Artificial intelligence (AI) is a branch of computer science and is based on highly intelligent algorithms. The goal is to enable machines to learn, understand and act from experience. The cornerstone for artificial intelligence was laid by the British logician, mathematician, cryptanalyst and computer scientist Alan Mathison Turing as early as 1936. With his Turing machine, a computation model, he proved that machines can execute cognitive processes if they can be broken down into several individual steps and represented in the form of algorithms. Today, artificial intelligence is divided into several fields. The term itself goes back to U.S. computer scientist John McCarthy, who first used it at the 1956 Dartmouth Conference under the heading Dartmouth Summer Research Project on Artificial Intelligence. The Dartmouth Conference also marked the birth of artificial intelligence as an academic discipline.
Brunel gives you the right experts in the field of artificial intelligence
Are you planning an AI project and still looking for the right experts? Our highly qualified specialists from the fields of computer science, mathematics, physics, statistics, data management and electrical engineering can take your next artificial intelligence project to the next level. You might also be interested in our range of services in the areas of temporary employment, work contracts and service contracts.
Artificial intelligence is finding its way into more and more industries. Why? Because huge volumes of data can hardly be handled manually in the context of digitalization. Yet harnessing precisely this data can be instrumental in optimizing products and designing more efficient processes. But how does artificial intelligence actually work and how can it be put to profitable use? The article below provides answers to these questions.
Table of contents
Strong and weak artificial intelligence
Methods of artificial intelligence
Subsets of artificial intelligence
Fields of application for artificial intelligence
Requirements for the integration of AI
Recommendations for the integration of AI
How Brunel can support you in your next project
Strong and weak artificial intelligence
Research and development work draws a distinction between strong and weak intelligence. Weak intelligence already accompanies us in our everyday lives and is integrated, for example, in voice assistants and chatbots. It mostly deals with specific use cases. Strong artificial intelligence, on the other hand, does not yet exist. The goal here is to develop intelligence that is equivalent to or even surpasses that of humans.
Strong artificial intelligence
Strong intelligence aims to give machines the same or even greater intellectual capabilities than humans. This would enable them to act intelligently and flexibly of their own accord. Characteristics of strong intelligence include logical reasoning, decision-making in the face of uncertainty, the ability to plan and learn, communication via natural language, and the combination of all these abilities to resolve higher-level goals. While researchers have not yet succeeded in developing strong artificial intelligence, they believe it is only a matter of time before they do so.
Weak artifical intelligence
Weak artificial intelligence describes systems that respond reactively to problems based on mathematical or computer science methods and are subsequently able to optimize themselves. In contrast to strong artificial intelligence, weak AI does not try to understand how to solve problems but focuses much more on performing clearly defined tasks and on subsequent optimization. Weak artificial intelligence is used, for example, in navigation systems, speech, text and image recognition, and in the personalization of advertising.
What methods is artificial intelligence based on?
Artificial intelligence uses two methods. One is neural artificial intelligence and the other is symbolic artificial intelligence.
Symbolic artificial intelligence
Symbolic artificial intelligence assumes that human thinking works on a logical-conceptual level in which knowledge is represented by symbols. By manipulating these symbols, machines learn to interpret and use them based on algorithms. The information that machines need to learn is provided by what are known as expert systems, which sort the information on the basis of logical if-then relationships. By comparing the information from the expert systems with their input, machines can thus learn constantly. The disadvantage of this heavily rule-based approach is that machines using this method of AI have difficulty responding to exceptions, variations and uncertain knowledge. They are also limited in their ability to acquire knowledge independently.
Neural artificial intelligence
Neural artificial intelligence emerged in the mid-1980s. It follows the structures of the human brain and the way nerve cells function. Similar to the human brain, neural AI segments knowledge into small functional units called artificial neurons. These lie on top of each other in different layers and are connected hierarchically. The top layer, also called the input layer, receives information from the outside. This can, for example, be data about patients, such as weight or body temperature. The information is then passed through one or more intermediate or “hidden” layers. The more intermediate layers there are, the deeper the neural network. In this case, we also speak of deep learning. When the input layer receives new information, each neuron in the input layer assigns a random weighting to the information flowing through it and adds what is called a neuron-specific bias term. The result is passed on to the next layer as input. The deepest layer – the output layer – marks the end point of the information flow and contains the final result of this information processing.
