Artificial Consciousness: Requirements and Implications. centuries is consciousness. True understanding of what consciousness is and what the

Transcription

1 Genac 1 James Genac Philosophy 308 Dr. Robert Greene 2/26/2011 Artificial Consciousness: Requirements and Implications One of the most interesting and complex topics that has been discussed by thinkers across centuries is consciousness. True understanding of what consciousness is and what the requirements are for being conscious have been quite elusive, even to this day. Recently, with the dawn of the information age, a particular amount of curiosity has been displayed by scientists in the field of Artificial Intelligence (AI), who seek to create an intelligent computer. Always dreaming big, designers of various robots and AI systems seek to simulate actual intelligence and transcend 'clever programming' (Henig). To achieve the goal of an intelligent machine, one must understand how the underpinnings of the mind are laid, how consciousness makes us differ from circuitry and electricity and thus allows us to solve problems in a manner that seems to be noncomputational. In this essay, I will first make an educated assertion of what consciousness is through a set of criteria; then I will determine if current technology meets these criteria and to what degree the criteria are met. Last, I will make some suggestions for possible research methodologies to steer scientists in the right direction to create a truly conscious artificial being.

2 Genac 2 Consciousness It is difficult to pin down exactly what consciousness is it is quite apparent that human beings are conscious, but what about other living organisms? Are apes conscious beings? What about rats, turtles, insects, and bacteria? According to Stanford's philosophy department, the idea remains disputed and highly controversial to this day. Sensory consciousness, which is an animal's ability to perceive its surroundings and 'access consciousness', which is essentially an animal's ability to mentally represent the world and choose actions based on those representations are the least disputed ( Allen, Animal Consciousness ). In this paper, I will refer to these first two as 'sensory perception' and 'cognition' respectively. The remaining two that Stanford lists are phenomenal consciousness, which are subjective experiences of what it feels like to exist (such as emotions, pleasures and pains, etc) and also self-consciousness, the ability to distinguish the self from other beings and objects ( Allen ). I will refer to these as 'subjective experience' and 'self-awareness', respectively, in this paper. Though these criteria catch a great deal of what many people see as consciousness, I believe that all of these presuppose a physical structure that acts as a unit, though it is comprised of trillions of atoms and billions of cells. The physical or metaphysical unity of a being allows it to perceive, think, and act. A definition of consciousness certainly wouldn't be complete without mentioning that conscious beings possess a unity, or 'organization,' through which they exist; it also provides a good starting place for a discussion of the requirements for truly intelligent robots. To be a conscious being, one must possess both a physical unity and a mind. First of all, it is important to note that all forms of life, even those that are generally considered unconscious, possess an organized physical form that works as a whole to reproduce, perform functions, and

3 Genac 3 survive. Single-celled organisms are a clear demonstration of how an unconscious living being can exist through this physical unity, but does modern technology have physical unity? The answer is clearly yes, and unity of this sort has existed for centuries in machines. It is not saying a whole lot to state that physical unity was designed into these machines to accomplish a goal: cooperating parts comprise a functioning, theoretical whole. Modern technology, as well as many of the lower forms of life, fail to possess what Aristotle would call a mind, and thus has no mental unity. In The Death and Life of Philosophy, Dr. Robert Greene quotes Aristotle's proposition for the unity of the mind: 'it is necessary for one [thing] to say that [any two other things are] different' and adds, [F]or the mind to distinguish any two or more items it is thinking about, it itself must be a unity... Furthermore, if the mind thinks them separately and doesn't combine them, how can it compare them? ( ). The question is then raised: Does a computer really possess the ability to compare two different things? I would argue that it doesn't. While an ALU (Arithmetic Logic Unit) inside a CPU can take two input binary numbers and run them through a series of gates to test their equality, it is not itself doing any meaningful comparison. The computer lacks the intrinsic ability to actually understand that 1 is not equal to 2 and is only mechanically processing electrical signals. Moreover, it lacks the ability to even understand what '1' or '2' actually means it is merely a shell with circuitry that is hard-wired to output 'not equal' when given two items (binary numbers) that its conscious designers have predetermined to be not equal. Computers also do not store, process, and compare data as conscious beings do. Binary application data is stored on a hard drive, and called to active memory to be processed. At first glance, this seems to mirror long-term and working memory in conscious beings, but at the low levels of memory storage in a computer, very few generalities

