Introduction

I’d like to share information that is worth sharing. So I ask the question—what is valuable information?

With the ubiquity of search engines and Wikipedia, much factual information can be found very quickly. However, in this era it is easy, even encouraged, to create “fake” information, including text, audio, video, and any other type of media that can be consumed by the use of computational devices. It is increasingly important that we can routinely discern whether the information we are seeing or writing is of any value.

To this end, I put forward a specific theory. It may be wrong or incomplete, but I find it is usually worthwhile to specifically test my ideas, by trying something that I at least do not know is wrong.

Valuable information could be any of:

1. true,
2. valuable, regardless of its truth,
3. both true and valuable,
4. useful, regardless of value and truth,
5. useful and true,
6. useful, true and valuable,
7. valuable, regardless of both truth and usefulness.

We will start with as a background prop the paper Objects of Consciousness, which says that evolution tunes perceptions toward fitness, rather than truth. Fitness being better expressed as a complete absence of unfitness, ie., smoothness of niche, round-peg round-hole and square-peg square-hole. Fitness is not universal, but a relation of an object or person and its environment together.

We have good reason to believe that the process of evolution does indeed solely tune perception for fitness, because it does so with everything else. Then for our purposes, if you want fitting and valuable and true and useful perceptions, there is only one possibility:

You must apply the Principle of Multiple Explanations, ex Epicurus, c. 300 BC, and strictly consider all and only the hypotheses not yet known to be incompatible with evidence, discarding with abandon those that do not, and then you may apply importance sampling to choose the order in which you consider their truth, value, or usefulness (or any other property or combination of properties). The point is that you need to start from a point which is at least not known to be unfit.

The maths on that tells us that we still must eventually consider all of the hypotheses that might be true in some form, so if we do not wish to do an infinite amount of work, we should choose the order in which we consider hypotheses carefully.

This way we can gradually improve our perception of truth by chipping away untruths, one by one, and reconsidering whether we should check something else first, before we proceed with expanding our previous pet theory.

What do I mean by “valuable?”

When we describe information as “valuable,” what are we really saying?

Whether a piece of information is valuable is relative to the person observing it. The same piece of information could be utterly irrelevant to one person, but absolutely vital to another.

So “value” isn’t just a property of information, but a property of information and people together; what we are saying when we say information is valuable is that it helps at least one person.

If we count the number of people for which this is true, and can quantify the positive impact that information has on those people, we can then quantify the value of information. As we will see later in the piece, the way in which information helps people is that it helps them to both be more present in reality and to better create it.

As a first draft of such a quantification, we could count the number of experiences people have with the information compared to without it, and count the number of meaningful choices people make with the information compared to without it, both over a specific period of time.

As a person does not control the experiences they have except by their meaningful choices, then we can decompose this into two quantities: lifespan, the length of time a person is alive during the period, and the difference in the number of meaningful choices made during the period with the information compared to without it.

If we make the simplifying assumption that the person is alive for the whole period (which is true for most periods of time we’d consider for a person), then this equation takes the form of a physical action:

So the value of information for a person is its ability to motivate their physical action compared to its absence.

As physical systems, humans need input flows of energy and mass. What we see from the above reasoning is that, even if they have those inputs, they also need a strictly positive quantity of veridical information, information about the real world, in order to be able to act in any meaningful way, even if they are physically able to.

We could consider integrating also over all people that the information could reach. Then we can calculate an objective value of information: by how much does it improve the audience’s capacity, or inclination, for physical action, compared to its absence.

In this way, by providing people with veridical information, you enable them to do things that were otherwise impossible.

In principle, in today’s world we could find out most things ourselves through original research, if we had the time and resources to do so.

But in practice, we don’t have time for everything. Use of language modelling to summarise information for you does not help at all: it can increase the quantity of information you receive, but it cannot improve its veridicality.

There is another objectively correct metric here, which is that the information that a language model (or any tool) presents to you must strictly be a sufficient statistic for all the underlying information it summarised to show you. If it does not, it is hiding information you could have received and so diluting the potential value you receive, or misrepresenting information that would have been relevant to you; that is, would change your actions if it were not misrepresented.

Do current language models do this? No — they guess, mangle, and confabulate. This is fine, for art, in humans, which is a method of communication: as it’s said in musical theatre:

“When emotions are too strong for words, sing. When emotions are too strong for song, dance.”

We reach for art when words fail us.

Language models are not alive things, they do not need to communicate with us as a thing in themselves. So it is strictly incorrect for them to behave other than in the way described.

We can see this progression of achieving the previously impossible most clearly through the progression of science, or scientific knowledge. Newton wrote:

If I have seen further it is by standing on the shoulders of giants.

How did he see further?

He could see further because he could get to “the same place” his predecessors were at more quickly, because they published their results and made the information available in such a way that he did not have to invest as much time and effort to come to the same understanding as they had. They could trust that it was, at least reasonably, veridical.

In todays world, we have forgotten this principle.

A sickness pervading the infosphere, writing in 2026, is that of inveridicality. It’s not just that most writing is “bullshit,” unconcerned with truth. What is written often does not even resemble reality.

When people absorb inveridical information, they are gradually fitting their perceptions to another world, a world not our own. They are losing their ability to take meaningful action in the real world, and gaining nothing, because it is not possible to physically act in a world other than the real.

In the past, words had meaning.

Go and read, if you haven’t, old writers. Adam Smith. Henry George. My current book, Kandinsky’s Point and line to plane, published 1926, freely available on archive.org. Passages like this, an ordinary paragraph from the preface:

Since photography and motion pictures today record all events, situations, or persons for practical or sentimental need, the skill of modern man has been freed from reproduction by hand, thus enabling him to cultivate a higher stage in art expression by following his creative esthetic urge. His eyes have become sensitized to realize the rhythmic life in the span of the in-between--the life that is the essence of a non-objective masterpiece. Such a masterpiece, due to those spiritual qualities, becomes everlastingly appealing in its endless combinations of colours, forms, and contrasts, in their relations to each other or to space. It can be easily observed that each colour and design motive is organized in itself, while constantly reacting and playing with its form or colour opponent. Thus it brings restful enjoyment, which is as peacefully uplifting as the observance of the infinity of the starlit sky. Out of such pleasures emerges the realization of the rhythm which lies in the in-between, realized by following the motives and discovering the meeting points of lines and forms, in contrast to a calm, harmonious unit.

Could a language model write this? Could a human, in 2026? Compared to most writing of today, I find this transcendently beautiful! What have we lost?

True, valuable information is, quite literally, healing. It helps us live and act more in the real world, while its opposite is, metaphorically, soul-destroying, and literally, this false image evaporates our ability to live, sense, be, and create, as we lose sight of what is important in reality.

In practice, then, how do we know when we have valuable information?

The answer is very simple: we know we have valuable information if it has been checked against reality, and that which did not match to it discarded, or represented another way, if still relevant to our actions.