of the subjective experience through language
- observation of electrical activity in brain areas mapped to subjective experience

Using those 2 as the recognition of ongoing subjective experience and with materialists like Dennett defining the processing that occurs in the brain as the subjective experience itself, how long before AI can claim they have machines 'experiencing subjective reality'?
After all, we have machines that can input,store and process the objective reality. It can always be claimed that the monitored processing of information in the form of electrical activity is the evidence of the machine experiencing a subjective reality since, besides language, that is the only evidence we have for it in the case of other humans due to the problem of Other Minds.

The only argument I can think to this is to show if machines can hallucinate. There is strong evidence to show that the majority of humans when exposed to sensory deprivation will experience some form of visual or auditory hallucination. Essentially, in the absence of sensory data, the mind will generate false data. If the machine AI does this then maybe they are truly experiencing subjective reality.....

  • You need a stronger sense of what 'hallucinate' means. To some degree (braindecoder.com/post/…), our entire perceptual mechanism is a hallucination that is checked by reality inputs. No inputs, means no checking, and a more bizarre hallucination. More input means more checks and a more realistic one. But there is no discontinuous point where something is 'just unrealistic enough' to be a hallucination. – user9166 Nov 21 '16 at 2:47
  • I essentially mean audio/visual experience in the objective world that does not exist there. In the case of the machine, I am willing to reduce it to observation of electrical output when there is no input where there is electrical output when there IS input. But maybe this question should be closed as some users have suggested. Hypothesized on unconstructed structures although I thought that was philopsophy was about. – Anoop Alex Nov 21 '16 at 3:12
  • But the combination of green and red to make yellow is a visual experience of the objective world that does not exist there. There is no radiation of the wavelength of yellow created by combining the two other colors. If that is a hallucination, then our notion of color is an ongoing hallucination, and since it characterizes almost all of our visual experience, we live in a continual hallucination. – user9166 Nov 22 '16 at 0:35
  • @jobermark There is a wavelength of yellow which is can be created by combining wavelengths of other colours, it's just that we don't have cones in our retina to pick it up, we only have long, medium and short wavelength receptors anything that falls in between is made up from the signals from these. In this case we are "deducing" a colour that actually does exist (other devices can distinguish it), we just didn't sense it properly. You're absolutely right, of course, about so much of what we sense being some form of hallucination, but this appears to be more of a counter-example. – Isaacson Nov 22 '16 at 8:00
  • I don't mean to be deliberately pedantic, I just think you made a very good point, and I didn't want it to be detracted from by a slightly flawed example. The way we put seamless images together from what is actually a constantly changing set of pictures every time we move our eye, or the way we invent the colours in our peripheral vision (which has no colour receptors) would be genuine hallucinations of something which might not actually exist in the real world. – Isaacson Nov 22 '16 at 10:30

Here is an article about AI image hallucinations:

Yes, androids do dream of electric sheep (The Guardian, 18 June 2015)

If a program can see hallucinations it could also attempt to interact with them. With some coaxing I imagine it could maintain a hallucinatory narrative.

Google frees its dream robots: half eaten doughnut. Illustration: Google

enter image description here

  • Interesting. So it seems my suspicions of some AIers claiming their machines could dream was not entirely unfounded. From the image you posted and those in the article, fractals seem to be the theme of recognition. Like it recognized one element and then looks for that element at different scales in the image. Seems to be a caterpillar above and them it emphasizes the curve of that at various scales and rotations. But you may note my earlier example of hallucination was output in the absence of input and we are definitely getting input here. – Anoop Alex Nov 24 '16 at 7:07

Hallucination in the sense of seeing things that are not actually there takes place all the time, we "see" colours by deduction from frequencies near to those colours, we also see an uninterrupted image of the world around us despite it being constructed from hundreds of constantly changing images. Broadly our brain's activity in response to visual stimuli can be divided into three types, those which use our limited sensory equipment to deduce something about the world which is real (detectable by other equipment) but which we have not actually seen, those which help to create a useful picture of the world which most others would agree with but which might not actually be there (optical illusions etc) and those where information from our memories or imagination get mis-identified as coming from our eyes.

An AI will by necessity suffer from the first type of hallucination unless their input systems are so advanced as to accurately receive all signals in the entire range of wavelengths from all sources. The second type will also be a necessity as the machine would be unlikely to function efficiently without some degree of inference where it's sensors are lacking.

Only in the third type do I think it is relevant exactly how the machine is manufactured

Hallucination in response to sensory deprivation is not an indicator of a "subjective experience or reality" it's just the way our brains are wired. All our sensory experiences, whether real, remembered or imagined are processed using the same part of the brain. Normally a second part of the brain provides us with information about the source of the sense, in the absence of evidence (sensory isolation) or in cases of Schizophrenia, it cannot do this and we mis-identify internally generated sensory experiences as external ones, see here.

So if an AI was wired to process sensory data in the same way (i.e, sometimes from its external sensors, sometimes from its memory banks, or its pattern-generating circuits, but with a subsequent program to tell it which) in the absence of that subsequent program, it may (if it were programmed to do so) decide that all sources of sensory data are probably external and so "hallucinate". It would not be difficult to make a machine generate the same data from its memory as it is generating from it sensory input devices and treat them as identically sourced.

The trouble with AI/conciousness type discussions is that we are speculating on a technology that does not exist and so that technology could theoretically do anything. We do not need a physicalist understanding of reality to speculate about AI conciousness because even if we accept a dualist perspective, why would we then dismiss the idea that technology could advance to the stage where it could be programmed to access the "realm of ideas"?

If we're going to accept the idea of a technology being able to do something it currently can't, what rational reason could there be to the deliberately constrain the ability of that technology and then claim to have discovered something interesting by doing so?

  • Research in AI has been for the 'simulation' of consciousness. Humans can only experience their own consciousness. They can try to understand how other humans understand their consciousness through the watered down medium of language. Currently the theories are: its all material, it all immaterial and its a mix of both. Possible currently AI research is trying to realize the reality of physical consciousness. – Anoop Alex Nov 20 '16 at 9:00
  • @AnoopAlex What makes you think that humans can only understand the concious experience of other humans through language? Does not neuroscience and physiological experimentation provide us with insights also? Either way it is still nothing more than speculation, to say the current theories are "its all material, it all immaterial and its a mix of both" is a very generous interpretation. To me, that just sounds like a list of all the possibilities. If we didn't know whether a box contains sand or water it would not be a useful insight to say it might be sand or water or both. – Isaacson Nov 20 '16 at 9:16
  • Until we knew definitively whether it was sand or water, saying it might be sand or water or both WOULD be the best explanation. And if our entire world was made of sand and we were only aware of the concept of water we might make the mistake of only looking at methods that would prove it was sand and ignoring ones that would prove it was water. – Anoop Alex Nov 20 '16 at 12:25
  • Sorry, I think my metaphor might not have been very well expressed, imagine the same scenario in a world, as you say, made up entirely of sand and water, then to say it is either sand, water or both would be entirely correct, but not useful. We are not in danger of ignoring one option simply because no-one has said it is a possibility, so I don't see the utility of the statement. – Isaacson Nov 20 '16 at 13:59
  • You seem to be confusing the neural representation of sensation with the experience of sensation. As Morten Overgaard points out, neuroscience relies on subjective accounts to gain insight into conscious experience: "To arrive at this association [with consciousness], one would have to conduct several experiments correlating recurrent processes with consciousness—using introspecting experimental participants. Consequently, this method would not be independent of introspection but would carry the strengths, weaknesses, and limitations of introspection." – user3017 Nov 20 '16 at 15:54

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.