4 Genac 4 are actually made depending on the algorithm, disk space may be filled with files in contiguous blocks, in blocks linked in a chain, or through trees of blocks containing index tables (Silberschatz ). Files may reference other files, but it is done so explicitly and a vast amount of true redundancy is created in a hard drive because of it. The genius of the structure of human memory is such that every time the brain encounters something new, it automatically links incoming sensory data to preexisting general data based on similarities with existing general knowledge, forming a hierarchical data structure (Greene, 160). To apply this to a computer with traditional computational algorithms, an entire sweep of memory would have to be executed each time a new input is received, and instead of storing the information as its own entity, the entity will be broken into pieces with connections to match existing data, while only the distinguishing characteristics are stored as the entity itself. Generalization is also unnatural to modern computers, and it is a layer of software on top of the hardware, using the hardware's 'equal' functions many times to find something 'similar.' A good example of this is stringcomparing applications such as spell checkers on a low level, each character is compared and matched up one by one to character strings in a dictionary file it's not actually telling a user that the result set of words are similar to the input word; it is telling the user that this word plus or minus a few characters is equal to this list of words. In other words, this is clever programming and not genuine generalization! To reiterate, computers lack the base of consciousness the genuine way the mind acts as a whole and identifies whole items to collect data from its senses, organize and generalize this data, and also derive meaning from its internal operations with the data. Though a mind is absent from computers, the question remains: To what extent do computers meet the 'physical unity' criterion of a conscious being?

5 Genac 5 Machines may, at the bare minimum, seem physically analogous to simple organisms in that they have a unified, organized form to match their function, but there is one significant distinction between the two, and I go back to Aristotle to examine this. In his book, Dr. Greene lists Aristotle's four causes for matter's interaction with form which is responsible for the existence of beings and objects. The first cause is known as 'from what', or what something is constructed of. All physical objects share some commonality in this cause because both beings and objects are made of matter (145). The second cause, or 'by what', refers to the processes an object or organism goes through to get to its current state. Again, there is no significant difference between beings and machines because all matter is subject to the same laws of physics (145). The third cause, or the 'form(al) cause' states that an object's structure is also a cause for its own existence (146). Greene's example of DNA's form or structure coding a living organism and itself being a cause for its existence in its current state shows that even modern day scientists can think of form as a cause; and it can even be applied to machines too machines are designed with a particular form because the form is necessary to accomplish its desired task. This ties in with the 'final cause', which deals with why the organism exists (namely, to fulfill its own natural function) (146). If real goals objectively exist or not, this is precisely where a significant difference between life and machines lies. Machines are produced by humans to accomplish a goal, and thus are designed with the physical form for a purpose. Living organisms are reproduced by parent organisms with the intrinsic intent to survive, while being chemically endowed to reproduce more similar organisms. This difference is due to over one billion years of natural selection that shaped living organisms, allowing the more competitively complex ones to reproduce more while machines have been designed to simplify repetitive tasks over a fraction

6 Genac 6 of the time that it took for conscious life to arise. Is it any wonder that programming a truly intelligent being is such a difficult task? The current process of designing AI is backwards and unnatural; we are trying to emulate the theory of mind in machines with physical forms designed for computation, not understanding, intelligence, and consciousness! So then, perhaps the 'final cause' or 'that for the sake of which' as Aristotle actually puts it, is indeed significant. Scientists lack the knowledge of the physical inner workings of consciousness and thus cannot even begin to engineer a conscious machine because they lack the means to do so. Simply put, if you have a goal of, say, traveling at the speed of light, light must be understood well enough to engineer a machine with this purpose. For this reason, it will take some significant advancements in neurological research to discover physically how the brain works. Until then, machines will not be able to possess a mind and cannot be conscious. Nonetheless, if machines can simulate some of the sub-criteria of consciousness such as sensory perception, incredible accomplishments may be wrought by them despite their limitations. Sensory perception is an entity's ability to experience the physical world around it through various means. As I mentioned before, machines do not have a central entity through which to experience anything, so even if we could plug in a human eyeball to a computer and harvest the data, the computer would not see anything because the computer does not have a mind. However, plugging in an eyeball, or more realistically using a digital camera with a computer can allow it to truly collect data from the outside world, store it, and recall it with stunning accuracy. In fact, machines can be designed to sense sight, sound, and touch (one could conceive of a robot that could smell or taste, too) and all of those capabilities allow astounding applications to be created. First, video and audio recording devices have changed life forever,

