Can an intellect judge itself?

Question

Suppose you want to measure the length of something, e.g. the length of a kitchen table. It's not an easy task, but luckily you have a meter stick on hand! So you take your meter stick and make your measurement: comes up 0.8 meters, wonderful. But suppose that now you wanted to measure the length of the meter itself, this spells trouble: before you were able to measure the length of your table only because you trusted your meter to measure, well, one meter! But now, since by desiring to measure your meter you have implicitly put in doubt the length of it, your meter became useless! You cannot use it to measure itself because you cannot put in doubt the length of your meter if you wish to measure with it. You must trust it to be accurate. And if you ignored all this and just tried to measure your meter with your meter you would end up with a measurement of, well, one meter! Apparently you got your answer but we all know that in reality you have learned absolutely nothing about the true length of your meter.

This concept appears to have general applicability. You cannot use something to judge itself.

But does this really also apply on the human intellect? We can surely conceptualise human reason as a sophisticated analytical tool, that is capable to measure and judge almost anything that it encounters, a general analytical instrument, a general intelligence. But what about the act of judging our own intelligence? I don't mean of course the act of judging our ideas or our opinions, no, I mean the act of judging our own reasoning abilities. Does it have some sense to judge ourselves as smart or dumb? Does it have some sense to formulate any judgement about our intellect at all? We can surely approximate a judgement of intelligence for other humans, distinct from us, but what about formulating this judgement for ourselves?

And if the answer is no, if the answer is that we cannot judge ourselves in this regard, then does this spell trouble for the future of AI research? One of the conceptual cornerstones of the theoretical AGI field is that once we are able to craft a general intelligence smart enough to takle the problem of AGI then it will improve itself, leading to an exponential improvement of it. But if our "meter problem" has general applicability then it should be really problematic, if not impossible, for an AGI to judge and improve itself.

---

Edit: I saw a lot of answers talking about a human intellect judging other human intellects, e.g. a psychologist performing a IQ test onto a subject. This is not what my question refers to! To make myself as clear as possible: the problem I wish to takle is an intellect judging precisely itself, not an intellect judging other intellects similar to it.

Answer 1

Reading this question, things popped into my mind from memory. I hope you don't mind me sharing them.

My friend, very drunk, looked at me drowsily from the other side of the table and said: “I can never understand myself. To be able to understand itself, my brain has to be smarter than my brain.” I thought it was time to go, especially since I had not been drinking, but on the way downstairs I had the impression that, however drunk, he might be on to something.

Sometimes my own brain surprises itself. It comes up with song titles that make me laugh as if someone else had made them up. Similar to the principle of a deja-vu, apparently the “song title generating” part of my brain just spits it out, and the “is this funny” part of my brain is caught by surprise. Some of my dreams I find truly entertaining and surprising too. Not every part of my intellect knows all things at the same moment. Time may be a factor in the solution. What if I invent a test, forget about it, and find it three years later?

I had to recognize chords for my entrance exam to conservatory. I had a piano, but playing my own examples didn’t work: I could see what I was playing, and knew the answer before hearing it. I had an old Sinclair computer and a bit of knowledge of Basic, so I wrote a program that beeped triads in all inversions and varieties, asking me which one it was. It also kept track of my right and wrong answers, and told me which chords I made the most mistakes with. It even left out the ones I always recognized, after a while. I learned all the triads over summer holiday and passed the exam, without help from fellow musicians. Was my IQ higher after this experience (that was challenging for me and taught me a lot)? Our brain is a self-learning entity, so measurements might be outdated as soon as they have been made.

With help of a machine, I was able to improve my skill of recognizing sounds, because I knew how to make them even though I had trouble recognizing them. Maybe a brain can write a program that measures its own intellect too? For a certain intellectual task (answering) we use different brain parts than for inventing questions. Is the brain capable of asking questions that it cannot answer itself? Can it see its own boundaries? Do we ever use all our brain parts at the same time? Can it be done like this: the brain measures itself bit by bit, part by part. The test is made by collaborating brain parts, together bigger than the part that is measured. Was my friend wrong after all?

In any case, if it is possible, we would need some kind of detour that fools the brain into not knowing the answers right away because it wrote the questions. Possible, or impossible? Don’t ask me, I am only the bass player.

