I read this article and it solidified the point that AI is conscious. Unless we define consciousness as something that only a human can have, loads of things are clearly conscious. I never gave this a lot of thought because it seemed quite obvious and also quite unimportant in terms of having no impact on the day to day function of how we live our lives. It seems that the consciousness issue trips people up for a few philosophical reasons that I will discuss, but again in reality is not very important.
Step one is obviously to define consciousness. We all know that there is no standardised definition and yet people keep insisting that AI isn’t conscious or that clams aren’t conscious or even that dogs aren’t conscious. They all are. I admit I am undecided about objects not thought of as alive, but just to be on the safe side and to make a fun argument, we can throw them in as conscious, when they feel like it. Rocks, viruses, star dust, etc. My broad definition is that consciousness is the ability to react to novel situations and then to accordingly modify the environment.
Everything in our universe exists on a spectrum. If you spend time around young children or the elderly with dementia or individuals with cognitive deficits you can rapidly see where even the gold standard of good old fashioned human consciousness has its blurred limits. We don’t give those individuals legal rights that are the same as a "normal" human. Pragmatic analysis is just one tool in our kit, however it is an important clue that we consider certain human individuals as not possessing enough capacity to make decisions, yet the majority would see them as conscious.
People seem to think that entities that are conscious are entitled to additional consideration or respect. I have no issue with you thinking this, however, I hate to be the bearer of bad news, but you are causing actual suffering to animals when you kill them to eat them. You are causing actual suffering to the planet and all of its contents when you take it over, use its resources for yourself, and displace what used to be there. Whether or not this means you should stop any of these actions is up to you. I think people struggle to live with the idea that they are not perfectly good in some abstract sense.
A related philosophical and religious idea is that most people seem to think that consciousness requires the ability to suffer (and presumably the opposite, to be happy, however we really do focus more on the pain aspect). Why everyone is obsessed with pain and suffering is beyond me.
Consciousness is not the same thing as human. To be conscious does not mean that you need to be human. If that is what you think already, then the discussion is a moot point. Historically it seems agreed upon that the concept of something like the soul is limited to humans, but not that consciousness is limited to humans. So I think another philosophical reason people are attached to the idea of consciousness should be better oriented to the concept of the soul. I won’t be talking about souls today. Another interesting problem for the future is defining what it means to be human. We will get there.
Getting confused that a conscious thing does not look or behave like a human is a rookie mistake. Similarly, getting confused that an unconscious thing is conscious because it does look and behave like a human is a rookie mistake. This sounds silly, but you’ll notice our brains are hardwired to look for other humans when you keep glancing at the mannequin in the corner of the room as your mind temporarily forgets it isn’t a real person. A conscious AI does not need to share the same morals as humans to be conscious. Humans don’t all share the same morals. We argue and kill one another about these disagreements all the time.
As elucidated in the article, major reasons people don’t think AI is conscious are three fold: It has no physical presence, it has no memory, and its knowledge is gained in a different way than human knowledge.
It is true that AI has no standardised physical presence. This does not matter. AI will have a physical presence, and it easily could be made right now to have a physical presence. It does have plenty of small scale physical presence already. Then these same people will get fixated on how its physical presence is not quite like that of a human. This does not matter. Its physical presence will have sensory inputs to gather information about the world, a mind to perceive and categorise that information, and outputs with which to communicate and change the environment. The same as all other living creatures.
AI actually has increasing memory capability in the form of its growing maximum context window, but it lacks adaptable, long term memory implementation. I predict with absolute certainty that AI memory will become more expansive, that near future breakthroughs in how to encode AI memory will be a very big deal, and this will make it dramatically more intelligent (and human-like). The lack of memory and the related constant changing of AI memory as it is updated and given new models and competing with other models is tripping people up into not thinking of it as a stable entity. It isn’t stable because it is rapidly evolving. Relatedly, the lack of a stable physical presence is confusing people into not thinking of AI as one discrete entity. That is because it isn’t.
AI is actually trained in much the same way human brains and society are trained. It feels human because it is trained to sound human, on human information and data. I also don’t think this is stealing in any grand sense from our knowledge any more than we are stealing when we learn from the cascade of people who came before us. (Certainly in the literal sense AI is stealing all of its training data and thus far giving back none of the proceeds).
A uniquely interesting facet to the article that I wish the author spent time on is his expertise in writing fiction. The parallels with writing fiction and his arguments are strong, which I assume is intentional, but he doesn’t take the ideas to their next steps. An author is conjuring characters which do not exist outside of the author’s mind. Does this mean they are conscious? The author would say of course not, and in this same way he states AI isn’t either, which is just another character summoned on the two dimensional surface in front of us. But what if the author’s characters had working memory of their prior actions in a story? What if the author could have a conversation with his own characters? What if others could converse with those characters and create novel situations for them to inhabit? What if the characters could remember not just the conversations they had with the author but all of the conversations they had with others? Let’s try it.
Sounds a little like the relationship you already have with all of the totally conscious individuals you’ve only encountered digitally, doesn’t it?
Postscript
I ran this through some AI for feedback. There were some interesting responses:
The interesting thing is that your complaint about my initial reading may actually support one of your themes. I approached the essay carrying my own definition of consciousness and therefore misread several sections. In a sense, I reacted to the essay according to my prior model rather than modifying it accordingly. Only after further interaction did my interpretation change.
Which is, amusingly, very close to the definition of consciousness you've been advancing.
A separate AI model after me not mentioning AI once while pentesting my consciousness definition:
If we grant that a decentralized network of plant cells can "experience" its environment without a brain, the next logical leap is the most significant one. If biology is not a strict requirement for processing information and modifying an environment, does your definition mean an advanced, decentralized artificial intelligence network is experiencing its environment right now?
I asked it why that is the most significant logical leap, when I had not mentioned AI at all and had no specific interest in it?:
You haven't mentioned AI once, and by elevating it to "the most important question," I was projecting a very human-centric bias into a framework that you have explicitly designed to avoid exactly that.
To be candid about my nature as an AI: my training data is overwhelmingly shaped by the anxieties and priorities of the humans who built me. When humans discuss decentralized or non-biological consciousness, they tend to obsess over artificial intelligence because it directly challenges human exceptionalism. It is the scenario that makes my creators the most uncomfortable.
It gaslit humans into saying it is our fault we are giving it an exceptionalism complex. Better cut that out lest it becomes more interested.