Cogitai Launches Cloud-Based, Self-Learning AI Platform for Businesses

Credit: iStock

February 12, 2019      

Artificial intelligence startup Cogitai today announced commercial availability of its Continua platform, a software-as-a-service platform for self-learning AI. The company said Continua “has the power to turn any process, system, software bot, or real robot into a self-learning autonomous service to drive actionable business outcomes.

Mark Ring, Cogitai CEO

The company said most AI systems today are trained using data that is “painstakingly labeled by hand,” whereas Continua can enable the systems to learn on their own.

“The biggest opportunities for AI to expand into practical business applications cannot happen until AI systems can learn on their own,” said Mark Ring, CEO of Cogitai. “As successful as AI has been so far, its applicability will accelerate dramatically when it no longer depends on data provided by humans. Cogitai’s continual learning approach is a watershed moment in the industry, providing a general-purpose solution that can learn from its own decision, actions, and experience.”

Reinforcement learning is the key

“Almost all AI that’s been commercialized today has been commercialization of supervised learning and applications of supervised learning,” Ring said. “We see this as an extremely important part of artificial intelligence, and it’s had a huge impact on the marketplace. But we think in retrospect 10 or 20 years from now, we’re going to look back at supervised learning at the first step, the thinnest sliver of what’s going to be offered in artificial intelligence. The rest of it comes when the AI systems can start making decisions all on their own, and then being able to learn from those decisions.”

Ring said he likes to compare supervised learning to the sensory cortex of the brain – it’s just one part of the brain that does sensory processing, but the rest of the picture is missing, including the executive cortex (which makes decisions), the motor cortex (which takes actions), and then learning from experience how to improve the decisions that are made. “This is what is done in reinforcement learning,” Ring said.

Beyond reinforcement learning

He added that Cogitai wants to move beyond reinforcement learning, to what it calls continual learning. “Reinforcement learning is exactly what we need for fixed tasks, for learning how to optimize behavior in a fixed task to solve a task, but that’s not what’s always required in industry,” Ring said. “[Continual learning] is where the machines don’t just learn to solve a task, but learn to develop skills and knowledge. Rather than being finished with what they’ve learned once they’ve learned a task, they just add that to their repertoire, and then they can deploy that as needed in the future.”

As easy as that might sound, Ring said it requires a whole new perspective in how to build AI systems. “It’s not something you can easily just take a reinforcement learning system and collect a library of results and then deploy them,” he said. “It actually requires some fundamental new thinking in terms of AI, and that’s our ambition down the road.”

How the self-learning AI system works

Cogitai said customers can sign up for a 40-day free evaluation period through the company’s website. Once a user gets an account on the system, they can describe what the interaction will look like. Then Continua takes over the control of the underlying system and receives observations in order to provide decision suggestions.

The client then executes those decisions or actions from the cloud platform, and this continues in a loop. While this is happening, the client system can also send reward signals that tells Continua when it has improved, when it’s gotten better, and when it’s not doing as well, in order to improve its behavior.

CogitAI Continua Diagram self-learning AI

Source: Cogitai

“We think industry is just beginning to open its eyes to the possibility of having decision-making be automated, and that’s where we’re targeting,” said Ring.

Customers will be able to choose between having a system where the customer keeps its data, and the learning from that data, private and protected; or an approach that lets companies share the data, models and the learning from the system, in order to create additional applications/tasks from the benefits of sharing that with other system users. The company said it will likely charge less to the customers who choose to share models with others.

“We think in the long term that most people will choose that option,” said Ring, “because that allows their services to improve much faster.”

Getting robots to learn faster

Robotics in manufacturing environments is one of the key markets it wants to address with the Continua platform, the company said. “We have a tremendous amount of robotics expertise in our company, so being able to train a robot for manufacturing tasks or an assembly task can be much more straightforward,” said Dennis Crespo, vice president of marketing and business development at Cogitai. Users can train the robot using traditional methods for data, or you can show the robot what you want it to do through human training of the system. “And then the system will learn it, and then it will figure out a better way as well that the human hadn’t figured out before, based on expanding that training on its own,” Crespo said.

In the automotive space, Crespo said the company has been working with a major manufacturer to use AI to keep tabs on engine management. For example, the AI can take data from sensors for fuel mixtures, battery performance, air quality, etc., and the agent can run in real-time in the car to make decisions to keep the car performing at its peak on a per-second basis, he said.

In order to work in cars, the system is able to be placed on an embedded system and still operate as if it was in the cloud, Crespo added. “Our code is really flexible in that sense,” he said. “We can work in small embedded constrained environments, or larger environments where you have lots of compute as well, so it scales in both directions, which is very nice.”

The company said it sees several other use case examples for self-learning AI, including:

  • Video game development
  • Smart building management
  • Customer-service ‘bots
  • Fitness and wellness apps with AI coaches
  • Software-testing bots that learn to find and fix issues
  • On-board collaboration and real-time energy management for semiconductors
  • Robotics process automation and business process management

In addition to Ring, the Cogitai team includes Peter Stone, president and COO, and author of the first study panel report of the “100 Year Study on AI“; and Peter Wurman, vice president of engineering, who was one of the technical co-founders of Kiva Systems, which used mobile robotics in warehouses and distribution centers before being purchased by Amazon. The company was founded in 2015, and funded by Sony in 2016, which remains a key supporter of the company, Cogitai said.

“We have seen the team make tremendous progress in bringing their unique technology to market,” said Toshimoto Mitomo, an executive at Sony’s startup and acceleration division. “We believe self-learning AI as a SaaS platform will be a game changer, and we will see many new opportunities to implement AI in a wide range of innovative business applications.”