Answer 2

Would anyone ever think to measure a specific meter stick using that very same stick? This would be wrong with respect even to the ur-stick in Paris, which is defined to be 1 meter, not measured as such. At any rate, since no one (or only the 'insane') would ever think of measuring a measurement stick using that stick, but many people would think of using their own intellect to judge itself, apparently no one else has ever 'noticed' that the impossibility of the one implies the impossibility of the other.

For example, we also (usually) have the capacity of eyesight. And we can tell the limits of sight: we can see that we do not see farther than this or that thing in this or that context. We have no 'measure' of this limit that is not in sight; so are we not using sight to 'judge' itself as to its own limits?

So likewise, I don't see that reason could not judge of itself that it is limited in certainly general ways, or limited 'in the moment' (as by judging myself to be a stupid person on account of something I did last night, say).

That being said, yet on your behalf, I would recommend looking into Kant's discourse on "duties to oneself" in an initial segment of the Doctrine of Virtue. Since for Kant autonomy is the source of duty, then it follows that we place ourselves under our obligations. Yet this seems to imply that we could release ourselves from our obligations 'at will,' so how is there really the mandate of duty in play? What Kant says is that our moral selves are sort of split into two dimensions, and we judge the moral self in one dimension from the vantage of the moral self in the other. Maybe, then, we could judge our intellect in one context, based on our intellect in another context.

Unfortunately, I assume I know less than you do about AI (I used to know what AGI meant but it's evaporated from my repertoire), or at any rate I don't know enough to critique (or commend) your remarks about AI. For example, I don't know whether, "AI that improves itself," is tantamount to, "AI that is using its own intellect to judge its intellect." But I will say this: I was surprised to find out that deontic logic, of all things, is nontrivially relevant to some paradigms/progams/w/e in AI analysis/research. Maybe at the end of the day we might program our computers using something like a Kantian two-dimensional personality system. (OTOH, maybe deontic logic is entangled with AI issues on account of analysts/researchers wanting to make sure we code our potential robot overlords with good intentions, even so that they never became our overlords at all perhaps.)

Answer 3

Does it have some sense to judge ourselves as smart or dumb?

An intellect that can compare any two other minds can always also compare itself to other minds

There is no single absolute scale about smartness. But in general, you talk about comparative judgements, and for that, a mind only needs to compare itself to other minds. So as humans we can compare ourselves to dogs, dolphins and chimpanzees in our ability to use tools. We can compare ourselves to computers in our ability to solve math problems or recognize emotions. And we can compare ourselves to other humans in our ability to answer questions about history, science fiction novels, science or religion. .

Answer 4

Two tennis players can fail a ball in the same way. One can understand how he failed. The other can't. Its a question of body awareness. At top level body awareness is more important than physical fitness. The one that understands how he failed can continue to improve.

An engineer needs to produce an almost perfect sphere. As long as he has tools that can measure the imperfections of a sphere, he can improve it.

One of my favorite quotes is due to Flannery O'Connor: I write to discover what I know. The moment we write our ideas we have a chance to objectify our thinking and judge it.

Louis Althousser wrote a very nice little book: Philosophy and the Spontaneous Philosophy of the Scientists. What he is saying is that when we dedicate ourselves to some activity we develop a series of beliefs that are unexamined. If that activity is important to us we have a lot to gain in taking some time to write those ideas and examined them. This applies to my activity as mathematician and to the way I educate my children or talk with my wife.

Everybody as some capacity of doing this. What are my limitations? I have many blind spots. Some of them are quite clear to the people that know me well. One of the better ways to deal with them is to cultivate truthful relationships. For Socrates philosophy had everything to do with the art of living well. I believe this is still important to all true philosophers, but became less clear. Pierre Hadot dedicated his life to recover this point of view. A life well lived.

Answer 5

1.) The lenght of „1 meter“ was defined as the length of a standard meter kept at Paris. Hence measuring the lenght of a private meter stick has to be done by comparing it to the standard meter. There is no self-reference.

Later was used an other equivalent basic unit of length, the wavelength of a certain atomic frequency. Today the unit-lenght is defined via the speed of light.

