Meet Cortica: An Israeli AI Company That’s Teaching Machines To Observe And Reason, Like Humans Do
The human brain processes all information via electrical impulses. You knew that, right? Well, that is exactly what inspired Igal Raichelgauz, CEO of Cortica, an Israel-based Artificial Intelligence startup. He saw the human brain as an electrical circuit and set out to replicate that circuitry to create an AI-based capability that would endow machines with a similar skill set.
Cortica wanted their AI to have a sight sense on par with that of humans.
And we do indeed have an astonishingly complex sight system — everything you see with your eyes, open receptors in your eyes convert to electrical signals. All that information is transferred by those signals, to a part of your brain which sorts and analyzes the color, depth, shape, and size of all those objects. This data is then received by the cortex — the part that most interests Cortica.
Remember poststructuralism? For those of you who need help with that preface, you only know a table as a table because you see it in relation to a chair. If the chair didn’t exist, how would you know what a table is, what it’s used for?
Something similar happens in your visual cortex. It classifies all the objects you see into different categories by assessing them in relation to all the objects you’ve ever come across.
That’s how you know what you just saw was a bird, or a bottle, or your friend, or anything else.
Sure, you know how little time it takes for our brain to perform the entire process since you experience it every waking moment of your life, but have you ever stopped to wonder, to revel or to acknowledge the sheer speed and processing power behind it?
You know what you saw the moment you saw it. Cortica believes it has reverse engineered this process, replicated the biological visual cortex of humans.
Guess how they achieved that?
They worked on a piece of rat brain, a piece that is still living. Yup, you read that right!
The brain gave them access to the electrical interface of all the neurons contained in that tissue. They were able to understand the input-output process of the neurons. They discovered that with some modifications, a neural network could create a “conceptual signature” — without any prior training. It would be able to recognize similar objects, and differentiate them from others.
Such an AI would be able to learn by itself, much like babies do — by observation and reasoning. While we observe and learn from the world around us, it would do the same from the data available on the web.
This is Cortica’s own, unique approach to what is called ‘unsupervised learning’ within the field of artificial intelligence.
Just so you’re on the same page, there are 3 kinds of machine learning — supervised, unsupervised and semi-supervised.
Supervised learning is when you teach the AI from a pre-determined data set, so you already know the output. This is the most commonly used one.
Unsupervised learning is when you give the AI no prior training, and you tell it to solve the problem with only the necessary input. The output from such an algorithm is unknown. For instance, you want your AI to categorize certain geometrical shapes into matching groups.
If you’re using supervised learning, you would have taught the AI about circles, squares, hexagons etc. before giving it the problem. In unsupervised learning, however, you would teach your AI nothing before asking it to solve the problem. It would see the various shapes, categorize them based on similarity, and give its own label to them. This is a process much more difficult to teach an AI.
Semi-supervised learning falls between these two. The AI would have an incomplete set of reference data, and it would hazard the best possible guess based on the limited data it has, and it’s own abilities to extrapolate the data.
Now do you see the ramifications of what Cortica has achieved? Two words — it’s huge!
But Cortica isn’t completely done yet. There’s still time before the technology enters the consumer industry, but Cortica claims to have created an AI that can see and process information like humans.
So many possibilities!
Self-driving cars have already entered the marketplace. But imagine if they could actually recognize and understand what an object or obstacle ahead of them is. The car would stop by itself if it sees a pedestrian crossing the road, thus preventing many road accidents.
The might be able to recognise accidents on the road and could call for help independently.
Your smart home gadgets would revert to the settings that are specific to you when they see you approaching. An air conditioner could increase the temperature if it sees a child in the room, so they don’t get cold. The refrigerator could detect what groceries are finished up and remind you to get more.
Amazon’s grocery store in Seattle is already automated, but what if it could actually see you? That would even remove the need to even scan the app at the entrance. You could just walk right in and it would recognize you from its database, and be able to process you, and your purchases independently and accurately!
The possibilities are truly endless.
Other AI startups such as DeepMind, RealFace, and Genee have been acquired by Google, Apple, and Microsoft respectively. Would Cortica too become a target to be acquired, or would it be able to hold its own against them? Its technology certainly looks powerful enough.
The world is changing, friends. Get ready to see it differently, soon.
Originally published at Chip-Monks.