You see what you need to see
The landscape of conscious awareness is defined by your motives
This is the twenty-fifth post in a series on AGI. You can read the previous post here. You can also see a list of all posts here.
“Inner experience” enters our consciousness only after it has found a language the individual understands — i.e., a translation of a condition into conditions familiar to him — Nietzsche, Will to Power
We all think we can clearly and directly identify what our subjective experiences are, such as the colour of a tree that we see, or the sound of a washing machine. Some philosophers have even argued that your experiences are the only things you can truly know. On the other hand, experience shows that if you tried to express the precise content of your experiences to others, you would find that language always comes short.
Consider how you might identify the colours you see. English words for colours like “blue” are socially determined groupings. They can never mirror the nuance of your actual experience of colour. So there is no way to prove to others or even yourself that you could identify what specific colour you saw. Nevertheless this is likely a shortcoming of language. Your eyes still see similar colours similarly. All things being equal, two examples of a shade of blue will cause the same thoughts. So there must be a consistent identity to your experience of a colour beyond merely the English word “blue”, right?
The above examples suggest that there are two distinct parts to visual identification: (1) the automatic response which takes place in the retina and visual cortex, and (2) the subsequent act of identification. An event may happen in your mind, such as when your eyes see a colour, but that doesn’t mean you will be consciously aware of it. Identification, specifically conscious identification, requires an extra step, a focused and purposeful one.
Anytime you try to identify what a colour is in itself, you are by definition paying attention, you are bringing something into focus. You are attempting to accomplish something, as indicated by the use of the word “try”. This implies you have a goal, and you are making a judgment in deference to that goal. The goal may be as simple as “how can I express what is there in English?” or even “what colour do I see?”
The result of this search must be a useful interpretation, i.e. useful to that goal. When you assign the word “blue” to a picture of something blue, it’s because you were trying to express its colour linguistically, and were not, for instance, asking if the image was visually pleasing. This highlights a difference between noticing and identifying. When you notice something, you select the subset of sensory inputs that are relevant to your goal, such as an apple or painting, and record those inputs (somehow). When you identify something on the other hand, you reinterpret the inputs as something you know, such as “a Van Gogh painting” or some other useful identifier.
The fact that ‘identification’ impels the mind towards what is useful to its goal means you are not identifying the colour as it is “in itself”, you are creating an interpretation that satisfies the motivating goal. It warps the outcome to its own needs, and will warp it differently at a different time, under a different need.
This is why you sometimes see the cube in the image below as blue…
…and sometimes, more specifically, as cyan.
There is no “neutral” standpoint from which you can see things as they actually are, in absolute terms. So-called “neutral” interpretations are just driven by another type of motivation, one based on the tension of being judged, and where the goal is to be perceived by others as being objective. You can sense such a pressure at work as you evaluate the colour of the cube above. When by yourself, and not particularly focused, you would likely identify it as simply blue, as you match it to the closest basic colour (red, green, blue, etc)¹. But if you were asked by a scrutinizing peer “what colour is the block?” you will sense a new pressure to be precise, along with the anxiety of being pedantically corrected. You now inspect the colour more clearly, and call it “cyan” or “cyan-yellow”. Both the word and the conscious experience itself will change. Your mind seems to focus more, and somehow notices more of the colour.
That last sentence is worth reconsidering. How do you know that you saw more of the colour the second time? While you are looking at the block you are simply looking, and evaluating its colour. You would only realize that you were “seeing more clearly” in hindsight, through introspection. You glanced back and became aware of a difference in quality between the two. Chances are you could still recollect that difference for a while after. This means a memory related to the colour was formed each time. You became aware of something, and becoming aware always means you can recall something after the fact, even if only for a few moments. That is what made it “conscious”.
Since the memory was different for the two cases, it was not a memory of the cube itself, but of some reinterpretation of it. When you looked at the cube in the context of each driving motive, you connected it to a thought that acted a useful interpretation, e.g. a word, or an image, or a colour. Going forward the new thought/memory will act as your interpretation of the image. The next time you see the cube, all those thoughts will come back, elicited by the cause. Moreover, when you reflect on your memory of the colour, or of the cube, you will become aware of this elicited interpretation, which means you will form a new memory (aka interpretation) of the prior one. And so on.
As is becoming clear, identification, interpretation, thinking, and memory are all interconnected, and their definitions overlap to a large degree. Interpretation is one aspect of identification. Generally, we use the word “identification” to indicate that you are connecting more than one input to the same interpretive thought; e.g. many objects can usefully be interpreted as “cyan”.
