Machine Dreaming and Consciousness

J. F. Pagel,Philip Kirshtein

• Editore: Academic Press
• Anno: 2017
• ISBN:
  9780128037423
• Argomento: Ciencia de la computación
• Categoria: Redes neuronales

Machine Dreaming and Consciousness is the first book to discuss the questions raised by the advent of machine dreaming. Artificial intelligence (AI) systems meeting criteria of primary and self-reflexive consciousness are often utilized to extend the human interface, creating waking experiences that resemble the human dream. Surprisingly, AI systems also easily meet all human-based operational criteria for dreaming. These "dreams" are far different from anthropomorphic dreaming, including such processes as fuzzy logic, liquid illogic, and integration instability, all processes that may be necessary in both biologic and artificial systems to extend creative capacity.

Today, multi-linear AI systems are being built to resemble the structural framework of the human central nervous system. The creation of the biologic framework of dreaming (emotions, associative memories, and visual imagery) is well within our technical capacity. AI dreams potentially portend the further development of consciousness in these systems. This focus on AI dreaming raises even larger questions. In many ways, dreaming defines our humanity. What is humanly special about the states of dreaming? And what are we losing when we limit our focus to its technical and biologic structure, and extend the capacity for dreaming into our artificial creations? Machine Dreaming and Consciousness provides thorough discussion of these issues for neuroscientists and other researchers investigating consciousness and cognition.

• Addresses the function and role of dream-like processing in AI systems
• Describes the functions of dreaming in the creative process of both humans and machines
• Presents an alternative approach to the philosophy of machine consciousness
• Provides thorough discussion of machine dreaming and consciousness for neuroscientists and other researchers investigating consciousness and cognition

Section II

Machine Dream Equivalents

Outline

Section II: Machine Dream Equivalents

Chapter Five Sleep Modes

Chapter Six Neural Networks: The Hard and Software Logic

Chapter Seven Filmmaking: Creating Artificial Dreams at the Interface

Chapter Eight The Cyborg at the Dream Interface

Chapter Nine Interpreting the AI Dream

Chapter Ten Creating the Perfect Zombie

Section II: Machine Dream Equivalents

It is now possible to create artificial software and hardware constructs of the underlying biological structures in which dreaming and consciousness form. As a result, and sometimes as a side effect of this process, machines have increased their capacity to meet definition and logic-based criteria for attaining cognitive processing states equivalent to biologic states of dreaming and consciousness. In this section, each chapter addresses a different machine dream equivalent as based on definition, phenomenology, technology, and/or metaphor.

Chapter 5, Sleep Modes, addresses the process of machine sleep in both finite state machines (FSMs) and in networks such as the Worldwide Web (www). In Chapter 6, Neural Networks: The Hard and Software Logic, we consider both the hardware and software of neural net systems. Fuzzy logic is approached from the machine perspective. Chapter 7, Filmmaking: Creating Artificial Dreams at the Interface, is a discussion of the manner in which the computer interface is structured on the same components of dreaming that are utilized in the technical process of film-making in creating artificial dreams. Bidirectional human–computer interfaces are addressed in Chapter 8, The Cyborg at the Dream Interface, as are novel interface systems using nonsensory modalities to extend the capacities of the interface into cognitive systems utilized more often in sleep and dream states than in waking perceptual focused attention. The data analyses, results, and outcomes produced by such systems can be hallucinatory and imprecise, strangely structured, altered, askew, and as hard to remember as any biologically produced dream. The interpretation of such outcomes and processes are the focus of Chapter 9, Interpreting the AI Dream.

At this point in time, we have the technical capacity to artificially create almost anything that we can describe—even systems which we do not fully understand. We are in the process of creating artificial constructs of the large, multiscale, nonlinear, highly heterogeneous, and highly interactive CNS processing system. Chapter 10, Creating the Perfect Zombie, considers the zombie systems under current development that will be part of our near future. It addresses the structure and processing of these systems, as well as their prospective capacity for dreaming and consciousness.

This section is perforce, somewhat technical. In constructing this section, the authors attempt to incorporate logical and inductive perspectives that they have come to their machines, machines that appear at least on some levels to already have the capacity to dream.

Chapter Five

Sleep Modes

Abstract

Sleep and the various finite state machine (FSM) sleep modes (Hibernate, Hybrid Sleep, and Screensaving) are defined and compared to sleep in biologic systems. There is a profound difference between FSM machine sleep (an off state of no function) and biologic sleep (an on state with functioning that differs from waking). The complexity theory of consciousness is discussed in detail as well as the defragmentation or screensaver of dream function. From a systems perspective, periods of low data flow in multiplex interconnected systems correspond to human sleep periods that can be viewed as periods of Internet sleep. This concept of system sleep is developed and conceptually applied to data flow, optimal load, flow congestion, bottleneck routers, and congestion collapse in network systems. During systems-defined sleep, systems actually operate closer to optimal performance than during periods of high data flow and congestion (wake). During sleep in such systems exceedingly complex processing can occur that is far more similar to what we now understand to be dream consciousness than the previously proposed correlates for dreaming as degraded thoughts occurring during screensaver mode. Internet sleep-associated data processing has phenomenological similarities with human dreaming including highly complex, fluxing, always-on mode of indeterminate associative, and potentially creative cognitive processing.

