In industrial contexts, artificial inelligence in isolation has proven to be an insufficient technology. Pure AI is often unreliable without access to human intelligence.
Complicated industrial environments often require completely different intelligence with the ability to look beyond the obvious.
“The problem with AI is that it works poorly without context. When it encounters reality in the form of complicated and, for the vast majority, incomprehensible industrial processes, it works worse due to too little and too old data, incorrect causal relationships, unstable optimizations and difficult-to-interpret end results,” says Leonard Johard, AIO at Calejo Hybrid Intelligence and the innovator behind the company’s groundbreaking technology.
To get the best out of AI, it therefore needs to be combined with human intelligence and human intervention in the form of human programming.
“Within Calejo, we have therefore developed a unique and patented method that integrates human knowledge and AI into a single system. We combine AI or neural networks with human intelligence in the form of acausal modeling and with scalability with the help of differentiable programming.”
Endless possibilities
In acausal modeling, which is the current standard for physical modeling, physical relationships are separated from computational structures.
“This makes the system scalable for those who want to build, reuse and maintain large models while taking into account the human modeling time required,” says Leonard Johard.
Acausal modeling is currently used by all leading modeling tools, such as Modelica, Dymola, Simcenter Amesim, Dassault CATIA, ModelingToolKit, JuliaSIM and Wolfram System Modeler.
To these two elements, differentiable programming is then added using an algorithm that can train billions of parameters in neural networks.
Ambitious goals
“We can train hybrid simulators and any program code in the same way we would in neural networks. With the help of hybrid modeling, AI gets a context to relate to with a causal relationship and interpretable solutions. The technology places lower demands on data quality and is also 10,000 times faster than regular AI and thus more scalable,” says Leonard Johard.
The technology is verified in a variety of messy industrial areas – paper and pulp production, metallurgical processes, electricity market, hydropower and water supply processes. Right now, market penetrations are being carried out in Smart Water, Smart Energy and Smart Industry in both Sweden and Italy.
“We have the patents and a professional operational team in both Sweden and Italy including business leaders, market experts and AI experts with international experience from global organizations. In the long term, we aim to conquer the entire global industry,” concludes Leonard Johard.
