The Elements of Intelligence
Christoph Adami · Artificial Life · 2023
Can machines ever be sentient? Could they perceive and feel things, be conscious of their surroundings? What are the prospects of achieving sentience in a machine? What are the dangers associated with such an endeavor, and is it even ethical to embark on such a path to begin with? In the series of articles of this column, I discuss one possible path toward “general intelligence” in machines: to use the process of Darwinian evolution to produce artificial brains that can be grafted onto mobile robotic platforms, with the goal of achieving fully embodied sentient machines.After reviewing the history of Artificial Intelligence research (Adami, 2021) and discussing the components, topology, and optimization methods used in artificial neural network research (Adami, 2022), we now take a step back to ask ourselves, What is intelligence? In our quest to evolve an intelligent system, this is not an idle question. In fact, asking this question will help us focus on essential features of what we call intelligence, rather than being distracted by incidental attributes. Our answer will be guided by the principle that intelligence is an evolutionary response to uncertain environments: that the primary purpose of intelligence is to increase the organism’s fitness.Just as it is unlikely that there will ever be a unique and universal definition of intelligence, it is also unlikely that there will be widespread agreement about what the processes are that contribute to intelligence: the elements of intelligence. The five elements that I will discuss here are rooted in the idea that intelligence is a (biological or computational) trait that enables its bearer to reduce the uncertainty about the world in which it lives (both in time and space) and harness the information it has gained to succeed against its competitors, cooperate with its supporters, and extract the resources it needs from its environment without coming to harm. Recognizing who is friend and who is foe (and using information to defeat the foe and support the friend) ultimately leads to a greater number of offspring.To leverage information in support of organismal fitness, the organism needs to perceive the environment; extract the salient features (those that matter to the organism and can be perceived by its sensory system); make predictions and plans based on the sensed world as well as on what was learned from experience; and, finally, act according to those predictions.1 Such a view of intelligence is very much aligned with the “knowledge-level systems” view of the late 20th century (Anderson, 1983; Newell, 1990), except that those attempts to formulate a “unified theory of cognition” made no attempt to quantify said knowledge in terms of information. An information-theoretic view of intelligence and cognition has the advantage that it can quantify the relation between the “symbols” manipulated by the knowledge system and the things in the physical world that they represent. This is important because, historically, one of the most common criticisms of attempts to formalize (and ultimately engineer) thinking systems conjured up an apparent dichotomy between the “zeros and ones” of computer systems, which are devoid of intrinsic meaning (“strings by themselves can’t have any meaning”; Searle, 1984, p. 31), and the fact that “thoughts are about things.” Information theory quantifies precisely that link, both in computers and in people.Whereas most theories of cognition posit that sensing and acting are integral elements of intelligence because they are clearly part of the “sensory–action loop” (Bongard & Pfeifer, 2001; Clark, 2016; Newell, 1990), here I take the point of view that the sensors and motors themselves are “given” (even though cognition does affect sensing and acting), and I discuss only the elements of intelligence that take place within the neurons of the brain, excluding sensors and motors (often called “peripheral neurons”; McCulloch & Pitts, 1943).2We will see that the elements of intelligence that I will discuss—categorization, memory, prediction, learning, and representation—are all tied explicitly to how information is acquired, shaped, stored, and manipulated.To make sense of the world that we perceive, it is imperative that we can tell one thing from another. How do we do that? How is it that a visual scene (say) evokes in the brain a set of objects and their relationships with each other, rather than a jumbled mishmash of colors and shapes? After all, the shapes and colors we perceive are not discrete but rather form a continuum. For us to be able to differentiate between things, we first need to be able to categorize.Our ability to place objects in the world (and behaviors and ways in which objects relate to behaviors) into categories is a crucial skill that develops early in infancy (Quinn & Eimas, 1997). According to psychologists (see, e.g., Karmiloff-Smith, 1992), categories (collections or classes of objects and events that exist in the world) are formed via a perceptual analysis that filters the raw data, leaving behind an abstract representation in the form of image schemas. We all carry such image schemas with us. If I were to ask you to imagine a chair, for example, you could do that very easily, and even though you might imagine a straight-backed chair with four legs, you would not regard a three-legged stool as something completely outside of this category. In fact, we are able to subsume thousands of different shapes under this one category “chair.” I propose that this ability to form categories is a central element of intelligence: all others that we will discuss build on this one. In particular, building categories allows us to quantify how much there is to know, using the information-theoretic concept of Shannon entropy (Shannon, 1948), a measure of uncertainty.Claude Shannon, the creator of information theory, called his measure entropy because a very similar quantity had been introduced in statistical physics much earlier.3 For the purpose of understanding the concept within the context of cognitive science (and to see its relation with the concept of “information”), I will use the word uncertainty instead of entropy. Shannon defined his uncertainty concept both for continuous (“blurry”) quantities and for discrete (or “sharp”) ones.Let us first write down Shannon’s uncertainty function in the discrete case. To do this, we have to introduce the concept of a variable X that can take on n discrete states xi, i = 1 … n. For the purpose of describing categorization, we might then ask, How do we associate objects in the world that have continuous shapes and colors and features with only one of the n categories defined by variable X? This process (called coarse-graining in the literature; see, e.g., Feynman, 1974), is without a doubt a complex one, involving (as I will describe) a shift from perceptual characters that are described by continuous values to conceptual ones described by discrete values. For discrete categories, Shannon’s (1948) uncertainty function is given byH(X)=−∑i=1np(xi)logp(xi),(1)where p(xi) is the likelihood of encountering an object within category xi in world X. In general, the number of categories (as well as the “distance” between