WASHINGTON (Circa) -- Recent advances in artificial intelligence have undoubtedly been impressive, but even the remarkable of today's machines have their limits. In order to advance to the next generation of AI, one company has gone back to the basics: mimicking the brain itself.
Cortica is an Israeli company with an AI technology that can learn in real time, and without a human feeding it tons of information. It drastically cuts down the time it takes for a machine to learn, and takes a totally unique approach to AI.
"While deep learning is based on a supervised approach, meaning people have to be in the loop to train the technology, Cortica's technology is fully autonomous and unsupervised," said Igal Raichelgauz, Cortica's CEO and co-founder.
Known as Image2Text, the technology is based on neuroscience research Raichelgauz and his colleagues conducted prior to the company's founding in 2007. After analyzing the cortical neural networks in rat brains, the researchers were able to digitally replicate the patterns they found in an electronic format.
"So the idea was not to copy the biology on a very granular level, but to isolate the computational principles at and build a mathematical model that can reproduce those," explained Raichelgauz.
Animal brains, including those belonging to humans, are able to process massive amounts of data in real time, despite environmental distractions. Cortica's tech essentially does the same thing. Current AI methods like deep learning require a human to input huge quantities of data in order for the machine to learn something. Cortica's tech can digest large quantities of visual data as it comes in, cutting out the human and the extended periods of time deep learning can often require, making it approximately 150 times faster.
There are numerous applications where this kind of technology can be useful, but Cortica's leadership is focused on a few, specific business segments where they see their tech being particularly useful. As the internet of things continues to make our devices "smarter" through data sharing over the internet, it's not hard to see how an AI that can process huge quantities of data as fast as a human could come in handy in several ways.
Autonomous vehicles are one of the most main industries Cortica is currently pursuing. The field is advancing rapidly, and some analysts predict it could one day be a $7 trillion industry. But despite the efforts of several flagship companies, autonomous vehicle technology has a hurdle it has yet to get over.
"You need to be able to let the car understand many scenarios that you haven't necessarily trained the system to cope with, to detect and perceive the situation," said Cortica's automotive head Barak Matzkevich.
Many autonomous cars simply can't do that.
There are six levels in autonomous vehicle technology outlined by the National Highway Traffic Safety Administration, ranging from the driver being in full control at level zero, to the vehicle being in full control at level five. Matzkevich explained that there are several models currently in development at level three and four. These cars are quite capable of handling everyday tasks on the road, but may not be programmed for every random situation they encounter, and will therefore need a human to take over. Reaching level five is tough. Cars will be expected to react just like a driver, even in extreme conditions.
"One of the main problems of [autonomous] cars is not necessarily to detect an object, in terms of distance, speed, and the vector that this object is moving, but to understand the behavior of the object."
It's a high bar that Matzkevich and his colleagues think Cortica's AI is capable of reaching, and they showed us how by taking us on a test drive in one of the most unpredictable driving environments today: downtown Manhattan.
While the car our intrepid photographer rode in was not autonomous, it was fitted with a camera, which was connected to a laptop loaded up with Cortica's software which gave us a taste of how the AI works. After a few moments on the road, the AI started categorizing what it was seeing with various colored blocks. Cars surrounding the vehicle were tagged in green, pedestrians were in red, and traffic lights were in yellow. In an actual autonomous vehicle, Cortica's AI would be using this visual data to avoid that distracted driver on their phone, or hit the brakes when an errant pedestrian crosses in front of you on Wall Street.
Cortica is also focusing on the emergence of smart cities. Many of the world's major urban centers are harnessing technology to improve things like sanitation, traffic, and security. Cortica's technology is capable of capturing both remarkable details and larger trends. When applied to the visual feeds in a city, it can use the same learning process it does with cars to find out when things might be out of place.
"What Cortica brings to the table is the ability to recognize and understand all this big data, and extract insights from the most simple tasks, like recognizing faces on a large scale, to more complex tasks, like recognizing different behaviors, subtle expressions, and predicitng different scenarios," said Raichelgauz.
For example, say a bag was left at a subway station and remains unattended. Cortica's AI could pick up on that bag and theoretically figure out if it was simply left there by a distracted traveler, or for a more nefarious purpose based on what it has learned. It could also help fight more typical crime by analyzing trends and predict when a violent attack might occur, alerting authorities before hand. It could also be scaled up for larger situations, like figuring out when a crowd at a peaceful demonstration might turn violent.
Cortica's technology is already about to be rolled out in India, and more cities could follow.
Of course, it's not hard to see how such a technology could be used for more nefarious purposes.
Julian Sanchez, a senior fellow with CATO Institute who studies security issues, expressed some skepticism regarding AI's ability to effectively prevent crime without some unethical consequences.
"Quite independent of specific concerns about AI, there are reasons one might might be reasonably worried about building those kinds of sensor networks and collecting that volume of data," he said.
Sanchez noted that such an infrastructure could be used by a less than ethical regime to ferret out whistle-blowers, or perhaps find embarrassing information on rivals that could be used in blackmail. But Cortica noted that it is taking privacy concerns into account, and said that its AI doesn't reproduce images it has seen in the past. The company is also exploring a decentralized data model like the block chain used in crypto currencies to ensure openness.
Despite the concerns, the implementation of AI across the spectrum of daily life appears to be more of inevitability than a possibility at this point, so don't be surprised if your car or city is smarter than you one day.
Circa's Jason Zucker contributed to this report.