More than 10 algorithms compete. The best answer wins.
There is no such thing as “the best approach” when it comes to building an AI system. ClarioEx runs multiple machine learning algorithms against your problem and gives you a result with well balanced accuracy and explainability.
An answer is not enough. Management needs the “why”.
Most of the AI systems used in industries are unable to explain “why” a particular answer is proposed by the system. In many cases, “explainability” of the systems is crucial. In a large number of situations, management may not be comfortable making a decision that follows what AI proposes simply because the system says so. Evidence, or supporting and similar experiences, has to be shown to management to support the claim or proposed solution.
In one of our experiences, unlike a traditional scoring system, loan approval officers are much more comfortable with a system’s proposed solution if the system can also provide the reason why it proposes to approve or reject a loan application. The supporting reason in this case can be statistics of historical cases that previously occurred.
By combining a machine learning approach called “Case-Based Reasoning” with statistical methods, the historical cases are retrieved based on similar cases that have been found. The result is an explainable AI system that can propose its suggestion with the supporting reason.
Explainability is now a business requirement
It is not only our view. Leading research institutes, analysts and technology companies report the same thing: businesses want AI they can understand, and most are not there yet.
of IT professionals say that being able to explain how their AI arrived at a decision is important to their business.
40% of respondents named explainability as a key risk in adopting generative AI. Only 17% said they were working to mitigate it.
AI companies grew less transparent. Their average score on the Foundation Model Transparency Index (out of 100) fell from 58 to 40 in one year.
Gartner predicted that organisations which operationalise AI transparency, trust and security would see a 50% improvement in adoption, business goals and user acceptance by 2026.
Two roads to an explanation
Across the research there are two ways to make AI explainable. Each has strong supporters, and each has a known weakness. This is the same trade-off we describe above, and the reason ClarioEx is a hybrid.
Train the most accurate model, then use a second technique to estimate which inputs drove each result. Methods such as SHAP, LIME and integrated gradients work this way. IBM and McKinsey describe them, Google and Meta ship them in their tools, and an MIT team has built a system that turns their output into plain-language text.
Known weakness: the explanation is an approximation. Prof. Cynthia Rudin (Duke University) argues in Nature Machine Intelligence that such explanations may not be faithful to what the model really computed, and that high-stakes decisions should use models that are interpretable in the first place.
Let the system reason in a way people already understand, so the reasoning is the explanation. Google’s Vertex Explainable AI offers “example-based explanations” that search for the nearest, most similar examples to an input. This is the same principle as Case-Based Reasoning, which we have used for many years.
Known weakness: it depends on a quality set of past cases, and it has not always matched the accuracy of model-based methods on every problem.
Where ClarioEx stands. The U.S. research agency DARPA set the goal for its Explainable AI programme as producing more explainable models while maintaining a high level of learning performance. That is the balance a hybrid engine is built for: model-based algorithms compete for accuracy, case-based reasoning supplies the supporting cases, and you get both in one answer.
The four principles of explainable AI
The U.S. National Institute of Standards and Technology (NIST) defines four principles that an explainable AI system should follow. Below each one is how we approach it in our work.
Explanation
“A system delivers or contains accompanying evidence or reason(s) for outputs and/or processes.”
Our approachEach suggestion is delivered with its supporting reason, such as the statistics of similar historical cases.
Meaningful
“A system provides explanations that are understandable to the intended consumer(s).”
Our approachExplanations are past cases from the business itself, which officers and managers already know how to read.
Explanation accuracy
“An explanation correctly reflects the reason for generating the output and/or accurately reflects the system’s process.”
Our approachThe cases shown are the experiences the system referred to when it derived the new solution.
Knowledge limits
“A system only operates under conditions for which it was designed and when it reaches sufficient confidence in its output.”
Our approachWe are frank about limits. If your problem cannot be dealt with using AI, we will tell you.
Regulation is moving the same way. The EU AI Act (Article 86) gives a person affected by a decision based on a high-risk AI system the right to obtain clear and meaningful explanations from the organisation that deployed it. The Act’s list of high-risk uses includes evaluating the creditworthiness of individuals. In Stanford’s 2026 AI Index, 43% of surveyed organisations said the EU AI Act already influences their AI decisions.