different categories) depends on how useful the differentiation is. For example, in some situations, it might be relevant to make a distinction between two categories (say, “chair“ and “stool”) that is not necessary in others. In other words, the brain tends to operate with just the categories that are necessary to best understand (and predict) the world, given the particular circumstances.How do categories emerge? This is a difficult question to answer, because although they clearly emerge over time via a process of use, feedback, and learning, those processes themselves are somewhat vague. Furthermore, some categories are clearly innate: The fear of the color red in certain birds (Pryke, 2009) is one such example. In a very real sense, categories evolve. Here we will think of the process of categorization as creating a certain number of image schemas that represent the different categories.4 Generally speaking, we can say that categories emerge so that there is a balance between a large enough number of different images to be able to describe the range of salient differences and a small enough number that manipulating these images in one’s head (or wherever they are is not This of categories is described by a shift from perceptual that very much the object in to conceptual that have most of the to the so conceptual that there is no to the perceptual such are called We that even perceptual a certain of because the sensory system can perceive only a range of values. for example, the of do not our perceptual of they without a doubt do so in (and for all we know, even in see & a of the evolution of a perceptual representation from conceptual to a one is the evolution of the The is the system and is to have the the used were the objects to which they the and into to the used to the via made by a into that the for into a the concept is in that the evolution of the number of by than can be via a of others about the objects in the world and is a necessary to intelligent it does not what to take to reduce We can say that what uncertainty is information. of evolution have of information about our world in our very about a of to and in a world that over a we need to information in a different We need measure for information can be as a between the uncertainty we have is our that the uncertainty given all the knowledge we we our world in which we to objects in categories to with then the uncertainty would all categories were in this p(xi) = In that this uncertainty would we instead that some categories with much greater likelihood than others some others might even have p(xi) = then this knowledge is by the of the of information as a of is that Shannon’s entropy to the discrete a that is to the of from the of us in the only the differences information is crucial to In our information in the is in the between neurons and can be and with the information our This to and to the sensory in the context of is crucial for intelligent is the in two other elements of intelligence I discuss and allows us to not make the and it it possible that we form of the world within our memory, the of is not something that is We are so used to information in our world that we take it for from the point of view of information is The of is are in place to the of by which we information has over In the need to information other than it in the organism’s the world on a than an organism’s the the world on the was The were and those that were not the in that no & the world it was also To in such a world, it is necessary to to those and it is that the precisely for this by sensory information to in a world that it is also necessary to from To do this, the brain has to be able to of needs form of is the (and of the of the environment is also called but others call it sensory For a in is not a because the to its the to the is to the of but how can that be is to for the to its by and in evolutionary with this is precisely what complex is an organism needs to of events (often called The of such can be using the a brain needs to a particular of In the this is a of two that the brain needs to (and then one of four possible This can be by a set of a sensory an and an In this the is necessary to and a particular can be will also so that the (and only a particular is within the This in the and the of the the time point to the of the and the the time This particular a If the of is is with the to be with a so so the a 1 is If this the to time as the the the If the is a 1 the a to the which we can as The was not and the is in the to for a a that 1 within this the within the The to of the the is to to I have described here is an of time series is the but it can be up to time For example, in we see the necessary to a which needs two neurons to The in will the the of the of you the with and the 1 with the of the that time series in brains a can this as I discuss of the described here used in brains to time This is difficult to answer because it is very to the of a set of neurons from the and given that the in the for example, from thousands of other neurons that on the it is that the and of the on the can with precisely such an of the is to be for any organism is is in in particular, in & The to the is For this the brain is described as a Clark, 2016; & & it is not only of the that is The information in is also used to the of an organism’s environment example, to that the are the the of the has other for example, it possible to of the is not even based on The that allows to a can be as a that the of the of the from do we that the brain what are the predictions based and the of information for by the of but their analysis is though they make a number of make the the is enough to a between information for and and that make within a using the information they in the as well as information from the The that an the information it has its then it its in terms of the of the an described by a variable that can take on a number of states The can states over so we have to introduce the variable and the over time is described by the we an environment variable that can take on states This environment over time in an that the that on is = = We now the function that the number of by the it is in the environment is in and we that the environment is in then there is a particular for which = the is In other words, to be the has to the Here we that the only the can do this is by sensing the For this, the is with a by variable and we also for that this is this that the of the the can its so that its the and that an sense the environment (or is of the sensed and its then the of a of such will is the of the of the over the possible states of the environment is to be the that the one from in the time in is the uncertainty that the has about the given in terms of the Shannon of is is a to being to the and the in is to the Shannon entropy of the an the environment with a X and an that will the according to the sensed then the will instead is the entropy of the environment given the sensed that the uncertainty the into We can write this using the which quantifies how it is that we will environment given that we sensed of this information for the in is the between and this is given precisely by the Shannon information gained by the This as I is the between an entropy there is to and a entropy