This leads naturally into “concepts”. The unity of concepts like cyan comes about when you interpret various experiences of cyan objects as a common symbol of some kind — e.g. the word “cyan”. A “concept” can be defined as a common symbolic interpretation applied to many specific experiences. So even though you believe you are discovering concepts in the space of your thoughts, you are actually creating concepts; they were not there to begin with. You find them because you are looking for them. Your desire to systematize your thoughts projects a consistency onto them, whenever doing so is useful to you.
The act of identification is obviously quite complex. I’ve left out a few pieces to make this post concise — for example, I didn’t mention what the source inputs are that an interpretation gets attached to, and how the underlying goal connects the interpretation to the source. These will be covered in a later post. The bottom line is that identification is the act of learning an interpretation which achieves a goal.
Despite this, when you think of a colour, it intuitively feels like you’re seeing some thing in your mind, a static entity with an identity that you are simply perceiving, like looking at an object in front of your eyes. This seemingly simple act is in truth an ever-renewing flow of ad hoc thought-memories and interpretations. To see this, take a moment and think of something blue. How long can you maintain this same image in your mind? You’ll notice that you can’t focus on it for more than a second before it disappears, and you must re-think the image to bring it back. When you identify something in your mind you are creating a constantly cycling series of memory-interpretations which takes effort to uphold. Just like in a dream, you cannot maintain focus on one object for long before it dissolves, since the object in question is an invention of your mind, whose concreteness is created on you seeing it.
Objectivity and free will
When we imagine the act of consciously identifying something we tend to imagine we are coming at the task from an objective standpoint. That is, you assume that you can choose to identify anything, at any time, as long as you know how to do it. Researchers carry this assumption with them into the lab. Identification is seen as an impartial act of free will, as arbitrary as deciding what direction you will turn your eyes. This is part of the appeal of the objective theories of object recognition; it gives the impression of full control, as well as impartiality. It creates the illusion that the object itself is dictating the outcome of the identification, and would drive it the same way for everyone, at every time. You rarely ask what happened right before you chose to identify the object. The act seems to come ex nihilo (out of nothing).
It didn’t — there was always a reason. For example, if you were to turn your attention to identifying the things around you right now, it would not be a free choice, it would be because this post suggested it, and you were perhaps wondering if you could prove the post wrong. The motive to disprove this argument would play a role in how you interpreted the experiences around you; the reason for which you engaged in the act determines what you find. If you felt that you could in fact identify the things around you of your own free will, then that thought, as a solution to the challenge of the post would find a welcome home in your beliefs going forward.
When training AI models, there is an assumption that the reason for coming to the data has no bearing on what is learned. Thus the data can be presented in any order. The training regimen for a Natural Language Processing model randomly presents text samples as immediate inputs to the AI. But in reality what you pay attention to and perceive in the data depends on what came before the data was consumed.
Humans pick up different aspects of a text, and read it in different ways depending on why they came to it. You may focus on the font, or skim for keywords, or try to extract a useful process from a DIY text, or look for entertaining ideas, or factual information, etc. If you suspect the writer was angry, you look for clues to that in the text. You may even project onto the text things that perhaps weren’t in the author’s mind at the time of writing. You are not extracting truth from the text, you are creating an interpretation for yourself. The act of identification always creates its own truth.
Empirical theories of identification tend to be based on “push” approaches; the statistical patterns of sensory experiences automatically push themselves on your mind and coalesce into interpretations and generalizations. In our inverted approach here, the mind is pulling together stimuli and building the most useful thoughts-responses, both linguistic and otherwise. It helps if you think of thoughts as being like actions; just as you may learn an action policy that is most useful to achieving a goal, you learn a “thought policy” that collects interpretations as they prove useful, and triggers them later.
In fact the mind cannot understand anything unless it finds it useful to its needs. Without the driving motives, no thoughts or interpretations would ever arise, and your mind and consciousness would both be a blank. The flavour and variety of all your thoughts is the flavour and variety of your driving motives. Together they create the dynamic landscape of the mind, as each motive searches out its resolving thoughts, often in competition with each other. Your stream of consciousness itself is a water-wheel of new thought interpretations. Every conscious thought you have, be it of your experiences, your feelings, their qualia, your emotions, even of your thoughts themselves, must be an act of awareness, an act of identification and interpretation, and thus an act of motivated learning.
In the next post we’ll continue building this exposition of the stream of consciousness, and discuss how the process of reasoning contributes to the flow of thoughts. We’ll then be able to address the gaps left by this post.
Next post: [Reasoning]
¹ This is why people often form false memories of their experiences — their interpretation retrofit the experience into a familiar category, to the point of being misleading.
