Create your own awesome maps

Even on the go

with our free apps for iPhone, iPad and Android

Get Started

Already have an account?
Log In

FFAI Research Agenda by Mind Map: FFAI Research Agenda
0.0 stars - reviews range from 0 to 5

FFAI Research Agenda

Turing Test

Natural language understanding, social computing and interaction as AI research topics.

Current AIs fail at "internal world model problems", metaphor.

Understanding metaphors and answering questions about mentally constructed physical models should be key problems in AI.

Self

Self, consciousness are important aspects of intelligence.

Psychology, neurobiology, and philosophy give us some clues about what "self" is and isn't.

How can we translate this into AI architectures?

Dualism

Dualism is a common (implicit) objection to AI; most AI researchers resolve it via functionalism.

The notion is analogous to classical (and erroneous) notions about organic matter and "life".

Dualism maps nicely onto the hardware/software distinction; many "properties of mind" are also properties of software.

Determinism and Free Will

Classical "free will" arguments are irrelevant to AI; the fact that computations are deterministic has no obvious consequences for the ability of computational systems to solve problems.

"Free will" probably represents a "theory of mind" and is important for multi-agent AI and social computing

Theory of mind and understanding of other agents' intentions and internal states is an important open question in AI.

Simulation

Simulated realities and AI are closely related issues.

If strong AI is true, AIs will likely exist in simulated realities.

Simulation is important both in the creation of AI (testing, learning, artificial evolution), and for internal modeling of the world.

Approaches to AI Research

Engineering Approach

subdivide the problem into subproblems

understand and solve each subproblem

assemble the solutions into an overall solution

assumptions, the details of hardware or implementation don't matter, intelligence can be decomposed in this way and consists of a large number of specialized solutions to identifiable problems

Neuromorphic Approach

understand the low level architecture of the brain

simulate it and train it, just like a real brain

assumptions, architectural details may matter, intelligence consists of a few general purpose modules and is an emergent property

Artificial Evolution Approach

create a simulated artificial environment

populate it with simulated agents capable of mutation, replication, and competition

select for desired traits

assumptions, architectural details don't matter, intelligence consists of a few general purpose modules and is an emergent property

Uploading Approach

determine the detailed architecture and structure of human brains down to the level of individual neurons

simulate this structure to the necessary level of detail

assumptions, intelligence and memory are relatively robust property and are encoded in connections and channel densities in the brain, chaos and quantum effects do not matter to intelligent behavior

upcoming topics

brain machine interfaces

decoding the nervous system and uploading

digital physics and hypercomputation

brain simulators

nanotechnology

3D fabrication and futurism

next semesters

physical self replication, 3D printing

intelligent agent simulations