to be we quantifies how well an can the of with the of X as it the of into Information can be into using an in to that brains of & 2016; & have that predictions on increase an organism’s for it is also important to have a to with in which a to be In general, the that allows us to our based on is called & is not an intrinsic of an and This needs to evolve. The on is and different (see, e.g., is an important concept in computer cognitive and Here I will focus on only one form of to neural I will discuss a very that is to is also and to the see also This is not a has been used in such as learning, theory, and evolutionary very well be an of what is on in our brains as we what I describe the how to one from a set of n and that We can imagine that this is on so we are in the that will our the particular environment that the each a particular based on information or the This could be one and the is given a and the all of the are An which of the will be to the which the are based on the to a or a This is the part of the We would to the to an based on the and how to best these given the Here we only the in which the likelihood that any particular will be for time is to the is the the that has the of it is make it from the how this to artificial we can think of each of the as a particular of a of neurons in response to a to a particular and the is to the of the given the In a sense, we are a particular sensed (and the it how we the likelihood of each of the possible response given the we for the the we are for the to the given the of the or a that the is that time point we i with a to its we a If the is then we increase the of this a by an to the If is we the is a small We see that the is then the time step this will have a likelihood of all other will be with a to we can say that those that to a will be that the likelihood of this to given the sensory is other are This is of the that the between neurons they (and a was only in the the not on the between two neurons but rather on the likelihood of a particular of a set of neurons which the is given depends on the a large tends to (and it also the than We that the = given the that on that it is possible to that the given by the over the is just a than the n is the number of (or the number of possible in a particular of and is the is by the who would have the a large might make the system but the on the the is a is to on as the are given by the in the for the that we For example, that the in as we the for the of has instead that to 1 in each the with = does not the all, but we can now this using the the of this using = and a = 1 any time the was it was to be (or the and a of = the was We also the to that how the can be to the the was the in the were very to the in well (and is there is an about the of the and the the relevant in a is to the brain example, by a brains have to what variable a that because is such a which sensory (or of useful In one such & with brains with a different of their neurons to their motors that the now is as what was now be the of a with the a system in which a by the with sensors the to the of without from a might ask this was and we only a particular and not discuss The answer is that an has learned to associate a particular with a given sensed we can view this as the the from a series of which are the behaviors used to the for each In the a set of is to which is precisely what in the step of the in the sense of the word by cognitive is a of the world that for the real world 1997). is an of the world within the brain that can be used to in the without to to the some of the world are clearly in our I will focus on neural here because they our behaviors for the most part does not take a of to how important it is to carry these of the The system, for example, on of what an object might so that we have to only a salient of a scene to in the so to to up with an will to discussing in in the can how much a brain about the world in terms of information If we a variable for the world states and variable for brain then what the brain about the environment is just the entropy this is not a because the brain states might be by what the in the sensors in are of the In other words, might be not because the brain is able to the world but because the brain the To a measure of we need to the sensory states so that they are to a brain needs to be able to represent the world there is no in the sensory Such a representation can be by the entropy that the information that the brain has about the world given the This entropy is in terms of an entropy in The representation = can be even than the information the brain has about the world by the of and because the can be This can the sensed information (see, e.g., are because they are relationships between and allows us to our so that our predictions example, what we to sense are how much a brain about the world it to be that representation with in the representation = could be a function for that we can imagine that what we about the world what we will increase our because we can make for intelligent are not has described an based on the entropy between sensors and what we would call are of this function is which is a measure of how much one’s in have what one will perceive some time in the is also possible to the entropy between the states one time point and a time to evolve This to some but does not to of these into of the and principle does take a variable into the context here is given by of we can say that they the of the variable the with is that the information can be (as see the in and is (as we see discussing it is to imagine that a has to a the representation = is the best function that into and using an evolutionary (see, e.g., in which representation is as a to than any other so according to the point of view I be into a it is from elements that build on each other, and have different on the we I have described five elements of intelligence but it is that the number is and there is between these Furthermore, other important elements as have been could that is part of prediction, but although there is we can also that a does not tell us how we use to that because for in an of the elements of intelligence would be both and for a column, I have to these five categorization, memory, prediction, learning, and In the I have described all these elements are rooted in the concept of so we can say that intelligent is possible (and allows us to how much we do not information information to make in allows us to our the was and information about the world in our brains so that we can make based on an understanding of the we use these also to a focus on the information-theoretic of the elements of intelligence is from a point of it does not on particular that our brains use for be that the to and time described is used in very well be the that in brains the In we do not have for or against such a and much is is about how brains perceive sensory such as or In the I will discuss some of and to a to how perceive in to a large number of and who have on the evolution of intelligence, in particular, and I in particular to who brains with and the (and a of as early as for an to We can only about we would be that had been The in this was by I also the in of the Artificial who have their time to has these articles and all are