WHAT DOES IT ALL MEAN?
As with any new area, buzzwords and opinions are abound. Is AI the same as ML, and what is Deep Learning? All are topics existing within the bucket of data science and are about the exploitation of data?
We can describe the most common three terms as follows:
Artificial Intelligence (AI)
The idea of modern artificial intelligence appears to have originated in the 1950s when Alan Turing wondered if a machine can “think”. In the article “Computing Machinery and Intelligence”, Turing studies this dilemma and suggests conducting a test (now called the Turing test) seeking to find if a machine could become “mindful”.
Machine Learning (ML)
In the 1980s, machine learning was developed, notably with the renaissance of connectionism. The computer begins to deduce “rules to follow” by analysing only data. We can see machine learning as a field of study of artificial intelligence which is based on mathematical and statistical approaches to give computers the capacity to “learn” from data.
Deep Learning (DL)
In practical terms, deep learning is just a subset of machine learning, attempting to model with a high level of data abstraction. In other words, it will be learning how your data is composed, for example, in the context of cat images classification: shapes, textures, colours, poses and background. Deep learning models are composed of layers of neurons.
Artificial Intelligence captivates, challenges and perhaps even scares our society due to the extensive use of the term in sci-fi movies. It does, however, deliver benefits and is crucial in sectors such as healthcare, manufacturing, retail, transportation, health, energy, logistics and finance.
During the 1980s and 90s, there were several false starts with companies promising AI and ML-based regression of data. A significant error was the assumption that with enough data, we can perform a regression to predict outside of what the data describes.
This is a total fallacy.
For practical use of AI or ML in our science domains there needs to be an in-depth knowledge and understanding of the area of science or its application. Treating the problem or question purely as a mathematical problem will lead to incorrect and non-reproducible conclusions. Taking a straightforward example, you could take a litre of water and apply heat to it. As heat is added, you could measure the change in temperature from 20°C to 30°C then 40° C. A regression would then predict a linear increase in temperature as the heat increases, but we know that at 100°C the water boils. There is a discontinuity, and the temperature does not increase in the same way as the earlier temperatures.
Whilst this is a crude and straightforward example, it is an excellent example of applying machine learning without having enough data – a flawed model – which makes extrapolation dangerous. This learning is applicable across our more complex domains and illustrates why we need a highly skilled team to make sense of data. The latest innovations are showing there are practical solutions and business-driven reasons to apply these data science techniques. A significant bottleneck now is the number of people who have the range of skills which are necessary. Zifo’s mission today is to provide scalable data science services to meet the needs of our global customers and help them to exploit their “renewable data”.
So – we see that data science is the marriage of different skills. Without the right balance of skills and knowledge, we can reach incorrect decisions and inferences, which we refer to as the “Danger Zone“. Zifo’s advantage is that we have been working in the scientific domain for over ten years so have a depth of knowledge in our company to help avoid the “Danger Zone”.