Fighting with machines …

Every once in a while, my wife finds me talking to myself – at least this is what it appears to be. In reality I am talking to my little friend, my voice-controlled, cloud-based smart speaker, trying to persuade him to play Lianne de Havas, whom he/she/it obviously does not know or expects to be pronounced in a different way. The response “I couldn’t understand that” usually triggers my nerves – ever more intensively the more often I hear it throughout our conversation. So I try different pronunciations – all fails. Then I change my tactics and try to come through the back door: “Do you know Lianne de Havas”? Hit – sunk! My little companion starts telling me about Lianne de Havas. Aha – I think: so why didn’t you play her songs when I asked you to? So now it is clear he/she/it knows her, I jump back in saying: “play songs of Lianne de Havas” and then it works finally. All in all however, it has taken longer than it would have taken to fetch the record, put it on the turntable and start the good old HiFi system …

If your cement plant happens to be a special case just like Lianne de Havas, you may have experienced identical frustration when trying to operate AI-driven systems. Especially running a kiln plant with complex alternative fuel scenarios and the usual proneness to multiple disturbances is an art, which AI-driven optimizers are still struggling with. Not to be misunderstood: machine learning is a powerful tool that can provide excellent performance for pattern recognition of any kind – think of recognition of faces, habits, traveling preferences etc. However, I am sure we are still going to need our human knowledge and expertise to really achieve our goals.

Yours sincerely

Matthias Mersmann