Let`s talk about AI

Artificial Intelligence, or AI, is all around us. We use it every day, even without noticing it. AI helps us find things easier, provides content we may be interested in, and helps us work more efficiently. But there is often confusion about what AI is and what it is not, because we do not have a shared understanding. Here is a conversation I often have with colleagues when talking about AI to help them understand how we are building AI solutions in HR at Bayer.

What is AI?

We have heard and even seen self-driving cars, robots, and other things the media touts as “AI.” But there are many more basic uses of AI that we experience every day. For instance, LinkedIn, Twitter, online shopping and newspapers, and many other online platforms are all powered by AI. Said differently, there are systems storing, analyzing and using data to provide relevant content or propose goods and services that may interest us.

When we talk about Computer Science, we can think of programmed algorithms that help us to come to the right conclusion: a certain input is translated into a result, such as a calculator. Next to it is AI, where the amount of data and decisions taken help improve the algorithm; this is what often is called Machine Learning. With the advancements of technology, we also now have “Deep Learning” that can cope and deal with much more complex data models. 

So, what is a concrete example of machine learning application in daily life?

One of the most used examples regarding machine learning is related to shopping. People often buy similar products over time and prefer certain brands. If that data is known and used by retailers, then the system can “learn” from it and propose certain products based on the statistical model “People who bought this, also bought that…”. If consumer preferences change, or you look for alternatives, the system will capture and learn and will change recommendations over time.

What is the relevance of Machine Learning? 

Machine Learning helps us learn from data. BUT, we must have a clear understanding of which kind of problem we want to solve first, because this determines which questions to ask and what data we need. Knowing the problem can also help us to create better insights and make better decisions.

We typically find three different dimensions of machine learning:

• Supervised learning: we use two different data sets. The first data set is clustered and labeled already e.g. this picture shows “A” and the other picture shows something else. The system is “trained” to recognize “A.” We then have a second data set which isn′t labeled but allows us to figure out if the algorithm can label the test data correctly and likewise the training data.
• Unsupervised learning: instead of having clearly labeled data, it is unstructured and the system looks for patterns, structures data and visualizes them.
• Reinforcement learning: is based on the feedback loop, e.g. at a later point in time certain kind of feedback is given into the system and used by the system to improve itself.

But is what we call “artificial intelligence” really “intelligent”?

It sounds like a rather philosophical question, because it is. When we talk about AI powered self-driving cars, it′s rather a combination of “robotics automation” and “machine/deep learning”. Based on data from sensors (e.g. distance, speed, image recognition) the algorithm is firstly trained on certain decisions and then improves by learning from them. The car doesn′t know what is right or wrong, but it is trained. The interesting challenge for this kind of AI is that the system first needs to detect the relevant items (e.g. trees, human, car) and then needs to make a decision (e.g. stop, drive around, etc) in ever changing situations. The challenge with these kinds of algorithms are not the detection, but the decision making based on probability calculations. Most important is to understand the logic behind it e.g. referring to odds, the Bayes formula and likelihood calculation.

I am not going into details here but can strongly recommend the MOOC “Elements of AI” from the University of Helsinki and Reaktor to learn more and try it out. So, while we might observe a certain kind of “intelligent behavior,” it′s our own conclusion that the system has made the right decision.

Conclusion

There are many more methods and definitions of AI, but to keep it simple, let’s think of AI as using “machine learning” to learn from data continuously to help you with gaining insights, making a prediction or support your decision making.

But keep in mind it′s called “machine learning” and not “machine knows it all”. You, as the human, makes the decision (purchase or not / read or not) based on the information presented; your decision continues to train the algorithm. 

If you are interested in learning more, I strongly recommend to check out the free and open online course https://www.elementsofai.com.

Specific credits to the "Elements of AI" Team from the University of Helsinki e.g. Teemu Roos and his team, and the team from Reaktor for helping all of us on this journey. 

Want to read some examples on how Bayer is already using AI:

Innovation to Help Farmers Enhance Their Harvests: https://www.bayer.com/en/crop-science-innovations-data-science.aspx

How Digitalization Supports Science For A Better, Longer Life: https://www.magazine.bayer.com/en/how-digitalization-supports-science-for-a-better,-longer-life.aspx