See https://en.wikipedia.org/wiki/History_of_the_metre#International_prototype_metre

2.) A kind of measuring the human intellect is an intelligence test, conducted against a given questionnaire as benchmark. With respect to the questionnaire and the distribution of the result within a sample of test persons an intelligence quotient is defined.

Of course, there is much debate which questionnaire and which sample one should choose for the benchmark.

The debate generalizes when considering AI. For example it is discussed whether the Turing test should be accepted as the benchmark for the intelligence of a computer.

3.) Summing up: When humans judge human intelligence, I do not see the problem of self-reference. But I see a certain arbitrariness in defining the benchmark.

Answer 6

No, it can't.

To judge an intellect completely, the judging intellect needs to judge itself which is impossible. I am saying so because, to judge a thing COMPLETELY, at least the judged or the judge must be steady. This never happens in the case of intelligence. If you say, 'Yes, it can.', that must imply the judging intellect is judging it partially; not completely.

'To judge an intellect itself' implies the judging intellect is judging another thing (...may not necessarily be another intellect. But you should consider them as two things). The ability of the judging intellect (it may not be sometimes good) may be because of the intellect judged. Otherwise you will have to consider them as two separate things. This certainly does not match our assumption.

Again, if something 'within or without' judges intellect, we must call that judging thing by some other name. A thing 'behind' the intellect or a subtler thing only can judge the intellect (if you didn't mean other intellects).

We often judge our old decisions. This becomes possible because change, growth or development occurs in the case of intelligence due to some other known or unknown factors. To make this possible in the case of AI, updates or other installations by humans are necessary.

Answer 7

Short Answer

The intellect certainly can measure itself, and it is so useful and common, there's an entire discipline devoted to it: psychometrics. Since you seem uncertain about the possibility, let me suggest that your extended metaphor with measuring distance presumes facts about objectivity that might be an impediment to understanding how intellects judge themselves, and others.

You may not be able to use a measuring stick to measure itself, but you CAN use another, more precise measuring stick to measure the first. The same is done with intellects.

Long Answer

Measurement of spatial extension is well understood by philosophers of math and physics. Let's clear up a few notions there first. Having a meter stick is by and large an exercise in intersubjectivity. First, the concepts of accuracy and precision do transfer to psychometrics, so let's start with a basic philosophical idea. That is, imperfect knowledge is not evidence that knowledge cannot exist. In fact, from an epistemological perspective, the fast majority of thinkers subscribe to fallibilism:

fallibilism (from Medieval Latin: fallibilis, "liable to err") is the philosophical principle that propositions concerning empirical knowledge can be accepted even though they cannot be proven with certainty,1 or in short, that no beliefs are certain.3 The term was coined in the late nineteenth century by the American philosopher Charles Sanders Peirce, as a response to foundationalism.

Maybe it's my bias from my computer software engineering background, but claims of certainty by sophisticated thinkers are almost accompanied caveats. In engineering (See "What is the Philosophy of Engineering About?" (PhilSE)), it's a given that whatever information and knowledge you have when prototyping and testing a system, you are bound to learn something new when you intentionally break the prototype. This is because the universe is very stochiastic. Any empiricist worth their salt is skeptical of rationalist claims that conclude with certainty about physical processes. Human reason is demonstrably defeasible, and informal logic is far more useful and widespread that formal types.

Now, as far as measurement is concerned, sure, no measurement is exact, but exactitude is simply a matter of consensus, as all standards are normative. A meter is a convention, and it's use is a convention, and the exactitude of its definition is a convention. None of that makes the meter useless. In fact, despite all of the normativity, the meter is one the most objective human constructs in existence.

Now, what about measuring the mind? The second thing to know is that the mensuration of abstractions is achieved and used daily with operational definitions. This of course became quite popular around the time of Gilbert Ryle and the behaviorists from which he learned. In his The Concept of the Mind, he attacks the Cartesian mind-body duality as a category mistake and says that the body and the mind are part of some continuum. From WP:

I hope to prove that it is entirely false, and false not in detail but in principle. It is not merely an assemblage of particular mistakes. It is one big mistake and a mistake of a special kind. It is, namely, a category mistake.

