Autonomous AI

powering the next generation
of intelligent machines

Artificial Intelligence has grown by leaps and bounds over the last 2 decades. Although AI is still in its infancy it has already begun to revolutionize our world. For AI to reach its full potential, machines will need to learn and “think” on their own without human intervention. This phase of intelligence is Autonomous AI.

Presented By
Cortica - Autonomous AI

What is Autonomous AI?

The New Paradigm in Artificial Intelligence

Autonomous AI - Unsupervised, Transparent, Intelligent
Autonomous AI is a new system of computational cognition enabling machines with the ability to truly think.

By reverse engineering the way that the human cortex processes information, Cortica scientists have developed AI that mimics the human brain. The result is technology that is capable of learning on its own in real-time. Autonomous AI is able to interact with the real world and collaborate with other machines.

Autonomous AI is the product of 10 years of Cortica’s research and development. The Israel based tech company has merged the fields of neuroscience and computer science to develop unsupervised learning methodologies, ushering in the next generation of AI to power intelligent machines.
Learn more at cortica.com

AI Evolution

To get a better idea of the future of AI it helps to understand its past.
Here’s a brief look at the key milestones of AI.
  • 1958

    1958- Perceptron Unveiled
    Perceptron was designed as an image recognition machine by researchers working for the US Navy. It quickly proved to be ineffective and its disuse stagnated the field of neural network research for many years after.

  • 1989

    1989- Yan Lecun
    During the late 80’s Lecun was a postdoctoral research associate in Geoffrey Hinton's lab in Toronto. Following his research Lecun released an early image recognition algorithm that started a new paradigm of machine learning.

  • 1991

    1991- Natural Language Processing
    Language-processing tasked historically relied on direct hand coding of rules. The introduction of machine learning calls instead for statistical inference to automatically learn rules by analyzing real world examples.

  • 1997

    1997- IBM Crowned Chess Champion
    In 1997 IBM’s Deep Blue defeated world champion Garry Kasparov in a 6 game series to become the first non-human chess champion in history.

  • 2007

    2007 - Image Net Founded
    ImageNet is a large visual database designed for use as training data for visual object recognition research. To date over 10 million image URLs have been annotated as part of the dataset.

  • 2011

    2011 - Watson Wins Jeopardy!
    IBM’s Watson defeated Ken Jennings, the previous record holder for the longest winning streak on the show, by a commanding deficit to become the show’s first AI contestant and winner.

  • 2016

    2016 – DeepMind Go Champion
    Google’s DeepMind created an AI that defeated South Koreas Lee Sedol in 4 out of 5 games of the series. This marked the first time a machine had defeated a professional player in a full game.

  • 2017

    2017 – Cortica Unsupervised AI
    Cortica unveils its Autonomous AI to power the future generation of autonomous machines. This revolutionary unsupervised AI is the key to unlocking everything from self-driving cars to instant medical diagnosis and smart cities.

AI Evolution

To get a better idea of the future of AI it helps to understand its past.
Here’s a brief look at the key milestones of AI.
  • 1958

    1958- Perceptron Unveiled
    Perceptron was designed as an image recognition machine by researchers working for the US Navy. It quickly proved to be ineffective and its disuse stagnated the field of neural network research for many years after.

  • 1989

    1989- Yan Lecun
    During the late 80’s Lecun was a postdoctoral research associate in Geoffrey Hinton's lab in Toronto. Following his research Lecun released an early image recognition algorithm that started a new paradigm of machine learning.

  • 1991

    1991- Natural Language Processing
    Language-processing tasked historically relied on direct hand coding of rules. The introduction of machine learning calls instead for statistical inference to automatically learn rules by analyzing real world examples.

  • 1997

    1997- IBM Crowned Chess Champion
    In 1997 IBM’s Deep Blue defeated world champion Garry Kasparov in a 6 game series to become the first non-human chess champion in history.

  • 2007

    2007 - Image Net Founded
    ImageNet is a large visual database designed for use as training data for visual object recognition research. To date over 10 million image URLs have been annotated as part of the dataset.

  • 2011

    2011 - Watson Wins Jeopardy!
    IBM’s Watson defeated Ken Jennings, the previous record holder for the longest winning streak on the show, by a commanding deficit to become the show’s first AI contestant and winner.

  • 2016

    2016 – DeepMind Go Champion
    Google’s DeepMind created an AI that defeated South Koreas Lee Sedol in 4 out of 5 games of the series. This marked the first time a machine had defeated a professional player in a full game.

  • 2017

    2017 – Cortica Unsupervised AI
    Cortica unveils its Autonomous AI to power the future generation of autonomous machines. This revolutionary unsupervised AI is the key to unlocking everything from self-driving cars to instant medical diagnosis and smart cities.

Where We Are Today

AI has evolved

Beyond Deep Learning

State-of-the-art Deep Learning systems work by having computer scientists annotate large databases of training information and painstakingly teaching machines to understand the data. This process effectively teaches machines to perform specific tasks but cannot scale to any form of true comprehension at any general level.

Autonomous AI’s ability to comb through data without supervision and look for patterns and commonalities makes it immune to many of the difficulties facing Deep Learning.

The Path To Autonomy

Programming

ProgrammingOriginally pioneered by IBM, Apple, and the like focused feature engineering with a rule-based methodology.

Deep Learning

Deep LearningUtilizes deep architecture and tremendous amounts of structured data to enable machines to complete specific tasks.

