In 2022 we will celebrate the 10th anniversary of [at] - Alexander Thamm.
In 2012, we were the first consultancy in the German-speaking world to take up the cause of Data & AI. Today, it can be said that artificial intelligence (AI) has the potential to make an important contribution to some of the major economic and social challenges of our time. AI plays a role in the energy transition and in responding to climate change, in autonomous driving, in the detection and treatment of diseases or in pandemic control. AI increases the efficiency of production processes and increases the adaptability of companies to market changes through real-time information as well as predictions.
The economic significance of the technology is increasing rapidly. More than two thirds of German companies now use artificial intelligence and machine learning (ML).
With our #AITOP10 we show you what's hot right now in the field of Data & AI. Our TOP10 lists present podcast highlights, industry-specific AI trends, AI experts, tool recommendations and much more. Here you get a broad cross-section of the Data & AI universe that has been driving us for 10 years now.
Enjoy reading - and feel welcome to add to the list!
Rank 10 - The right team
Building a talented data science team can be costly and time-consuming due to the skills shortage. Small and medium-sized enterprises in particular often do not have the ability to hire data science and data engineering professionals to tackle the use of AI. Without a team with the appropriate training and expertise, companies should not expect to achieve much with AI. The costs and benefits of building internal data science teams need to be analysed or outsourced to external service providers.
Rank 9 - Bias in AI
An AI can only ever be as good as the data it has been trained with. Therefore, high-quality and unbiased data are particularly important. But the reality is different: Data that companies collect every day is often skewed and has no meaning on its own. Often, this data only represents a part of the entire data basis and is influenced by previous user input or individual approaches. This challenge can only be overcome by defining algorithms that detect these problems and make the data less "biased".
Rank 8 - Data protection and security concerns
The most important factor on which all Deep and Machine Learning models are based is the availability of data and resources to train these models. When data is generated from millions of users around the globe, there is a risk that this data will be used for dishonest purposes. Some companies have already begun to innovate around these obstacles. Google, for example, has developed an approach to this problem called 'federated learning': it trains an ML model with personal data, for example from phones, on the device itself so that the data is not sent to the servers. Only the trained model is sent back to the company and so no personal data is stored on the servers.
Rank 7 - Lack of explainability and comprehensibility of AI
In most cases, AI is a "black box". Not even a data scientist can clearly explain why the AI made a certain decision. That is why 'Explainable AI' is a broad field of research with the goal of making the technology more transparent for humans - which will reduce any reservations and concerns about AI in the future. So far, the basis for decision-making can only be explained to a certain extent. But until there is full transparency, we have to accept that AI makes decisions whose solution is not always comprehensible to humans.
Rank 6 - Availability and costs of large computing capacities
The high power consumption of power-hungry algorithms is a key factor that prevents many developers from using AI. Machine learning and deep learning require an ever-increasing number of processing cores and GPUs to work efficiently. There are many areas where Deep Learning can be used to gain valuable insights, but some algorithms require the processing power of a supercomputer. Thanks to the availability of cloud computing and parallel processing, developers can work more effectively on AI applications, but the computing power required comes at a price. Not everyone can afford to process these huge amounts of data and rapidly increasing complexity of algorithms.
Rank 5 - Use of external data
The inclusion of external data is an important part of data analytics prorams when companies are looking for strategic insights outside their organisation. With so much data available, it is difficult for companies to know what kind of external data they are looking for and where to find it. Data marketplaces provide a platform for buying data, but typically do not help buyers understand what kind of data is needed for their use case or problem. For example, it can be difficult to ensure good data quality and understand what impact datasets have on predictive models before buying them.
Rank 4 - Lack of guarantee of success
Another challenge in the implementation and integration of AI: the lack of a guarantee of success. The introduction of AI and the implementation of ML projects in a company always involves a lot of effort. To initiate an AI project, the available data must be evaluated and experimented with. Then, the ML model's chances of success are examined with regard to the desired goal. In some cases, the desired result of the use case cannot be achieved with the available data and further strategies must be examined to solve this problem.
Rank 3 - The data must contain patterns
What if the data does not match? A common problem in practice is that the data does not contain a particular pattern. In some cases, data changes randomly and therefore cannot be profitably predicted or analysed. Then using an ML model with this data does not lead to the desired accuracy. Here, in some cases, the data sources can be further evaluated and possibly further processed, cleaned or replaced. Otherwise, the problem of the use case must be questioned and further specified.
Rank 2 - The more data, the better
Data plays the main role in training ML models. Therefore, it is always helpful to have too much data rather than too little. Because ML models need large amounts of data sets to make meaningful predictions, too small a data set can make the ML model inaccurate or even unusable. Only if the training data sets represent all possible constellations and anomalies can these also be recognised later in the application.
Rank 1 - A unified understanding of AI
There is no single, universally valid definition for AI - everyone associates different ideas with the term. It is precisely this circumstance that can become a real challenge in a project. Therefore, all those involved in the project should be clarified in advance. It must be discussed exactly what artificial intelligence is capable of and what it is not. Myths and misunderstandings must be dispelled. Any concerns and worries of the employees - for example, the fear of losing their job or of critical situations arising from AI predictions - must also be taken seriously. Only in this way is it possible to create a uniform understanding so that all those involved are on board and pulling in the same direction.
These are our top 10 challenges we've experienced implementing AI in over 1300 use cases.
Click here to go to our Use Cases Database.
What challenges have you already experienced in implementing AI?