Skellig blog AI machine learning

Preparing for AI and Machine Learning

As Henry Ford said, “If I’d asked customers what they wanted they would have said a faster horse.” Something similar is brewing in terms of machine learning in our industry.

Machine learning and AI could be great for pharma and biotech but our technology choices over the past 30 years are a bit of a problem. In industrial pharma automation, there was no apple computer to make some tough decisions on behalf of the customers. Instead, we have all sorts of half-decent “I want!” technology choices.

Back to Monopolies?

The big suppliers have taken the “give them what they ask for” approach too far. Clients are presented with a plethora of similar spec options that still need to be custom integrated. We start projects in a way too open-ended manner.

The issue here is the platforms have failed this industry in making hard calls. Nobody wants a faster horse.

Instead of focusing on being flexible enough to support every field bus ever invented, we should pick a bus technology and focus on making it really great.

It’s 2018 and we don’t have plug-and-play figured out. We can’t unplug a piece of equipment easily and plug it back in without a custom integrated, “kinda works” solution. Plug-and-play was figured out by the PC industry in the mid ’90s.

Our industry is led by big suppliers with a disease. This is the disease of give the customer more choice. It’s not working. We need focus. We need better choices, not more choices.

There are rumors the big guys are working on plug-and-play. But it goes to show how far behind we really are. Are we going to see an ad from one of the big guys in 2019 announcing a plug-and-play solution and have any reaction other than, “This should have been delivered 20 years ago”?

Moving Ahead with Machine Learning

We have widely adopted virtualization, so there is hope for us yet.

Now we see that the next big thing is machine learning and AI. Its promise is huge. However, we can’t take advantage of this until we decide on a few things. We don’t want to start integration projects with a blank slate. Instead of class-based control modules for devices, we need class-based equipment like bioreactors. We need these standard designs so that machine learning and AI can look at a wider dataset and learn. It can’t be applied in a useful way when everything is designed so custom and complex.

The big platform companies should have simplified their offerings before now. They haven’t. Now the ultimate power is with the customer. When they say “Enough is enough, we don’t want your platform only, instead we want your solution”.

The solution is more standard, less custom. Then machine learning and AI can compare bioreactor to bioreactor and we can see useful patterns. Hopefully we won’t have to wait till 2048 to see this happen.