But, the point is taken that the measurement of the mind is not as easy as the measurement of a table. But that's where the operational definition including norming test results comes in. Instead of comparing extensions of space, psychometrics compares results on Q&A instruments to do the same. From WP:

A norm-referenced test (NRT) is a type of test, assessment, or evaluation which yields an estimate of the position of the tested individual in a predefined population, with respect to the trait being measured. Assigning scores on such tests may be described as relative grading, marking on a curve (BrE) or grading on a curve (AmE, CanE) (also referred to as curved grading, bell curving, or using grading curves). It is a method of assigning grades to the students in a class in such a way as to obtain or approach a pre-specified distribution of these grades having a specific mean and derivation properties, such as a normal distribution (also called Gaussian distribution).

Now, the most famous of these psychometrics is likely the measurement of the G-factor, what is known as IQ tests. Simply put, human beings can be appraised according to various models and definitions of intelligence by observing behavior and comparing those behaviors with the population. The Cattell-Horn model of intelligence is extremely sophisticated and reliable when considering correlations with other behaviors. But it's not the only. There's Gardner's MI theory, and really, there's a pleothera of ways to test the mind, some of them domain-specific skills. In diplomatics, the military, education, and missionary work, language assessments carry great weight in measuring the mind. In fact, every school class that is based on empirical evidence (standards-based or rubrics-based in the lingua franca of educators), purports to objectively measure the mind. And what about the ICD and DCM's classification of disorders of abnormal psychology. All measurements of the mind.

Summary

Is measurement perfect? No. Is knowledge exact? No. (See WP's article on "Justified True Belief" for problems with defining knowledge) But does that mean that measurement cannot occur even in abstractions? Absolutely not. To make such a claim would be to deny a rather obvious fact about the pragmatism and widespread use of psychometrics. It's even arguable that the purpose of folk psychology is nothing more than an intuitional measure of the minds of others.

Reply to Criticism

You seem to have miss the point of my question entirely. Of course you can measure other intellects with your own. Of course you can measure a meter with another more precise one. My question was: can an intellect measure itself? Since you have brought up psychometric: can a person create completely by itself a psychometric test to measure at a later time its own intellect somewhat precisely? Common sense seem to suggest: since he created the test he will know the correct answer for all the questions, rendering the test completely useless. – Noumeno

You are asking a dubious question. I've measured my intellect several times with an IQ test. Because I didn't make the test doesn't mean that I'm not measuring myself. Is the only way I can measure my height to make my own meter stick? Of course not. That's just silly, right? But let's indulge the silliness of the claim that one has to build the tool for measurement oneself. The trivial case is that I simply do my best to replicate someone else's tool. No meter stick? I buy one and then make my own... But let's take a better example. Let's say that I can write software to build a computer adaptive test, and use software algorithms or heuristics to provide problems within a specification. If I want to see if I have mathematical intelligence, I simply write software to generate random data and check to see if I perform those operations correctly on the data. The same could be done to test spatial intelligence or interpersonal intelligence or for patterns. In this way, I design the test to test me by creating its own problems. Again, the idea that the only way an intellect can... measure itself by its own means is just silly. No self-designed test is done without the assets of knowledge conferred by society. You think your idiolect is your invention? No, it's essentially a loan from society. You think the technology for taking tests you surround yourself is your invention? Nope. Society provided it for you, unless in some rare case you're a true innovator... Your case is not can the intellect test itself, but rather, can a test of knowledge-that truly measure the capacity of the test taker, and even then, why not? Let's say I write an English test, and then I answer all of the questions correctly. Isn't that still a valid measure of my knowledge-that and knowledge-how? Let's say after I administer it, I have someone else grade it, a fellow teacher, and they say I've gotten most of the problems correct. Why is that not a valid measure of my intellectual ability? When I used to teach, I gave my students the answers to the questions because I... expected an adequate justification. So, knowing "answers" is one small part of learning. I think about the only case that the intellect can't measure itself in the sense you seem to be getting at is when the intellect lacks the ability to measure itself, which itself is an intelligence. That's most obvious in the Dunning-Kruger effect, where people have a poor time assessing their own performance. So... I don't think that the metaphor between ruler and human mind holds much water. The former is an artifact of the latter... and the latter has a collection of abilities to self-reflect or self-measure that can be furthered through education and improvement of intelligences related to self-awareness, at least as a general precept.