7 Genac 7 from the first camera to modern-day video chat and voice and face-recognizing software. Though no understanding is achieved by the machines capable of mimicking these senses, they serve as powerful information-gathering and organizing tools for their users. In fact, computers record data in a detailed and contiguous manner that is far more accurate than most humans are capable of. Take for example, a security system in a bank which can record and store detailed videos of bank robberies from many different angles and document physical characteristics of the perpetrators this is something that no human being could do. Probably the most obvious fact about computers is that they make massive computational problems fast and easy to solve, which happens to be the bane of humans. Humans, having minds, can interpret a machine's output in a meaningful way humans can use machines to augment their own abilities, taking advantage of the strengths of a computational machine while compensating for its lack of true intelligence. This synergy between humans and machines clearly demonstrates some of the incredible accomplishments of machines simulating sensory perception; we shall now continue to ask if machines can simulate cognition, and if so, what they can accomplish. Cognition, or access consciousness, is the term used to describe an entity's ability to mentally represent and process data gathered through sensory organs. In living organisms, cognitive processes take place on both an active and passive level, typically dubbed 'conscious' and 'sub-conscious' thought. When people think of the word 'cognition', conscious thoughts will often be the first thing to come to mind (literally). In The Death and Life of Philosophy, Dr. Greene again quotes an elegant description of conscious thought from Aristotle: [D]eliberative imagination is in [animals] that have reason... [who] can make one [piece of data] out of many (160). In this way, presupposing that a mind acts as a unity, a function of this unit is that it takes

8 Genac 8 sensory data, say from a cow, and combines it into a whole unit with a hierarchy of generalities below it. Suddenly, from a spotted, four-legged, warm-blooded mammal with an udder that makes a moo sound comes the idea of a cow, and from the cow, its attributes can be further unpacked and used for comparison to other objects like a stuffed animal that looks like a cow (we get the looks like a from the generalities that we form internally). It is this ability that machines precisely lack: They cannot organize data into conceptual wholes and unpack the wholes into generalities. Moreover, they cannot even derive meaning from the data that they are able to store as no unity or mind exists to perceive it. Indeed, machines are currently incapable of what we would call conscious thought, but what about sub-conscious thought, and is it even cognition? I believe that subconscious thought should still be thought of as cognition as it is a way of responding to an environment through mental control, given a certain stimulus. For example, the way most people commonly have the same order and motions of washing themselves is indeed unconscious, but it is still highly dependent on sensory data to function (i.e., if the person just got a haircut, the hair-washing subroutine would be adjusted, yet unconscious motions would still be largely used). Other examples of subconscious cognition in animals are commonly practiced movement patterns, better known as 'muscle memory'. Muscle memory allows us to do a variety of tasks instantly with nearly no conscious thought good examples of this are climbing stairs, walking, speaking, or playing an instrument, all of which are learned at some point in one's life and tend to remain learned as long as they are regularly called upon. Interestingly enough, computer programs, which are series of commands and routines that perform a predefined task, are almost exactly identical to the kind of processes that living beings store. The medium of

9 Genac 9 storage nerves versus a magnetic disk as well as the overall method of storing and recalling the data differ, but the same general concept of subconscious cognition exists in the form of a stored template of behavior that relies on sensory data to be activated and executed. But still, a machine has its limitations; it is unable to manipulate its own applications as we are able to manipulate our subconscious processes, such as washing our hair after getting a haircut. One may think at first glance that a program's capability of tolerating errors would be analogous to a human's ability to react to changes, but I assure you, it is not. This is another instance of clever programming to deceive your perception of the machine's lack of intellect. The key difference is that computer programs are purposely hard-coded to catch errors that could be predicted to happen each situation that a machine seems to handle was thought of and coded by its designer; conscious beings can handle and effectively respond to anomalies that were completely unforeseen, such as having to dodge a dive-bombing attack of a hungry bird at a picnic. With this in mind, we have again returned to the fact that a unified mind that can react to dynamic situations through generalization and learning is needed for a machine to have consciousness; in other words, such a conscious machine could write and modify its own applications to adapt to its environment. It is curious to ponder how fast conscious machines with lightning-fast computation speed would adapt and change themselves, especially when the domains of their own minds are be understood and accessible to them. That day is likely far away, but these beings may possess the last two controversial criteria of consciousness self-awareness and subjective experience. To have self-awareness and subjective experience presupposes a unified mind that, as we determined earlier, machines do not have. Simply put, they have no self to truly be aware of, nor

