Artificial Intelligence Research

But they represent a new forefront in the field of artificial intelligence. Today’s AI doesn’t try to re-create the brain. Instead, it uses machine learning, massive data sets, sophisticated sensors, and clever algorithms to master discrete tasks.

The theoretical setting of hierarchical Bayesian inference is gaining acceptance as a framework for understanding cortical computation.

Friston recently expanded on this to suggest an inversion method for hierarchical Bayesian dynamic models and to point out that the brain, in principle, has the infrastructure needed to invert hierarchical dynamic models [6].

In a recent review, Hegde and Felleman pointed out that the “Bayesian framework is not yet a neural model. [The Bayesian] framework currently helps explain the computations that underlie various brain functions, but not how the brain implements these computations” [2]. This paper is an attempt to fill this gap by deriving a computational model for cortical circuits based on the mathematics of Bayesian belief propagation in the context of a particular Bayesian framework called Hierarchical Temporal Memory (HTM).

Even if these simple AIs could modify their own utility functions, why would they? Nothing outside the utility function has the power to generate base “motivation”. I’m not sure why this is hard to understand. It’s not going to magically change when AIs become smarter.

An MIT team led by Institute Professor Ann Graybiel has found groups of neurons in the primate brain that code time with extreme precision.

The neurons are located in the prefrontal cortex and the striatum, both of which play important roles in learning, movement and thought control.

The research team trained two macaque monkeys to perform a simple eye-movement task. After receiving the "go" signal, the monkeys were free to perform the task at their own speed. The researchers found neurons that consistently fired at specific times -- 100 milliseconds, 110 milliseconds, 150 milliseconds and so on -- after the "go" signal.

True Artificial Intelligence would bypass problems of biological complexity and ethics, growing up on a substrate ideal for initiating Recursive Self-Improvement. (fully reprogrammable, ultrafast, the AI's "natural habitat".) This Artificial Intelligence would be based upon: 1) our current understanding of the central algorithms of intelligence, 2) our current knowledge of the brain, obtained through high-resolution fMRI and delicate Cognitive Science experiments, and 3) the kind of computing hardware available to AI designers.

Humans cannot conduct any of these enhancements to ourselves; the inherent structure of our biology and the limited level of our current technology makes this impossible.

Recursive Self-Improvement is the ability of a mind to genuinely improve its own intelligence. This might be accomplished through a variety of means; speeding up one's own hardware, redesigning one's own cognitive architecture for optimal intelligence, adding new components into one's own hardware, custom-designing specialized modules for recurrent tasks, and so on.

И вот что интересно: в первых опытах часть животных изучали сразу после визуального стимула, а часть — после того как они провели некоторое время во сне. Реорганизация зрительной коры наблюдалась только у тех животных, которым дали поспать.

Ответом явилась молекула-рецептор N-метил-D-аспартат (NMDAR). Она "следит" за изменениями в межклеточной коммуникации во время бодрствования и включает цепь новых биохимических сигналов во сне.

Another experiment designed to test how well the subjects could control the single neurons was a fade experiment in which the subject was shown a combined image of two faces: Josh Brolin (star of Goonies) and Marilyn Monroe, and told to think of Josh Brolin. The electrodes sent data on the Josh Brolin and Marilyn Monroe neurons to the computer, which brightened the image of the one causing most neuron firing. As the subject thought of Brolin, the image of Monroe faded out.