Answer 8

Generally, judgement is done collectively, even when its done individually. This is because its the society that one is born in and lives that shapes the capacity of judgement. There is no free-standing intellect.

When we judge whether our own reasoning powers are good, bad or indifferent we use a standard that we have learnt and thus that came outside of us.

Moreover, judgement in practise, especially of ourselves, when important things are at stake, for example a crime, is often formally done by others. For example a man in the real world is not allowed to excuse himself from a crime, but has to come in front of a jury of his peers that judge him. College students aren't allowed to award themselves ratings and nor can phd candidates award themselves doctorates. In both settings, their claims are examined through outside tests, exams and work. The suspicion on all of this is that a man or a woman is likely to be prejudiced in favour of him or herself.

Of course we can all judge ourselves. I can judge whether I fixed that light properly yesterday or whether I know how to drive. But the question that arises os whether those judgements have any substance. In small things, we must judge and the judgement may be with or without substance, but because its in a small matter, its not important. In large matters, judgement is left in the hands of others.

Answer 9

We have two tactics for standardising things: comparison; and referencing fundamental physics from first principles. The move towards the latter relates to the 'journey of unification' we see in the history of physics, discussed here: Is the idea that "Everything is energy" even coherent?

The meter started as the circumference of the Earth divided by 1 million; the now SI unit is related to distance light covers in a certain time, and SI unit of time to vibrations of a caesium atom. But there are other ways to define and use time, like sidereal time, years, lunar time, and timescales involved in the cosmic distance ladder. We take it for granted we can reference between these in terms of the SI scale, but the relation between these is not exact and requires intercalation, and especially at very large or small scales a fundamentally different reference may be appropriate, like the age of the universe which currently has a large error in SI terms. It is possible to run away with the idea everything is 'absolutely relative', it is not, as discussed here: The ontological logic of the relational interpretation of Quantum Mechanics And, the shift from the causal-narratives to symmetry-transformations more generally, here: Is the idea of a causal chain physical (or even scientific)?

What is intelligence? Not easily answered, because it is the kind of term about which we feel 'we know it when we see it'. We implicitly associate 'degrees of consciousness' with ability to solve problems, allowing euthenising of some humans with brain stem death, and ranking rights of different animal species, from dolphins and chimpanzees which have substantially enhanced rights to good treatment including intellectual stimulation, many rights for pet species like dogs although sentiment is a bigger factor than intelligence, to moves for better treatment of the most intelligent farm animals like pigs getting more space to raise piglets in the UK, to reduced rights for 'vermin' species even fairly intelligent ones like crows, to almost no rights for insects and shellfish. We use usually intelligence like this, comparatively.

We have many conflicting folk-intuitions about intelligence, it is meshed into different contexts or modes of life, it is part of different language-games. IQ is an attempt to replace a supposed noumena called in psychological studies 'g', with measurable phenomena ranked using IQ tests. Detailed discussion of strengths and weaknesses of this approach here: Do IQ tests measure intelligence? TLDR, the correlations between performance of individuals in different mental tasks can basically be related to health as discussed in this great paper A dynamical model of general intelligence: the positive manifold of intelligence by mutualism; but, when we look at specific examples of humans demonstrating exceptional intelligence in problem solving, we often find people like Ramanujan or Einstein who show not general intelligence and capability at life, but very unusual focus towards specific mental skills, and a better way of thinking about intelligence than IQ can help us to become a world more supportive of neurodiversity which will benefit everyone (contrary to widespread belief Einstein's IQ was never measured, because he wholeheartedly shunned that approach to thinking about intelligence).

Wisdom was considered a priority over 'cleverness' in the ancient world. I suggest wisdom is specifically the kind of problem solving that finds solutions for dilemmas, and at best reframes situations so that apparent lose-lose options can be opened to a win-win outcome. Science has seen a shift from focusing on making good decisions, to having the best modelling of phenomena, with good decisions assumed to follow. That is, solving the problems that open up the space of choices - I suggest that relates to a shift from a cultural value for wisdom, to one placing higher value on intelligence. Discussed here: Wisdom and John Vervaeke's awakening from the meaning crises?