Sources
- IBM, What is explainable AI? and Global AI Adoption Index 2021
- McKinsey & Company, Building AI trust: The key role of explainability
- Stanford HAI, AI Index Report 2026, Responsible AI and Transparency in AI is on the decline
- Gartner, CISOs need to champion AI TRiSM to improve AI results (27 September 2023)
- NIST, Four Principles of Explainable Artificial Intelligence (NISTIR 8312, 2021)
- Google Cloud, Introduction to Vertex Explainable AI
- Meta, Captum: model interpretability for PyTorch
- MIT News, Enabling AI to explain its predictions in plain language (2024)
- C. Rudin, Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead (2019)
- DARPA, Explainable Artificial Intelligence (XAI) programme
- European Union, AI Act, Article 86: Right to explanation of individual decision-making
Two kinds of machine learning. We combine the goodness of both worlds.
In general, there are 2 types of machine learning algorithms that can be used to create an AI system. Each has its own strengths and weaknesses.
Approach used by most AI systems
“Analyse data to get a model, where the model is used to propose a solution to the problem at hand.” It focuses on producing highly accurate solutions for its users. Most AI systems today use this type, including the most popular algorithm, the Neural Network (NN).
- Highly accurate solutions
- Cannot really explain why a certain solution is proposed
Explainable AI systems
These algorithms “do not create any model; they propose a solution by reasoning with the data every time they are given a problem to solve.” Humans make decisions mostly by referring to their experiences, and can explain a decision by referring back to them.
- Can explain why a certain decision is made
- Needs a very good set of well-analysed experiences to back it up
ClarioEx (Ditto): Hybrid AI Engine
Our approach at Innocop is quite simple: just combine the goodness of both worlds. Since both types of machine learning algorithms have their own strengths and weaknesses, we combine both types when we design or customize an AI system for a specific business question.
However, unlike most of the hybrid approaches used by others, even for the same type of algorithm, ClarioEx allows you to get a number of algorithms to compete with each other for the best result. The final result is an engine that can provide users with well balanced accuracy and explainability for their problems.
Best machine learning algorithm?
We are AI researchers who have been doing research on machine learning and AI related issues for more than 20 years. There is no such thing as “the best approach” when it comes to building an AI system. There are hundreds of machine learning algorithms that can be used in building systems, and every single one of them has its own strengths and weaknesses. Even in the same domain, small changes in the data can easily lead to inaccurate results proposed by the AI system. So, why not use multiple approaches in building a system?
We do not believe in a single machine learning algorithm
ClarioEx, an enhanced version of Ditto, our hybrid AI engine, is a highly customizable AI engine that contains more than 10 machine learning algorithms to choose from. Unlike other tools for AI systems development, our engine can also be customized to run multiple algorithms in order to automatically find the best answer for you.
Why explanation matters to business
Despite its success in a number of areas, in many real-world cases NN alone cannot really solve a problem, especially when it comes to making a system that can explain why a certain solution is proposed. Without explanation, for businesses, what is proposed by AI systems (or even by humans) is useless, because it will be considered a guess. This is one of the most talked-about issues, discussed widely in NN communities around the world.
Why does our engine have various algorithms?
AI is not new. There are hundreds of algorithms available for dealing with different problems and conditions. Even within the same type of algorithm, every one of them has its own strengths and weaknesses. As a result, ClarioEx contains a number of machine learning algorithms that can be customized to race against each other in order to get the best solution for its users.
On explainable systems: it is not easy to find a quality set of experiences from the data of day-to-day business operations. A number of large AI systems of this kind had to be shut down because such experiences could not be found for them. This is the weakness the hybrid approach is designed to cover.
All problems are unique. And AI cannot solve every one of them.
All problems are unique. There is no such thing as using a single approach as a silver bullet to solve all kinds of customer problems. In fact, we do not believe that such an approach exists anywhere in the world. Most of the problems to be solved in business are unique. Each of the issues requires in-depth analysis before we design and customize a solution for the problem.