In order for the system to learn, the method of “supervised learning” is very often used in which the system learns on the basis of examples, i.e. real input-output data pairs. In this process, the neural network calculates a result based on the input data received. The system compares the result with that of the sample data set and calculates the magnitude of the deviation. This deviation is fed back into the neural network, and the weighting and the bias term are varied such that the deviation becomes smaller. The more often comparisons are made and the more examples a system receives, the better it can learn and get closer to the actual purpose for which the system was set up.
What subsets of AI exist?
Artificial intelligence is divided into three main subsets, each of which breaks down into four subdisciplines.
Action
The “action” subset encompasses language processing and knowledge representation and is itself divided into four subdisciplines: natural language processing, expert systems, predictive analytics and robotics.
Natural language processing: Natural language processing (NLP), also known as computational linguistics (CL) or linguistic data processing (LDP), attempts to enable computers to understand and interpret human language on the basis of algorithms. The aim is to enable humans and computers to communicate on the same level, effectively as equals. Google's translation function is an example of the use of natural language processing.
Expert systems: Expert systems are computer programs that use AI to derive solutions and recommendations for action. These systems require a knowledge base in which human knowledge about if-then relationships is processed in a way that is comprehensible to computers. Expert systems are used, for example, to automatically wean respiratory patients off breathing apparatus, or to predict earthquakes or floods.
Predictive analytics: Predictive analytics is a method that can forecast future events based on historical data. Predictive analytics is used in the insurance industry as well as in meteorology and marketing.
Robotics: Robotics concerns itself with the construction, operation and use of robots, and with their control, sensory feedback and information processing. Artificial intelligence is used in this context to plan, execute and monitor the actions of robots.
Perception
The "perception" subset likewise breaks down into four disciplines: image processing, speech recognition, text recognition and facial recognition.
Image processing: Image processing involves processing signals that are represented in the form of images or videos. With the additional use of artificial intelligence, image processing can support skin cancer diagnostics in dermatology, for example.
Speech recognition: Speech recognition is about enabling computers to process spoken language. Examples include voice assistants such as Alex or Siri, which already use automatic speech recognition.
Text recognition: Text recognition can be understood as the written equivalent of speech recognition and is already being used in chatbots.
Facial recognition: Certain programs can be used to identify people on the basis of certain facial features shown on digital images. Examples of applications in which facial recognition is already used include unlocking smartphones and passing through airport security checkpoints.
Learning
The third subset of artificial intelligence is that of "learning", whose specialist subdisciplines include deep learning, machine learning, reinforcement learning and crowdsourcing.
Machine learning: Machine learning technology "learns" by applying highly complex algorithms to a large set of data. The more data there is, the more the technology learns. As a result, tasks no longer need to be programmed separately.
Deep learning: Deep learning is a subdiscipline of machine learning based on neural networks – computer programs that use multiple layers of nodes, recognize patterns and make human-like decisions – and follows the lead given by the human brain. By drawing on the neural network in combination with enormously large amounts of data, the system can link what it has learned to new content and thus make predictions and decisions. It can also question its own predictions and decisions. Deep learning is, for example, already being used in autonomous vehicles.
Reinforcement learning: The special feature of this machine learning method is that it is based on a reward system: The machine learns independently which action is the best. The underlying data derives from extensive trial-and-error processes within a simulation scenario. Reinforcement learning is used, for example, in traffic light control.
Crowdsourcing: Crowdsourcing is also known as human computation. It involves getting a large number of users to solve tasks that cannot be solved by artificial intelligence alone. In this context, the very concept of crowdsourcing highlights the continuing need for human intervention despite artificial intelligence.
Where can artificial intelligence be applied?
AI can be used for all kinds of applications, both in the digital environment and in processes in the physical world. Examples of the use of AI in digital processes are chatbots, which are text-based dialogue systems that are able to learn on their own and independently derive answers from observed laws. Chatbots are often used in e-commerce customer support, for example. In retail as well as in logistics, AI can also be used to optimize inventory to reduce out-of-stock rates and increase sales and profits. Private households too are increasingly coming into contact with AI-based technology. Product recommendations that target users with personalized content via e-mail, social media channels or the web are one example. Another includes personal voice assistants, such as Alexa or Siri, that can act on spoken instructions thanks to voice recognition.