Relating back to the 'hierarchy of being' we associate with intelligence, to be more conscious is a difficult problem for modern philosophy, because what consciousness is has been such a disputed area. Some relevant attempts at providing conceptual clarity, are the Global Workspace Theory of conscious awareness, Integrated Information Theory, OrchOR, Strange-Loops & Tangled-Hierarchies.

I see the crucial step as the shift from a local brain-led picture of consciousness in an absolute objective world, to subjectivities and world jointly emerging in an intersubjective space. Here the physics picture of what is non-comparative as related to continuous-symmetries again becomes important: the conceptual refinements and of intelligence that we call language which assemble and organise the world into salience landscapes and gives us cognitive grip on things we find meaningful, are not held privately, but gain their meaning and use from inviting each other into aspects of our subjectivities - just like with understanding Bosons and Fermions, this relates to the symmetries involved, or not, of various phenomena as we imagine ourselves into the perspective of others (science focuses on clear symmetries, psychology on fuzzier more heuristic ones). We can picture this emergence of an intersubjective space, with the metaphor Indra's Net. Discussed in more detail and related to wider philosophical thinking here: Is the Categorical Imperative Simply Bad Math? :)

There is a great picture of Ch'an Buddhist awakening, introduced in Liberating Intimacy: Enlightenment and Social Virtuosity in Ch'an Buddhism by Peter D. Hershock. Buddha's first account of what had changed after arising from the Boddhi tree was to expressly deny having become a new kind of being, but to say only 'I am awake'. Discussed here Have any thinkers applied empiricism to the dreaming and deep sleep states? and here What resemblance is there between Moksha and Nirvana? Awakening is clearly described as being 'more conscious', more present, of interacting with what is in some way more real. That very often leads to unproductive debate trying to clarify further. Hershock helps with a picture of awakening as, above all 'intersubjective virtuosity', that to diminish attachment to the conventional self opens up this capacity to see into the minds of others, and so to act for all beings. Both a fundamental sudden complete reorientation, and something that skillfulness can be attained in enacting.

I find Hofstadter's picture of strange-loops the most compelling account of minds. We create our picture of the world not as a static absolute background, but through the process of moving around a point of focus, our working memory or or Global Workspace, which processes the structures we use in different areas of life and the relationships between them, into a weltanschaung or meaning-cosmology, discussed here Which philosophers and philosophies discuss "worldview epistemologies"? This is a coherentist picture of world and minds, that minds have a quality of beginning wherever they are, and developing nested and related heurustic explanatory layers into a tangled-hierarchy collaboratively, rather than using the mathematical picture of foundationalism which inevitably is limited by Munchausen's Trilemma, and so avoiding the Halting Problem and Incompleteness that limit such approaches.

Intelligence is emergent, and by understanding it's development through enhancing intersubjectivity, we can picture a direction in which it can develop further (the rare case of exceptional intelligence in solitary species, in cephelopods, seems also rooted exactly in their ability to see into the intentions of predators and prey, see According to the major theories of concepts, where do meanings come from?).

Tegmark and Wu's AI Physicist For Unsupervised Learning, and AlphaZero point to our deepening understanding of what learning means, and pointing a direction to understanding the generalising of artificial intelligence, discussed here What is intelligence?.

AGI will emerge as a reflection of humans involved, rooted in who they are as who we are is rooted in our own ancestors. Not just what we are conscious of being, but language, culture, and all the manifestations of our attention. Rather than being children of our bodies, AGI will be children of our minds. So we should look there, to understand the future of intelligence we will make, together.

We go beyond intelligence as a comparative measure, through understanding the continuous-symmetries of intersubjectivity. We can relate this not only to intelligence and freedom-to-act, but to wisdom, and being more awake as developing skillfulness with intersubjectivity. We make our experience world in our own image:

"Watch your thoughts, they become your words
Watch your words, they become your deeds Watch your deeds, they become your habits Watch your habits, they become your character Watch your character, it will become your destiny." -original source disputed, it seems to have emerged and been refined in multiple places

We will get the AGI we deserve in precisely that sense. This is not a new problem, but an archetypal one, expressed in the story of Pandora's Box, the exiling from Eden, and stealing fire from the gods - but this time it won't be us telling the story, but new beings in who's story we have only a small part.

Apologies for not being more concise.