AI cannot solve every problem. No matter what your problem is, with our researchers and world-class advisory team, we can analyse and discuss with you the most suitable solution for your business. However, since we are scientists, not just someone who only wants to sell you solutions, in some cases, if your problems cannot be dealt with using AI, we will also inform you rather than wasting your time and resources.
Worldwide, only around 25–40% of AI projects can really answer business problems.
Your assistance on “Making use of AI”
While most businesses target their next move with AI as part of their engine, we, a group of AI researchers with more than 20 years of experience, can help you identify whether what you would like to do is achievable using AI technologies.
Unlike many others, honesty is our key value. We may even tell you: “No, AI cannot help you in solving your problems.”
What if AI is really your viable answer? Since we are researchers, again unlike those who have their predefined solutions, you can rely on us to find you the best solution that AI can offer.
So, why wait? Give us a call. Let’s discuss the wishes and problems you have in mind, and see what we can work on together.
Where our systems are already at work
AI for Loan Approval
A system that proposes to approve or reject a loan application and provides the reason why, using statistics of similar historical cases, so loan approval officers are comfortable with the suggestion.
AI for Fraud Detection
Every organization detects transaction fraud with a different way of thinking. We design the detection around how your business works, not around a predefined solution.
AI for Auto Update Enterprise Architecture
A customized system that keeps an organization’s enterprise architecture updated automatically.
Who we are
We are a group of AI researchers who have been doing research and developing customized AI systems for more than 20 years.
Unlike most AI practitioners today, who focus mostly on using existing tools to develop their systems, with our experience in AI research and development we are able to develop customized solutions (including inventing new algorithms) according to the detailed requirements of each business.
In other words, when designing an AI system, for us, no predefined solution is set. Based on the business’s requirements, the strengths and weaknesses of all possible machine learning algorithms are taken into consideration.
AI Winter?
We have already been through 2 “AI Winters”. Be careful of the 3rd one!
AI is a field that has recently received a lot of attention from businesses. Lots of claims, promises and resources have been put into the field. Unfortunately, based on our experience, the road for AI has never been a smooth one. Therefore, businesses who would like to utilize AI technologies have to be really cautious, especially about what they wish for.
Hype is common for most newly introduced technologies. The term “AI Winter” was coined to refer to a situation where interest (particularly funding) drops due to the disappointment of businesses. There have already been 2 AI Winters in the past, where the 2nd one occurred in the late 1990’s. The setback occurred because of over-claims made by a number of scientists in the community, which led businesses to unrealistic expectations of AI technologies.
Just like every technology that exists today, there is no such thing as a silver bullet that can deal with every problem. With the high expectations of businesses today, what a number of AI scientists worry about is that the 3rd AI Winter could be just around the corner, if we keep over-claiming what can be done by AI systems.
What we believe
Every business problem is unique
Regardless of the type of technologies used in businesses, the main goal of spending on technologies is to “solve a certain problem” for the business. Since each business can consist of a large number of different components, structures and constraints, in most cases each of these problems is unique.
To our knowledge, for example, there is no single or universally applied “Loan Approval Process” used in banks, every bank detects “Transaction Fraud” using a different way of thinking, and every organization we have met treats issues with their mobile applications differently. Consequently, each of these problems requires a unique solution to deal with it.
There is NO such thing as a “Silver Bullet” in AI
Depending on its complexity in solving a problem, each AI system consists of one or more components. Examples of these components include machine learning algorithms, knowledge representations, and reasoning engines. Since the field is not new, there are hundreds of these different components in the field. Each one of them has its own strengths and weaknesses. There is no such thing as a silver bullet approach that can be applied in any domain or solve any problem.
Regardless of various claims in the AI community, developing an AI system is not just drag-and-drop from a tool. News about AI project failures, mostly because the systems do not meet the specific requirements of the business, can be found every day. For us, every problem has to be truly analysed. Then, specialists in all related fields are required in order to understand the problem and come up with a design. Finally, the solution has to be uniquely customized by putting together the most appropriate components in developing an AI system.
So, why wait? Give us a call.
Let’s discuss the wishes and problems you have in mind, and see what we can work on together.