Keywords

Sleep; finite state machine; Internet; Internet sleep; defragmentation; complexity theory; optimal load; flow congestion; congestion collapse; bottleneck router; dream-equivalence

SLEEP (3) (suspend execution for interval)

unsigned sleep (seconds)

unsigned seconds.¹

One of the most profound realizations of our technical era has been that sleep, despite its lack of perceptual input, is an active state of cognitive processing. Before EEGs, and before real-time fMRI and PET scanning, sleep had most often been viewed as an off state of the body and brain. The only logical basis for inference otherwise was the reports of dreams on awakening.

Conceptually adapted, the biological state of sleep has been used to describe a series of “sleep modes” in electronic systems. The original purpose of the sleep mode was to offer a consciously controlled option for extending biologic sleep time. The offended human could reach out from bed and hit the “sleep” button on the alarm clock to shut down and delay the annoying alarm of waking. In this form of sleep mode with the electronic system turned off, sleep is a state of suspended execution with its limitations an extent defined by a timer (see initial quote).

Since their initial use (in mechanical alarm clocks), multiple forms of “sleep” analogues have been incorporated into electronic computer systems, denoted on the personal computer as low-power states (e.g., sleep and hibernate). In these states, the computer is not actively involved in the processing any data. These are suspended states in which the computer is preserved in either the system RAM or the system mass storage device. These sleep modes are power-saving states similar to when pause is pressed on a DVD player. In sleep modes almost all operative systems are turned off. Computer actions are stopped and open documents and applications put into memory. Full-power operation can be reassumed within a few seconds. Sleep, in a personal computer, is a low-power state in which the software state of the system memory is maintained. During sleep mode, your personal computer is not actively processing. The memory is not changed. The computer is in a quiescent state. An external stimulus can cause the “sleep” mode to be exited. The source of this stimulus can be from either a timer, a user action, or such a programmed cue as a “wake on LAN magic packet.”

Many currently utilized personal computers have an “Advanced Configuration and Power Interface” through which multiple designed global power states can be defined for a computer connected to a power source. These states vary from fully operational to connected but mechanically switched off. Four of these modes can be considered forms of global system sleep mode:

S1, Power On Suspend: CPU state and memory state are maintained, peripherals are powered off;

S2, Standby: CPU state and memory state are maintained;

S3, Suspend to RAM: Lower power Standby; CPU state and memory state are maintained “sleep”;

S4, Suspend to disk: “hibernate.”

S2 and S3 are for the most part identical and are often collectively called sleep.

In the Hibernate mode, open documents and running applications are saved to the hard disk while fully shutting down the computer. Once a computer is in Hibernate mode, it uses zero power. In Hibernate mode, a personal computer applies no power to either the system memory or the processor. Memory and processor states have been stored to the mass storage system to be restored at a later time. As in the sleep modes, the system may resume on timer, user action, or “wake on LAN.” Once the computer is powered back on, it resumes where it left off. Some computers utilize Hybrid sleep modes that are a combination of the Sleep and Hibernate modes. These Hybrid modes put any open documents and applications in memory on the hard disk, and then put the computer into a low-power state, yet one that allows the user to quickly wake the computer and resume work. This Hybrid sleep mode is enabled by default in many systems. Hybrid Sleep mode is useful for desktop computers in case of a power outage. When power resumes, work can be restored from the hard disk, even if the RAM memory is not accessible.

Sleep in Finite State Machines

Computer processing can be broken into the categories of real-time and batch processing. In real-time processing, the processor performs calculations in response to changes of inputs, state, and time, and then generates output as a function of both the inputs and the state of the machine. The state of the machine changes as a function of the inputs and the defined present state. These systems are called Finite State Machines (FSM) the actions of which are described by:

$\begin{array}{l}\mathrm{state}(\mathrm{t}+1)=\mathrm{Fs}(\mathrm{state}(\mathrm{t}),\mathrm{inputs})\hfill \\ \mathrm{outputs}=\mathrm{Fo}(\mathrm{state}(\mathrm{t}),\mathrm{inputs})\hfill \end{array}$

For such machines there is a historical record of previous inputs and states implied by the present state. For a finite number of states there is a finite history. Such FSMs are the basic building block for computer software and hardware.

FSM microprocessors have become ubiquitous in appliances and other consumer products being marketed today. These processors usually have either a “low-power sleep mode,” a “wait for interrupt” implementation, or a “wait for input change” mode that corresponds to an idle state for the machine. Fig. 5.1 is a flow chart for a simplistic “wait for input change” machine for opening and closing a chickenhouse door. This machine implements this program in two states, both of which are waiting for a change of the time of day from day to night or night to day; and then opening or closing the door. There are two basic types of FSM; Moore state machines in which the outputs are solely a function of the states and Mealy state machines where the outputs are a function of both the state and inputs. The act of closing the door is performed either by (Moore – state ‘close door’) or Mealy (state ‘open door’ and ‘is it night’). In both cases ‘Is it day’ and Is it night’ act as the stimulus for inducing a transition to another state functioning as transition operators. An FSM typically utilizes parametric programming in which set numerical factors and operations define the conditions...

Indice dei contenuti

• Cover image
• Title page
• Table of Contents
• Copyright
• Dedication
• Acknowledgments
• Section I: Machine Dreaming and Consciousness—The Human Perspective
• Section II: Machine Dream Equivalents
• Section III: The Philosophy of Machine Dreaming
• Index