Embedded in physical processes, AI can also be used in the form of robots – for example to take over tasks that are harmful or dangerous for humans, such as welding or painting, or to complete activities that require particularly high precision. Artificial intelligence is also used in autonomous driving, the current technology trend in the automotive industry. In the future, vehicles will use deep learning to better understand their environment and ensure greater safety and comfort.
What does a company need to use AI profitably?
The use of artificial intelligence is becoming an increasingly important topic in numerous industries and business sectors. Depending on the objective, AI can deliver competitive advantages, reduce costs or guarantee security. However, certain requirements must be met if companies are to make profitable use of artificial intelligence.
Using high quality data
The basic prerequisite for the use of AI is large and high-quality data sets. Inaccurate data usually leads to unusable results. The time and place of origin of the data should be taken into account, for example. Data from external data sources should also be checked for its trustworthiness.
Consideration of unstructured data
Unstructured data is information whose file type is known but which cannot easily be stored in a database or other data structure (e.g. text files or e-mail messages). Compared to structured data, which is easier to find and process but only makes up a very small part of all available data in companies, unstructured data is a significantly larger data resource and should therefore definitely be taken into consideration.
Qualified experts for AI
In order for AI to succeed and be used properly in the company, it is crucial to have the necessary knowledge available in house. Developers should therefore receive appropriate further training and/or developers specializing in AI will need to be employed.
No success without human intelligence
Even though AI offers an almost infinite number of possibilities to optimize business success, it can only learn what developers actually give it in the form of algorithms. Only through efficient use can AI unlock its potential and sustainably improve business decisions.
Creating transparency
Artificial intelligence is regarded as a megatrend and more and more companies are investing in it. Often, however, only those who have developed the algorithms genuinely understand how the AI arrives at this or that decision. To avoid this situation, it is important to be transparent with all stakeholders and fully disclose algorithms within the company.
How can AI be integrated in your business?
Integrating AI enables companies to adapt products and services to the needs of their customers, to personalize experiences and thus to gain a clear competitive advantage. At the same time, AI forces companies to rethink, plan precisely and have skilled professionals.
Your application process
Define your objective
AI can be used for many applications. But not all possible applications will apply in each individual case. Therefore, the first step is to define what you want to achieve or improve with the use of AI.
Check data availability
Once the goals to be achieved with AI have been defined, the availability of the requisite data should be checked. It is important to ensure that the data suitable for the application is of a high quality and that there is enough of it to obtain meaningful results. Furthermore, the data should accurately reflect characteristics or criteria such as age, gender or other attributes of a group.
Perform initial data exploration
The third step is to start to explore the data. This gives you an initial understanding of the data you have available and its characteristics. Based on this step, important variables can be identified and categorized for potential models.
Define modeling methodology
Once the data you need has been selected, an AI model is created and you can start training the AI. Information is fed into the AI model. From the input data, the AI model attempts to identify patterns and arrive at a decision-making process. During this step, it is important to involve the stakeholders to a sufficient degree and to regularly compare the goals and requirements for the AI.
Validate the model
The fifth step is to evaluate the results of the algorithms, which means that performance metrics must be defined. These metrics in turn help to further refine the models. It is best to split the data into a test set and a training set. The AI is further trained based on the training set, and the results are compared with those of the test set.
Launch the product
Validation is followed by product rollout. Initially, the rollout should only apply to a limited group of users. If everything works as intended, the rollout can then be extended. In this step, regular feedback should be gathered from all users to further optimize the AI.
How Brunel can support you in your next AI project
Our experts boast an outstanding depth of training and experience in the field of artificial intelligence. They have acquired specialist knowledge in the programming and refinement of algorithms, in AI training and in the analysis of existing data and processes. To find the right experts for your project, we first obtain a comprehensive overview of your project: We want to know what tools, experience, qualifications and language skills our experts should have. In this way, we can ensure that our specialists develop AI solutions that meet your individual quality requirements.
Subsequently, our talent scouts and account managers search for suitable candidates via various channels and networks. A proactive and forward-looking approach to candidates as well as our global network of technical and IT experts are of great benefit to us in this context. Would you like to put our recruiting expertise to the test? Go here to find out more.