10 Genac 10 a self through which to experience the world. Though it is pretty clear that machines today cannot truly possess self-awareness, clever imitations have been made such as M.I.T.'s Nico, a robot that is designed to look into a mirror and rate the reflection with a score on how likely it is to be 'itself' (Henig 9-10). Nico performs trials by comparing its own movement to the movement in the reflection, but is this even hitting the mark of self-awareness? In a sense, Nico possesses the knowledge of where its body parts are, but that is about as deep as Nico's 'mind' goes, as it lacks the capability for knowledge of itself other than through an image, and would even be tricked by a reflection of a human mimicking its every move. Humans can be aware of much more than this, and often other forms of self-awareness delve into subjective experiences, such as emotions, pleasures, and pains. In The New York Times article The Real Transformers, Henig quotes Daniel Dennett, a philosopher who was consulted on the development of a robot named Cog: [it is] unlikely [that a conscious robot could be built] but thinks it possible for a robot to be built that could know what it is doing and feeling and why (Henig 10). Dennett may be playing fast and loose with his vocabulary here, because in order to truly feel or experience, a unity must be present; something must be there to subjectively experience a feeling. This is not to say that a simulation would not be possible one could surely conceive of a robot that would be programmed to respond with a simulated emotion, and perhaps even alter its internal states as a response but with current technology, all that would exist is an empty shell with clever programming. To summarize and reiterate, it is pointless to speak in terms of robots being selfaware or having feelings when no mind exists within them. That being said, with my final remarks on the subject I will offer some suggestions for designing a conscious being along with an interesting example that I believe is heading in the right direction.

11 Genac 11 Creating a conscious being requires some understanding that we do not currently possess; we need to understand how the mind works if we are to artificially create one. The big question is, how did consciousness come about with seemingly no designer? Moreover, how did material living beings over time gain complex structures that enabled them to perceive and observe matter? Organisms exist in their form because other organisms survived in an environment long enough to reproduce themselves; changes in the system and energy allowed more competitive (often more complex) beings to come about in this manner. This is a refreshing change in approach that may eventually aid in intellectually advanced robot design, and I am not the only one to think so. Hod Lipson, a professor at Cornell University, has been active in research on machine evolution in a competitive environment ( Hod Lipson ). In one of his experiments, a mother computer uses an evolutionary algorithm that searches through robot schematics and tests them against simulated real world conditions and keeps the successful ones as its products ( Hod Lipson ). Another popular genetic algorithm on the internet is called boxcar2d, in which a series of randomly generated cars undergo simulated natural selection when tested via a physics engine on a randomly generated track. Though it is greatly entertaining and it does demonstrate that design does not require a designer, it cannot truly be considered true evolution; these cars do not reproduce themselves, but instead they exist as mere models on the program that are chosen for performance and the generated 'children' are randomly created with the phenotypes of other successful cars (Boxcar2d). While it is almost certain that consciousness will not arise from current experiments such as Lipson's and boxcar2d, it is at the very least a step forward. Considering the billions of years that it took for life to arise on earth and the infancy of the science uncovering life's history, it would be unwise to expect consciousness to be developed in

12 Genac 12 this manner anytime soon truly, even single-celled life is much more complicated than our most advanced machines. With all that in mind, it is not to say that building conscious life is impossible, but rather improbable with our current understanding of life. Realistically, the day in which a truly intelligent machine will be built is far off and we will have to just settle for the incredible, yet deceptive empty shells of our current computational machines that surround us and interact with us as if they are alive, and accomplish so much as a result.

13 Genac 13 Works Cited Allen, Colin. Animal Consciousness. Stanford Encyclopedia of Philosophy. 13 October Web. 10 December < Boxcar2d 23 February < Conscious Understanding: What is its Physical Basis? youtube.com. Google Tech Talks. 17 March Web. 10 December < Greene, Robert. The Death and Life of Philosophy. South Bend, Indiana: St. Augustine's Press, Print. Henig, Robin Marantz. The Real Transformers. New York Times. New York Times, 29 July Web 10 December < Hod Lipson cis.cornell.edu. Cornell University Web. 10 December < Silberschatz, Abraham et al. Operating System Concepts. Hoboken, New Jersey: John Wiley & Sons, Inc., Print.