Biological path toward strong AI

The Biological Path Toward STRONG AI

Matt TaylorNumenta

OS Community Manager Principal Engineer

Summary• Intelligence, Weak and Strong

• Some minor brain science

• Hierarchical Temporal Memory (HTM)

• Sparse Distributed Representations

• HTM Algorithms

• Encoding

• Spatial Pooling

• Temporal Memory

– ROBERT MCMILLAN (from a Wired article title, 01.16.15)

“AI Has Arrived”

– me, just now

“Weak AI Has Arrived”

Weak AI ≠

Intelligence

What is Strong AI?

Wikipedia requirements for AGI:

Weak AI won’t produce Intelligence

what’s missing?

Realistic Neurons

Movement

Numenta’s Dual Mission

1. Understand how intelligence works in the mammalian neocortex

2. Create software based upon those principles

To create intelligence, study what we know is

intelligent today

Let’s talk about brains

Let’s talk about brains

Let’s talk about brains

HTM is an evolving theory of intelligence based

upon this neuron model

• capacity • matching

Sparse Distributed Representations

EncodersExample: Date Encoding

• proximal feedforward input activates mini-columns • normalizes sparsity of encoded representation • maintains semantic meaning of representations

Spatial Pooling

• identifies context of input based upon distal segments • works within mini-column cells • puts cells into predictive states for next time step

Temporal Memory

HTM Neuron Model

These are just the foundational components of neocortically-inspired

intelligence

HTM School

It’s all open source

Join our community

numenta.org

Thanks for watching

Matt Taylor [email protected]