Mission
To Give Human-like Cognition to the World's Robotic Hardware
Robots today cannot pass kindergarten. The Aquarius Cognitive Engine will provide them those skills, and take them beyond.
Embodiment with a Cognitive Engine
Safe AI
Creating Mutual Understanding Between People And Robots
The advent of highly skilled, automated systems is inevitable. What often concerns people is not necessarily the technical capabilities of these systems, but the uncertainty around how we can communicate with them.
What allows us to feel safe, and to be safe, is a mutual understanding of intentions between both the robot systems and ourselves. The only way to accomplish this is to implement a sophisticated cognitive control system in humanoid robots. If we can communicate with these machines (like we might communicate with a friend), and we know that they can empathize with us (like a mother can empathize with her child), then we can continue AI progress in a safe manner.
But how can we implement "empathy" into an AI model? As mentioned above, it needs to have a compartmentalized system that can form complex decisions and plans from their complex experiences. Goals to complete certain tasks may sometimes have to take a backseat to a deeper goal of maintaining social acceptance with the humans they have learned to love. That learned love is something that will be trained in them just as human toddlers learn in the first several years of their life.
Embodiment with Modern AI Alone
Current AI
Current AI
Current AI
Current models need to be trained on massive amounts of data
Current AI
Current AI
Robotic arms, or other applied robotics rely on a team of engineers and consultants to tailor a rigidly trained model to complete a specific task
Current AI
Aquarius Engine
Problem: The learned skills will prove to lack flexibility and reasoning characteristics, leading to machines with severely limited range and a waste of resources
Aquarius Engine
Aquarius Engine
Aquarius Engine
Install a Cognitive Engine that consists of functional modules inspired by research on the human brain. The engine can also integrate advanced AI models and run in series.
Aquarius Engine
Robot learns about objects, concepts, and human communication like a toddler
Aquarius Engine
Solution: Robot learns to engage with people and rest of world like a developed human with self-starter characteristics, leading to complex planning capabilities
Competitive Advantages
Current AI models alone cannot control an embodied machine. The Aquarius model will prove to have a few advantages for robotic applications in the near to distant future over the current approach of training models on large, but rigid datasets.
This cognitive control system will allow robots to predict simple events in the world. Also, equipped with goal-driven curiosity and sociability, they'll be able to ask questions on general tasks, a lot like a young student learning in school.
1) Causal Reasoning Centered on Goals
AI development companies and teams like OpenAI and DeepMind are working on some impressive deep learning models. OpenAI has recently unveiled their ChatGPT model that can mimic human text. They have also created a model trained on 70,000+ hours of MineCraft videos, which allows it to perform complicated sequences of actions in the game. DeepMind (Google), which had created AlphaGo, has also created models that have learned to play video games. There are many other companies that have invested millions of dollars into creating other deep learning models that can operate chatbots, detect human faces or objects in images or videos with object recognition, read medical records/images, create floor plan designs, and many other tailored applications.
However, these models do not learn cause and effect with respect to its own self the way humans do. The Aquarius model learns the world the way a human toddler does (not with millions to billions of input-output machine learning training sets as the other models).
The Aquarius model learns how to form a plan in relation to a final goal by fixating on simple, but more critical pieces of information in its world. It can be adaptive like a human on how a subgoal step might change in order to achieve a higher goal (This functioning is due to both the Abstraction and Flexibility systems mentioned in the Background section). The causal inference skills will be more in line with human goals, and will exhibit characteristics a lot more like how humans behave.
2) Social Intelligence
While there are some models that claim to detect human emotions based on voice output or facial features (HireVue, Emotient, Affectiva), these models do not carry a representation of actual human social goals. These models do not fine tune its own behavioral output as if to try to position itself in a way that it can continue some type of human interaction in order to maintain some relationship the way that humans do.
The Aquarius model has been both hard wired with a low-level social goal hierarchy and designed to be flexible and adaptive to sub-steps in order to stay on track with a particular level of social connection. This functionality allows the model to learn and grow the same way young humans learn and grow.
3) Human Compatibility
As a third advantage, or perhaps more of a summation of the first two properties, the Aquarius technology will prove to be a more effortless and cheaper method for training AI to be incorporated in human society.
Current models developed by groups like OpenAI and DeepMind are not designed to act as a system within a robot body. They are also trained on very specific or rigid aspects of human learning, and are not necessarily truly in line with actual human behavior.
As AI continues to grow, companies will be looking to integrate human-like robots in their daily tasks (be it construction, mining, space exploration, or other services that might require human-level flexibility, mental or physical).
Current models will need teams of AI consultants to understand how to tailor their systems to new human jobs. On the other hand, Aquarius technology will allow an already mechanically-developed robot, equipped with human-like cognition to essentially be employed to any company. The combo will allow this agent to learn a job like a new actual human employee might learn. Exhibiting artificial curiosity and social intelligence, it will be able to ask its own questions to its trainers. This will allow the system to quickly learn the essential tools of the trade, rather than having to invest millions in consulting groups.