Artificial Intelligence

00:06 Artificial intelligence, or "AI" for short, is definitely one of the topics that are currently receiving a lot of media attention, and it has many expectations.

00:16 But let's take a look at where artificial intelligence can be classified in the so-called “hype cycle”.

00:22 The hype cycle divides the process of technology adaptation into five different phases along two dimensions.

00:28 From the horizontal perspective, we see the technological maturity, and on the vertical level, we see the visibility of a technology in the sense of the expectations that are formulated for this technology.

00:39 What do you think? Where is AI in this representation? Are we in the first phase, the triggering of innovation, where we first see examples of using AI that stimulates creativity for future situations? Or at the peak of excessive expectations, where a great number of maybe even unrealistic expectations are seen towards technology? Maybe we have already passed this peak and we are in the so-called “valley of tears”, where we are confronted with unmet expectations regarding technology? Or are we already beyond this step and are on the way to enlightenment towards a productive application of AI in many scenarios? As we will see below, the question cannot be answered clearly.

01:30 This is mainly due to the fact that the definition of AI is not clearly defined.

01:37 When we think of, for example, the area of autonomous driving, we will certainly encounter many applications of AI, which are to be classified at the peak of excessive expectations. At the same time, there are already many productive applications of AI.

01:54 Let's think of automated passport controls at airports, for example, where image processing and image recognition processes are in use.

02:03 To start with AI, a structured database and a large amount of data is necessary.

02:10 You are probably familiar with the very striking formulation that data is the oil of the 21st century.

02:17 For companies, this means that they have to think about how the resource data can be used sensibly and in which areas this resource is already available in sufficient amounts.

02:30 That means there are areas yet to be identified where larger amounts of data of a similar structure are available- where this structure remains relatively constant over time. If data is the oil of the 21st century, we can also say that AI is the electricity of the 21st century.

02:49 Because exactly like electricity is created from the raw material oil, AI feeds off of raw material data.

02:57 And just like oil, the data is refined and used to automate or fuel complex processes.

03:03 This results in strong efficiency gains, such as the automation of decision-making processes that can currently only be carried out by humans.

03:11 And for companies, new business areas are emerging-new opportunities to actively shape the competitive environment in the future.

03:18 Small companies, in particular, also have the opportunity to catch up with large companies and remain competitive in the future.

03:26 If we ask ourselves what exactly is meant by AI, we are faced with a whole range of terminology. Robotics, machine learning, deep learning, natural language processing, and chatbots are just a few of the terms.

03:40 All of these terms have in common that they are sub-areas of AI.

03:44 However, they are not suitable for deriving a definition of the term.

03:48 So let's take a closer look at some uses of AI below.

03:53 One of the first major applications for AI was in the automation of board games.

03:59 In 1997, the IBM chess computer "Deep Blue" managed to defeat the reigning chess world champion Garri Kasparow.

04:07 This success is considered one of the key milestones in AI research.

04:13 A second example came in 2016 when the "Alpha Go" system defeated the active world champion of "Go," the Chinese board game.

04:23 Another example is in the area of personal assistance systems that are controlled using natural language, such as Siri or Alexa, which are part of every smartphone today.

04:34 Similar applications can also be imagined in the corporate context, such as to query sales figures for a product via natural questions like these: "How many model A, B, and C shoes did we sell in Singapore last month?" Or these applications can find contact persons for a specific topic.

04:52 For example: "Who can help me with tax matters?" That could be a question that can be used to query employee databases in the future.

05:00 The two examples show that what is understood by AI depends very much on the point of view and changes over time.

05:07 While many people today primarily think of autonomous driving when it comes to AI, dealing with the problem of "chess" is not the first use that comes to mind today.

05:18 So we can see that what is currently understood by AI is constantly changing over time. To look at this problem from a slightly different perspective, let's take a look at the complexity of the issue and the question behind each of these problems.

05:34 I invite you to consider how difficult or easy it is for you to answer the following four questions. Question 1: Which of your work colleagues is particularly personable? Question 2: Does an invoice contain all legally necessary information so that it is formally correct? Question 3: What is a suitable gift for a close business partner? And Question 4: How do price increases affect total demand- let’s say, an increase of 4% for product A and by 5.6% for product B, and is this demand elastic? If we now consider how difficult or how easy it is for us to answer these questions, then you will probably come to a similar classification as shown here.

06:18 We have problems reaching from simple to difficult on the horizontal level.

06:23 You probably had an easy time determining which colleague had a particularly nice attitude. You simply have an intuitive feeling for it, and you have a very precise idea for the likable manner.

06:37 Question No. 3 -What is a suitable gift for a business partner?- is probably also relatively easy for you to answer.

06:44 You would probably exclude extremely expensive or very special gifts.

06:49 Question No. 2 about the invoice and the necessary information is a little more difficult to answer.

06:55 You are, of course, able to identify this information, but on the one hand, you need to know what information is necessary so that the invoice is formally correct.

07:04 On the other hand, you have to identify this information on the document itself.

07:09 That means you have to have some cognitive performance.

07:12 The fourth question about the price increase and the impact on price elasticity is probably the most difficult one for you to answer because you have to gather different information, structure it, and do calculations.

07:25 Even if the calculation is not too complex, the answer as a whole requires a relatively high effort. If we now look at which question is easy and which question is difficult to answer for a computer, the result is an inverted picture.