Autonomous AI

Autonomous AIThe next generation of artificial intelligence providing machines with the ability to learn in an unsupervised way and truly think.

Deep Learning

Deep Learning is a subset of Machine Learning. It operates on the principle of training machines manually from large datasets and utilizes cascading layers of nonlinear computational processing. The output of one layer becomes the input of the next while features of the data are extracted at each level. This differs largely from the process of the brain, which is made of interconnected neurons working in parallel.

Deep Learning has become the standard type of Artificial Intelligence over the last decade and is excellent at accomplishing predefined tasks based on heavily annotated data. However it lacks any ability to learn on its own. Deep learning works well in heavily rule-based scenarios that require a simple response, but it can only be applied to tasks where enormous amounts of structured, annotated input datasets exists. It fails when applied to spontaneous or general human tasks.

Deep Learning is a supervised processDeep Learning is a supervised process. Programmers design the architecture and feed it structured, annotated data to teach it new concepts. Humans learn on their own. We have experiences that we observe, infer and comprehend simultaneously.

Deep Learning utilizes complex layersThe multi-layered approach of Deep Learning adds complexity that does not always equate to accuracy. These layers can lead to overfitting, a statistical error where the model becomes too complex in an effort to explain the idiosyncrasies of the data. The multitude of layers also creates a system that needs massive amounts of computer power and lacks optimization.

Deep Learning lacks conceptual understandingDeep learning lacks a true conceptual understanding of objects, environments, and concepts as they primary are built for object classification. Humans understand subtle and abstract concepts, context, behaviors and inferences that allow us to make predictions about future events.

Neural networks have millions of connectionsDeep neural networks have millions of connections that help them make decisions. Any slight change can lead to an unexpected outcome with no rationale about the final conclusion. This leads to an inherent bias based on the training data, and makes it nearly impossible to diagnose errors.

Autonomous AI

Autonomous AI takes a very different approach from modern, state-of-the-art technologies. Autonomous AI learns from unstructured, noisy datasets by combing through the information in search of commonalities. In this unsupervised process, the AI clusters and organizes data into concepts. These concepts are given a lightweight structural representation called a signature.

Autonous AI learns without supervisionAutonomous AI does not require scientists to annotate data to manually teach the system. The AI combs through datasets identifying commonalities automatically and creates new concept signatures. This allows the system to garner a true understanding with extensive edge case coverage.

Autonomous AI translates natural signal dataAutonomous AI translates natural signal data (video, sound, radar, and more) into highly compressed searchable and universal structural representations. By translating natural signals into these light computational concepts the system is able to operate and analyze with ease and speed.

Autonomous AI turns data into knowledgeAutonomous AI is able to collect, discern, and organize the represented information into true knowledge. As this system identifies commonalities between data it becomes both more robust and intelligent - identifying links and parallels between different concepts.

Autonomous AI is a universal platformAutonomous AI’s concept signatures are universal and work across devices. With light, shareable computational files the system is able to share information between machines in real-time to improve recognition on the fly.

Autonomous AI is transparent for predictable resultsUnlike the “black box” approach of Deep Learning, Autonomous AI’s results are completely predictable and transparent. With a fundamentally different, and largely flat approach Autonomous AI brings about simple error detection and correction.

One self-driving car produces 4,000GB of data per day
By 2020 security cameras will produce 30 billion images per second
A small fleet of drones can produce 150,000GB of data every day
Powering The Future

Autonomous machines are no longer the dreams of science-fiction writers. In the near future we will have widespread self-driving vehicles, automated security systems, rapid medical diagnoses and intelligent robotics. All of which will rely heavily on the unsupervised learning of Autonomous AI.

Autonomous AI for self-driving carsAutonomous Vehicles
As vehicles move toward real-time autonomy, unsupervised AI is the only solution that can bring about level-5 autonomy. A car cannot be trained to understand and react to all possible scenarios on the road. Autonomous AI is the only methodology to achieve this level.

One self-driving car produces 4,000GB of data per day. Autonomous AI analyzes the data automatically, in real-time to gain a full comprehension of the patterns, changes, concepts and contexts that are necessary for safe vehicle operation in any and all conditions. The AI goes even further to predict outcomes and events to allow for truly intelligent autonomous driving.

Autonomous AI for smart citiesSmart Cities
By 2020, global security cameras will produce 20 terabytes of visual data per second. these images and video are key to solving crimes, reducing traffic and congestion, improving efficiency and creating safer infrastructure. Autonomous AI unlocks the power of that data by organizing it into clear, searchable and actionable insights.

Autonomous AI for medical diagnosisMedical Imaging
Autonomous AI has the crucial ability to comb through massive data sets to find commonalities as well as anomalies. this ability to discover anomalies in the medical sphere provides tremendous impact in everything from large scale cell analysis, to MRI review, andand electronic medical records (EMR).

As technology advances diagnosis accuracy will improve. patients will be able to upload their own data and images taken from mobile devices. In the future patients will be able to perform routine tests, get results and seek treatment while rarely setting foot in a medical facility.

Autonomous AI for roboticsRobotics
The impact of automation is set to be massive–changing everything from manufacturing to media to engineering. This trend toward automation will have tremendous impact on society’s output and may fundamentally change our economy. The key element to true autonomation lies in AI systems that are able to employ unsupervised learning across unstructured data to complete complex series of tasks with consistency.

Cortica - Autonomous AI
Learn more at Cortica.com