Artificial Intelligence, consciousness, perception and the uniqueness of the human brain-by Edward Benjamin Ziff
The world is quantal, nearly infinite in its complexity. We must navigate this world to survive. To navigate successfully, we must reduce the complexity of the world by making categorizations. This process depends on establishing relationships between world entities and between these entities and us.
The properties of objects exist in time and space. An object is never the same twice. It may be seen from a different angle or with different illumination. Its apparent texture may change and it may be seen in different contexts. To recognize an object in spite of its continuous variation, we must see the relationships between two or more of its different states. This necessitates comparison of state 1 with state 2. Because the two states do not exist simultaneously, to make the comparison, we must hold in mind a memory of state 1 when state 2 occurs. Or we must hold in mind memories of states 1 and 2 and make the comparison at a 3rd time. Because objects are continuously changing, the comparisons are taking place continuously as well. The properties of the object that emerge as most constant (least variant) become the hallmark characteristics of the object.
The Cubist painters recognized that endless variation is inherent to our world. Cubist art is one stage in the evolution of the two-dimensional representation of the physical world. Egyptian art lacked perspective. Renaissance art incorporated Euclidian perspective into object representation. Cubist art incorporated the continuous variation of object properties perceived by the viewer. However, representation on canvas can only allude to the viewer’s perceptions. Indeed, no physical medium can represent the full aspects of perception by the brain, since these perceptions incorporate physical components (color, texture, size, taste, odor, motion) and psychological components (potential for harm or reward, emotional implications, etc.), the latter not directly connected to the physical representations that can appear on canvas. Also, these perceptions are a part of consciousness, which is only known to self.
Categorization makes possible prediction. Properties associated with a category of objects can be assigned to a new object once it has been recognized to belong to that category. Prediction that is made possible by categorization also benefits navigation, since to choose an optimal action from the many alternative possible actions we must be able to compare the outcomes of these alternatives. This suggests that actions, like objects, can be categorized.
Some theories of perception hold that information about the external world that is captured by the senses is written into the brain much as sound and video are written onto electronic memories of cameras and recorders. Even in the cases of man made video and sound recorders, to represent the physical world information in an electronic form and then retrieve it we need a conversion algorithm. In modern electronics, the algorithm may be a software program that governs the operation of the computer, camera or microphone hardware. The software is a representation of the functional properties of the hardware in a form that can be appreciated by human consciousness. The software can exist in multiple states whose properties are familiar to the educated brain. This enables us to adjust the operation of the software in a purposeful way. We can start and stop recording. We can adjust the brightness or hue of an image. We can even convert speech to text using a combination of hardware and software.
In the brain, the cellular detectors of sight, sound, touch, taste odor and body orientation each have their own biological algorithm, which is provided by the brain circuits that process this information. We could imagine that the visual image of the world that projects upon the retina creates some sort of direct representation in the brain. However even in the retina, a complex processing of the visual image starts immediately. This initial capture is in a sense digital, since the capture is by rods and cones, each contributing a unit (byte) of visual information. However, the activities of rods and cones are multivariant and are transferred to bipolar cells and the bipolar cell activities are modified by amacrine cells, all within the retina, the visual signal metamorphoses during a process of “information” extraction that continues as the activities are passed on to the visual regions of cortex. A similar process takes place with taste receptors on our tongue and odorant receptors in our nasal cavity, and with the hair cells in our cochlea that vibrate in response to sound. These sensory inputs are combined with other forms of information about pleasure, pain and reward. This process is vastly different from making a sound recording or taking a photo, whose purposes are to recreate the original sound or image in the future (listen to music or view a photograph). Furthermore the brain can relate these types of information, visual, aural etc. plus pleasure, pain and reward to establish relationships that make possible more complex categorizations that combine object categorization with possible harm or benefit. A camera has no capacity to ask “When did I see an object like this before and did it make me happy and if so, what did I do to achieve the pleasure?”
The relational activities of our brain are manifest to us in our consciousness in a way that has no counterpart in machine or computer representations or our environment. We may pass a familiar street corner and recall past events (my friend of 20 years ago lived in that building and her mother made delicious chicken soup). The brain also creates anticipation, which is an expectation of the future based on the experience of analogous past events.
An inability to deduce relationships may arise from developmental defects or brain damage and can lead to a breakdown of brain function and reduce our capacity to navigate the world effectively. Such inability may result in prosopagnosia, dyslexia, impulsivity and Capgras syndrome. Perhaps the simplest malfunction is red-green color blindness, which is a malfunction of retinal cones, a defect at the very start of visual processing. Prosopagnosia, dyslexia, impulsivity and Capgras syndrome reflect defects in deeper brain functions that impair object (person) recognition or the attribution of value to objects and actions.
Language represents perhaps the most complex of the relational capacities of our brain. Through language we can make apparent to others aspects of our own consciousness. Language builds upon our ability to perceive relationships. It converts these relationships into an acoustic (spoken word) or visual (written word) code that is familiar to others and that reproduces our conscious relationships in the consciousness of others. The equivalence of the acoustic code (spoken language) to the visual code (written language) depends on brain relational functions. Notably, when the use of the written code is impaired, as with dyslexics who cannot learn to read written text, implementation of the acoustic code to teach dyslexics can overcome the written code learning barrier, as with the language skill methods of Talal and Merznick.
Poggio and others attempted to develop Artificial Intelligence machines that could reproduce aspects of these brain functions. One was a machine that could identify faces or whole persons in photographic images. Our power to do this has greatly increased since Poggio made his first efforts and now cameras can recognize faces in the images that they are poised to capture and can even delay snapping a photo until the identified faces are smiling. Also, once these camera-obtained images are entered into a computer, individuals can be identified. Some of the software for recognition of faces and persons employs procedures that resemble what goes on in the brain, including feature extraction. Thus, we might argue that the brain only differs from the computer in the magnitude of its computational processes. However, remember that the individual human brain has about 10×14 synapses, which is more than the number of stars in our galaxy. Moreover, each synapse can exist in a myriad of states, changing its activity hundreds or thousands of times per second. Yet even at this enormously greater complexity, we could argue that the difference between brain and computer is one of magnitude, not kind. Yet the brain differs in other fundamental ways. A computer only has attributes that are provided by human designers. In this sense, the computer is simply a mechanical extension of the consciousness of the designer (programmer or engineer). So we could argue that the computer is imitating the brain. A carousel horse is an imitation of some aspects of a true living horse, but is unquestionably not the same. By analogy, we could argue that no matter how complex the computer, there will always be some aspect of the brain that is not represented. To overcome this limitation, the computer would have to be a true, living flesh and blood brain, in which case a human engineer or programmer could not design it.
Edward Ziff received a bachelor’s degree in Chemistry from Columbia University in 1963 and a PhD from Princeton University in Biochemistry in 1969. As a postdoctoral student with Nobel Prize winner, Fred Sanger, in Cambridge he conducted early genome sequencing studies. Ed served on the faculties of the Imperial Cancer Research Fund in London and Rockefeller University in New York. In 1982, he joined New York University School of Medicine, where he is Professor of Biochemistry and Molecular Pharmacology and Neural Science, Investigator of the NYU Neuroscience Institute and was an Investigator of the Howard Hughes Medical Institute. Ed researches brain function and neurological disease and was a Visiting Researcher at UFGD in Dourados. He lives in New York, has written for The New York Review of Books, coauthored a popular book on DNA, and is an amateur photographer, video maker, and painter.