07:38 The question of price elasticity is very easy to answer for a computer because it is a formally describable mathematical problem that can be calculated in fractions of a second. The same applies to the identification of invoice information and a comparison against formalities, since this information can also be identified in a fraction of a second using optical image recognition processes.

08:01 The question of the right gift for a business partner, on the other hand, is not that easy to answer for a computer, since the question of what is a good gift and what is a bad gift depends very much on cultural aspects and cannot be clearly said.

08:15 The same naturally also applies to sympathy or antipathy toward a work colleague.

08:20 This depends on the preferences of the answerer and is highly individual.

08:25 A computer, per se, has no opinion of its own or preference toward a particular person.

08:31 An answer to this question will be difficult to determine mechanically.

08:35 If we now swap the scale from simple to difficult for clearly defined to ambiguously defined problems, three major AI processes can be identified in order to address these problems. Traditional calculation methods represent the first class of procedures.

08:52 They are suitable for solving clearly formalized mathematical problems: for the calculation of different scenarios for market developments, for route optimization, for the delivery of packages, and for the efficient design of production plans.

09:05 If we expand these approaches with knowledge-based assumptions, we are able to solve other problems that involve less formalized knowledge.

09:16 So, via a combination of greater amounts of data with different assumptions, for example, for the statistical distribution of certain data, a knowledge-based conclusion can be realized. For example, it is possible to make a diagnosis for diseases from the presence of certain symptoms.

09:37 Another example is the interpretation of large amounts of data and the identification of certain structures in these data collections.

09:44 The so-called learning processes represent an extension of knowledge-based approaches.

09:50 Learning processes are able to independently derive rules and structures from large amounts of data using examples.

09:57 Examples include image recognition, face recognition, or object identification, so the question: "What is on a picture?" or the methods of speech recognition and machine translation are also used in the area of anomaly and pattern recognition in very large amounts of data. If we consider all three methods in combination, we see that a very large area of the problem spectrum can already be covered.

10:22 However, even learning processes cannot address all sub-areas, as there is always a certain, so to speak, intuitive aspect within a specialist domain, which cannot be taken into account by these processes either.

10:36 What could a productive implementation of artificial intelligence in a company look like? To evaluate this, we have to deal with three questions.

10:44 First, what is the objective? Second, what could a concrete use case look like? Third, how does the concrete implementation work? When asked about the goal, we first have to clarify: What is AI supposed to achieve? Meaning: What is the goal of the mission? Is it about automating an existing routine job? Or should the business model be changed sustainably? The question of economic impact is closely linked to this- for example, the impact on an existing business model.

11:14 And finally, the question has to be answered: What is technically possible or can be realistically implemented? Questions arise about the availability of data or the availability of computing power for processing this data.

11:27 Second: The identification of a use case is based on the target definition.

11:32 A case must be selected that can be clearly defined and outlined in terms of content.

11:38 Many companies make the mistake of choosing an overly complex case that might not be suitable to be solved by AI at all.

11:45 So, the first step in the use case definition is to make a realistic estimate of what should be implemented within the scope of the case and what skills are required.

11:54 Technical details play a role in the implementation.

11:57 This part is particularly about the question of data and resources that are required to implement the use case, that is, whether this data already exists or still needs to be collected. Furthermore, aspects of the technical infrastructure play a role, such as the specific project setup with regard to members of the team and a time frame for the implementation. After these fundamental considerations, we will now look at a process model with three steps for introducing AI.

12:27 The first step involves starting a pilot project.

12:30 The main thing here is to gain experience in the use of technology to develop a feeling for what is technically feasible and for how this development can be integrated into your own company. If we take a look at the problem areas of AI that we looked at earlier, it would certainly be a good idea to start with a problem that is clearly defined so as not to choose an overly complex ambiguous problem.

12:54 The main intention of the pilot project is to gain traction and to place the topic of AI in the company. An important aspect in this context is the acceptance of AI within the workforce.

13:12 You should choose a project that clearly shows the added value and has a good chance of becoming a success. The second step is to build an AI core team.

13:21 It is important to use the experience from the pilot project to build a core team for AI that will develop its own skills in this area in the long term.

13:30 While external experts have to be used heavily, especially in the beginning, the generation of long-term, strategic competitive advantages focuses on building one's own competencies. Also within view of compliance, it is not a real option for many companies to permanently purchase external technology competencies, since the complete process sequence from data collection to processing and ultimately use in AI applications must remain in-house.

13:55 The third step is to develop an AI strategy for a company.

13:59 It is important to convert the experience from the pilot project into a long-term strategy, taking into account the skills of the core team in order to clearly define how to deal with AI in the company.

14:13 The main focus here is to address fears and reservations about the new technology and to establish a transparent and responsible approach.

14:23 An essential element of AI strategies is the creation of a strategic competitive advantage. This is where the concept of the Virtuous Circle comes into play.

14:35 The Virtuous Circle consists of three consecutive sub-steps that represent a self-reinforcing feedback loop.

14:42 Starting from an initial product, the continuous use of this product creates a volume of data, the evaluation of which in turn leads to an improvement of the original product. So, the continuous use of the product leads to an increase in quality, which represents an increasingly higher entry hurdle for future market participants, and continuously adapts to changing usage scenarios.

15:09 In summary, it can be said that AI is not a single technology and is not limited to a single application.

15:18 AI has numerous possible uses.

15:22 I invite you to identify a use case in your own company and quickly start implementing AI.