Let me tell you about a Microsoft AI researcher […] who recently spent quite a lot of time considering whether the legendary Microsoft real time strategy game Age of Empires II is conscious, and built a basic neural network within the video game using digital goats to prove his point. […]
De Wynter built an LLM within AoEII using goats. “The point of the paper is to formally show that we anthropomorphise too readily, and that sometimes the claims we make with regards to LLM capabilities are too strong,” he told 404 Media. “It’s not an easy task, given that ‘human-like attributes’ is a bit of an abstract term.”
From a theoretical standpoint, there are three options:
Keep in mind that all Turing machines can be computed using a pen and paper (although the paper must be very big). Or goats in age of empires
You dare to doubt the sentience of the AI paper?
Sorry to have to be that person but for clarity of the discussion the word you want is sapience.
Sentience is something that, one might argue with a bit of effort, machines have already achieved: they can use sensors/senses to gather information about their environment which one might call sensing things. They can even react to that information that their senses have gathered.
Sapience is the ability to think about that information beyond the immediate impulse. It’s necessary for what we usually refer to as “intelligence”.
While I take your meaning I’m not sure this is a helpful path forward.
We’re not fully sure what conciousness is or how it happens. We could split hairs about sentience, sapience, emergent properties, etc, and maybe even do so meaningfully but our semantics, today, will remain necessarily continental. Even if we progress our understanding beyond that point and discover real answers to those hard questions… Will we be able to define the boundaries between states? Will we be able to even sat such boundaries exist?
Addressing your argument directly: can sentience exist apart from perception of environment? I’d say it must. If so then how is sense and reaction an indicator such a system carries within it a theory of mind? It can’t be. It’s not enough.
The cart is firmly in front of the horse in all discussion and debate of LLMs as concious, sentient, or sapient. Partly because such claims in the affirmative are not yet testable (nevermind falsifiable) claims and, more to the point of the discussion at hand, partly because we do not know what it is we are talking about.
Anyone claiming otherwise is guilty either of magical thinking or of peddling snake oil and, in either case, has misjudged the importance of first answering more fundamental lines of inquiry.
Wait, really? Can you give several examples of this, then? Sentience involves an organized, (hopefully at least sometimes) self-aware response to one’s environment. For example, the chance for something to become sentient is never increased by a lack of stimuli, right?
Sorry, I don’t follow. I’m talking about the definition of those words.
Not talking about consciousness.
Sentience IS perception of environment.
I don’t know and I’m not saying it is.
Do we not? That’s precisely why it’s important to make that difference. Sentience is relatively easy to define and can be assigned with a lot of confidence whereas sapience isn’t and can’t.
I’d conject that our brain could be computed on paper, given a large enough sheet. We arent that special, we are biological computers. We decide things based on a huge amount of factors but they could be 100% deterministic Do I want another Coffee? Brain checks tiredness, stomach fullness, time to get to next task, body temp, outside temp, flavour from last cup, and so on… Given 1000s of sensors it may be that our consciousness would “freely” choose the same option if all parameters were exact same.