September is the best month to prepare a nice tomato soup! It will be the tastiest tomato soup you’ll ever experience. And it is so easy to do: about ten tomatoes, a medium onion, a small carrot, a celery stick, a little olive oil, a little sugar and one-liter water. This is a five stars recipe according to those who tried it.

We have all we need: all ingredients and the best method to prepare the soup. Let’s do it. It will be so tasty! After a good hour, the soup is ready, the table is dressed, and we start to eat. But the experience is not that good. Honestly, the soup taste awful. Why? It is a simple recipe, easy to prepare and we did it the same way many times, we have a good kitchen and made all this, like always, with heart. But well, we must admit that it does really taste bad!

What went wrong? Any idea to avoid this next time? Could the issue come from the kitchen? No, all is new and never had any problem. Could it be the method? No, we already made the same recipe many times and it was always delicious. Then maybe one of the ingredient: tomatoes, the onion and the carrot came from the garden and celery, olive oil, sugar and water from the shop nearby. Could have been the celery since we already used the olive oil, the sugar and the water for other things and all was ok. But the celery…

We really hope that you will remember this little story when you will confirm that all data collected and used for your analytics are good data, that you can be sure of their origin, their “pedigree”. If it is not the case, are you sure your analytics will be good? Are you sure you don’t have dark data, duplicates… Are you really sure that when you use the results from those analytics within any decision-making engine that the decision is made in a good way? Are you 100% sure that as you trained your AI with a massive load of data, that the machine really learned from data that is reliable? Will this decision be reused anywhere else? Have you measured the impact? Such mistakes could have killed people (2017, IBM Watson for Oncology, Florida’s Jupiter Hospital)! Because “knowledge” build like this can only lead to inductive reasoning, introducing uncertainties here and there, twisting results and as a consequence are making recommendation that could be wrong. You are sure that you will rely on these recommendations to drive your business? Good luck!

What is your need to use such solutions?

Prediction for prediction, going to the story teller is cheaper.

But that is not all. What was the root cause for this bad recommendation made by Watson? The AI learned from test data and no one checked. The AI learned wrong things, and this can explain the result. But it is not only that. The whole system was built by experts who put all their knowledge to solve the problem for such cancer cases, but they forgot to draw the problem space were many other cases would have appeared. One of them is when a patient has a lung cancer with severe bleeding, administering chemotherapy and the drug Bevacizumab. According to a warning on the medication, it is fatal in case of severe bleeding and it is not like other side effects who could lead to increase the risk, it is fatal.

Maybe the AI was trained using a wrong data set or a test data set but as the AI had to make a decision, this dependency was not checked and this has nothing to do with data. It was neither checked nor available because it was not something to look after. Within your business context, you did it? The biggest Fortune 500 companies did this error and this error will be reproduced many times since their digital transformation methodology is not considering this. The promoted recipe is not taking care of situations where the expert way of doing things cannot be gone, for any reason.

If a problem space would have been used, if the machine would have learned using data for witch a pedigree is available, if the data structure would have been categorized, the resulting AI would never have taken such a decision! At MacAnima, we think that this is the good way forward. It is not a matter of technology, tools or solutions, it is only a matter of conception.