The Paperless Lab Academy®2023 Europe agenda is rich with hot and trending topics for the lab sector. On stage will be prominent keynote speakers who will come to the Congress to share their experience and knowledge on specific topics.

A major contribution to Agenda 2023 will be a presentation we are all eagerly awaiting: "Artificial Intelligence in Action in the GXP Environment". The presentation will be given by Toni Manzano, Chief Scientific Officer at Aizon.

I had the pleasure and privilege of interviewing Toni to get some insights into his presentation and to start the discussion before he covers the topic on stage at the Paperless Lab Academy®

What do you think has changed technologically or conceptually in the last 10 years to get to the artificial intelligence we have today? 

From today's perspective, I would say that not too much has changed technologically since the disruption. 10 years ago, there was a lot of talk about Big Data, not really about AI. Even though the term artificial intelligence was coined as early as 1950 ^(1).

Technologically, all the tools we need for AI were created 10 years ago with the cloud, Big Data, and computation.  Those 3 concepts, 3 ingredients made it possible for AI.

It was necessary to have data of all kinds and to break down data silos that prevented the information from which knowledge is created from being brought together. The cloud makes it possible to democratise global access to data. It overcomes the barriers on premises and enables logistical distribution of data around the world by overcoming VPN and all technological barriers.

in my opinion, these are the 3 ingredients that make artificial intelligence possible. In fact, AI has already had 2 winters. When the scientific community realised it needed more space and did not find it, and when it realised it needed more computing power and did not have it. That has now been resolved.

From 10 years ago up to now, technology has not evolved, it has not changed as drastically as it did with the advent of the cloud.

The term artificial intelligence or anything to do with digital twins, augmented robotisation, augmented knowledge are concepts that are part of this popular digitised culture. Today, nobody leaves home without asking google Maps or similar app: When is the next train coming? No one looking for a picture looks for the photos one by one, we use search tool.

As a society, we are maturing in terms of digitalisation, and we are maturing faster than the pharmaceutical industry. But if you look at the projects, we have done at Aizon. Companies have a need that can only be solved if you deal with the complexity and variability of the information; realistically, you should not try to simplify, you should accept it. In this case, you cannot work with classical statistics because it is very complex, so you have to work with artificial intelligence. In this transition, digital maturity has emerged but in the pharmaceutical industry we are still far from the maturity that society has today.

As we have all matured digitally as a society, as you say, we use digital information every day, perhaps even unconsciously or automatically. My phone connects to my car and immediately tells me how long it will take to get home. What do you think are the obstacles that hinder or slow down the digital maturity process in the pharmaceutical industry?

In a recent study of the average age of CEOs, boards of the twenty largest companies, the average age is 57.

I am 51 and have long considered myself a digital person, yet already I struggle to keep up.

When there is no urgency to be optimal in a company that has classically always lagged behind technology, it turns out that there is no need. These people do not have that need either, they adopt the part they can from a social and personal point of view, but without designing a strategy to bring it into everyday industrial life.

A strong message to the boards! Toni, help us better understand AI then.  Artificial intelligence is about the data, but what about the algorithm you need to design for every specific process?

Do you know that there are 75 new patented artificial intelligence algorithms every day? You cannot imagine how many unpatented algorithms there are every day. let me explain you why there are so many.
Imagine a production line, a 6000-litre fermenter. It turns out that you need a different model for each stage of the fermentation. A model being the combination of data and algorithm. A single product in different formats, 20 models for 24 hours of fermentation Imagine that multiplied by all the bioreactors for all the lines. Also, every bioreactor needs a different model because every bioreactor is different.
AI cannot tell you why, but it can tell you what is happening.

In other words, the new challenge is how to manage AI, am I understanding right?

Indeed. There are two types of data: the historical data used to build the model, and the real-time data used to feed the model that outputs the real data value. Then comes the accuracy between the reality and the prediction to know if your model is good or even deteriorating.

Who is behind all this work to make it happen?

It is a combination of data scientists and subject matter experts in the process. They work to identify the algorithm that will provide the best approximation. The algorithm is fed with data, and this is how the model is then created. The model that represents reality is ultimately a statistical mathematical model that, after review by the subject matter expert, can be deemed ready for production

And then there's the compliance part. I mean, is this all approved?

We are part of a committee coordinated by the AFDO, Association of Food And Drug Officials ⁽²⁾, which includes the FDA, which has the greatest interest in this development.

In fact, the FDA is already building artificial intelligence into its own processes for two reasons: first, of course, for the patients, to get a safer, higher quality and more efficient product; and second, because the agency itself needs to coordinate better to cover the entire pharmaceutical industry, which extends to outsourced CMOs. They need to automate their monitoring and control process.

Your talk is entitled "Artificial Intelligence in Action in the GXP Environment". Will you come on stage with a use case and explain to the audience what is happening today in terms of compliance and the FDA situation?

I will definitely bring in use cases, otherwise it is too abstract. I think it is necessary to present some use cases so that people can see what kind of applications are running with AI today. The FDA question is crucial, and the audience needs to understand that there is explicit support.

Another point is that many people of the audience do not know where to start. the message will be that without the right data, without reliable data, you cannot run AI. AI is pure statistics, nothing more than statistics, and you cannot trust a result whose final statistics are based on unreliable data.

Pure statistics? I suspect it's more than that, given the complexity and multivariable you explained earlier. How do you start?

The first step we take when we are sure of the data is to find out which of the hundreds of variables really explain the problem.

How can you be sure of the data? How can you check its quality?

An excellent question. Data quality can only be validated if you have a lot of data. That is the first premise. Only when you have a good amount of information can you determine which of that information is valid and which is not. With the help of algorithms, you can identify the outliers on the one hand and the truly representative samples on the other. So the data cleaning starts automatically. This only happens when you have a large amount of information.

Thank you, Toni, for the valuable messages you have conveyed in this interview. We look forward to meeting you in April at the 10th edition of the Paperless Lab Academy ® and diving deep into the topic of "AI in GXP environments, compliance and the FDA approach to AI".

Isabel Munoz Willery, owner and organiser of Paperless Lab Academy ^® 

**************

1.By the 1950s, a generation of scientists, mathematicians, and philosophers with the concept of artificial intelligence (or AI) culturally assimilated in their minds. One such person was Alan Turing, a young British polymath who explored the mathematical possibility of artificial intelligence. Turing suggested that humans use available information as well as reason to solve problems and make decisions, so why can’t machines do the same thing? This was the logical framework of his 1950 paper, Computing Machinery and Intelligence in which he discussed how to build intelligent machines and how to test their intelligence.

2.About The Association Of Food And Drug Officials (AFDO). The Association of Food and Drug Officials (AFDO) is a well-recognized national organization that represents state, territorial, and local regulatory. The Association’s principal purpose is to act as the leader and a resource to state, territorial, and local regulatory agencies in developing strategies to resolve and promote public health and consumer protection related to the regulation of food, medical products, and cosmetics. afdo.org