13 June 2005 TOWARD A HIGHLEVEL DEFINITION OF

13 June 2005 TOWARD A HIGH-LEVEL DEFINITION OF CONSCIOUSNESS Manuel Blum Ryan Williams Brendan Juba Matt Humphrey

From Collins English Dictionary, 1995: conscious adj. • 4. a. denoting or relating to a part of the human mind that is aware of a person’s self, environment, and mental activity and that to a certain extent determines his choices of action. • b. (as n. ) the conscious is only a small part of the mind. Compare unconscious.

My credentials for studying consciousness:

• - My credentials for studying consciousness: - 3 rd grade, Bernard Blum (Dad) - brains - Richard Schoenwald - Freud - In the Neurophysiology Lab of Dr. Warren S. Mc. Culloch and Walter Pitts – A Logical Calculus - Hartley Rogers - Recursion Theory - Marvin Minsky (Ph. D advisor) - AI - When I got tenure, …

A complete summary of my results: • Consciousness is a kind of flashlight. The flashlight beam is a limited resource (what Minsky calls a serial bottleneck) • Our conscious self is offered a set of choices. The beam of consciousness selects one. This repeats ad infinitum

Applications Philosophy • Understanding “understanding, ” “selfawareness, ” “free will, ” resolving: “All theory is against the freedom of the will; all experience for it. ” — Dr. Johnson Science • Building good theorem provers (as opposed to proof checkers), and better CAPTCHAs

Why Consciousness Now • There are many researchers trying to understand consciousness, including – Philosophers: since forever – Neurophysiologists: f. MRI studies! – Cognitive psychologists – Roboticists • No complexity theorists that I know of – Les Valiant comes closest – Leslie Lamport “How to Write a Proof”

Why should Complexity Theorists study Consciousness? • They have not thought about it, but they are well-suited to develop theory – Automata and Turing machines – Knowledge (Zero-knowledge) – Randomness (Pseudorandomness) – They understand that complexity rulez • Researchers in The Scientific Study of Consciousness cry out for a theory

Our approach is to define a formal concept of consciousness, called CONSCSness • CONSCS = Conceptualizing Strategizing Control Systems • CONSCSness = the state, condition, or quality of being CONSCS

Informal Notion of Entity & Environment • An ENTITY is an organism (such as an ant or a person) or an organization (such as an ant colony, a cultural group, or a nation) — in relation to an environment • An entity's ENVIRONMENT at time t is the entity's (relevant) world at time t: the context in which the entity functions

More Formal Definition of Environment & Entity • An ENVIRONMENT is a computational device having the full power of a Universal Turing Machine (UTM). It is connected to a collection of other computational devices, called ENTITIES • An ENTITY is a computational device that (together with memory supplied by the environment) has the full power of a UTM

You have just seen the only “definitions” in this talk… …unfortunately

The CONSCS-defining game 1. Formal definition of CONSCS should be mechanically applicable to any entity in a well-defined environment – One should not have to be conscious to test if an entity is a CONSCS – This requirement has a similar (but weak) analogy with CAPTCHAs, which are bots that distinguish humans from bots – We do NOT require that our definition of consciousness be “black-box. ” One may be forced to “open up” an entity

The CONSCS-defining game 1. Formal definition of CONSCS should be mechanically applicable to any entity in a well-defined environment 2. The question whether an entity is CONSCS is a function of its algorithms, not the stuff (silicon or carbon) that implements those algorithms 3. In the real-world environment, people are CONSCS while rocks are not (CONSCS)

Informally: An entity is CONSCS relative to a given environment iff: 1. The entity has a model of its environment, and a model of itself in the environment 2. 2. The entity is motivated/driven towards goals 3. 3. The entity is able to “try on any 'story' for size. ” Story provides strategy for dealing with the world 4. 4. The entity has simple serial circuit — the CONSCS processor — as interface between self and world

1. The entity has a model of its environment and a model of itself in its environment Entity Model of Environment

1. The entity has a model of its environment and a model of itself in its environment A. These models are for making predictions B. The models are derived from experiment C. The models are constantly being revised • A, B, C are done unconsciously/automatically. • Revisions C are called TACTICAL revisions.

2. The entity is designed to achieve certain goals. • Goals are determined by the values of two real variables, “pleasure” and “pain, ” and a “current story (CS). ” • Options for conscious consideration (evaluation by below formula) bubble up to consciousness. • The entity is designed to consider each such option and select the option that maximizes net payoff under CS (current story) = Expected prob distribs inherent in models of environ and self [ pleasurecs(t) – paincs (t) ] dt t 0 = current time

Options bubble up • Options are generated by systems of the (parallel) unconscious brain, which present their few best ideas to consciousness. • Options presented to consciousness are considered for a vote by the entire brain*. – the entity uses its current story/strategy to decide which option to select * a vote taken by the entire brain accounts for the (apparently) serial bottleneck.

Two kinds of options bubble up from the unconscious: Some are stories/strategies, such as: I am a scientist or I am a crook. Some are not stories, including: Get a glass of water or Get a cup of coffee. Go for a swim or Write up your results.

3. The entity has the ability to “try on ANY ‘story’ for size” A story is a theory about the world, such as: . Other entities are built just like me Other entities are mindless robots A story suggests a strategy* for dealing with the world: Treat other entities with respect Treat other entities any way you like ANY ‘story’ means any story/strategy that any entity could use to make decisions We are not claiming that any story will or even could come up, only that if it came up it would be considered.

models/tactics vs stories/strategies • Models are updated unconsciously. We call these tactical updates. • Stories are updated consciously. We call these strategic updates.

4. The entity has a simple serial circuit — the CONSCS processor — as interface between Entity’s internal unconscious (parallel) brain AND Entity’s external (massively parallel) world • CONSCS processor is for choosing a story, and using story to select strategic option to focus on • The options are offered up by the entity’s internal (unconscious parallel) brain • Choice made according to current story/strategy

What humans are conscious of: all options that bubble up to the CONSCS processor (but not lesser options that don’t rise to that level). This includes: • inputs from the external environment, • inputs from the internal environment, • actions that one can perform on the two environments. • pleasure and pain: current values.

Applications Philosophy • Understanding “understanding, ” “selfawareness, ” “free will, ” resolving: “All theory is against the freedom of the will; all experience for it. ” — Dr. Johnson Science • Building good theorem provers (as opposed to proof checkers), and better CAPTCHAs

Bibliography on the Scientific Study of Consciousness • David Chalmers, philosopher logician, has a huge excellent online bibliography • Rodolfo Llinas, neuroscientist, is finding “Neural Correlates of Consciousness” • Bernard Baars has a long history of good writing; his theater metaphor is widely accepted • Daniel Dennett, Antonio Damasi, … • Marvin Minsky never uses “consciousness” in his “emotion machine, ” as it means different things to different people.