Intelligent machines on the rise
Are we experiencing an unprecedented technological advance through artificial intelligence (AI)? There is a lot out there that suggests this! Still, it will take a long time for machines to truly become equal or superior to humans. Until scientists and researchers have developed super-intelligence (strong AI), we humans will continue to co-exist and work well with weak AI.
While strong AI can replace humans, weak AI is an extension of our cognitive skills and gives us already today great advantages in mastering specific application challenges. AI will become a core component of the modernization of society and the economy. It will support us immensely in coping with large-scale global challenges — for example, in developing more intelligent cities and safer and congestion-free traffic, in lowering energy consumption and optimizing our power grids, in cutting down on carbon dioxide emissions, and in protecting the internet more effectively. In the global race to boost productivity, AI will be a decisive factor with regard to demographic development.
Scientists like to present us their latest innovations in the battle between humans and machines. Just look at chess and the popular Asian game Go and you will see that, in these cases, the ma-chine is cognitively already far superior to humans. It would be vastly exaggerated, however, to say that machines are generally speaking superior to humans because of this example. It will take a while for strong AI and humanoid robots (androids) to acquire the skills and capabilities of humans.
More than a string of ones and zeros
AI is supposed to mimic human cognition and actions in machines. In the case of a car navigating its way through urban landscapes, this can take place using a relatively simple algorithm which finds the shortest route; or, it can result from a complex neural network that, for example, steers autonomous cars and all traffic towards the most efficient and congestion-free route. This kind of neural network is being trained with huge data volumes to recognize patterns and derive its own conclusions. There are various approaches to achieving artificial intelligence. So far, however, no approach has emerged that has proven successful. Development is still in the starting blocks.
Although deep learning as a method is a good three decades old, it has experienced a revival of sorts lately. You can imagine deep learning as a type of filter that works its way from rough to fine results, thereby increasing the probability of generating a correct one. It is made up of layers that build on one another and uses the results of the previous layer, which creates a continuous learning process. With its immense computing power, deep learning is able to dig through enormous amounts of data. It succeeds where other approaches have failed.
AI as a black box?
How AI gets many of its results remains largely a mystery; the more complex the machine, the more intricate the task is of reading these black boxes. This may be due to the fact that, in contrast to structured data storage in, say, a database, AI data is stored in fragments that are practically unattainable for humans.
The need for a glass box instead of a black box is more than understandable here. Researchers are working hard to find a way to make computer algorithms more comprehensible for humans. Advances in AI mean that the above-mentioned forecasts, decisions and actions of a machine are based on criteria that constantly update themselves as the algorithm continues to learn.
It remains to be seen if deep learning is the end-all technology. There are other promising AI approaches. Time will tell which approach is the right one for which use case. The best approach is probably a combined one, for example, with semantic (ontologies), static (deep learning), etc.
With value comes acceptance
Given the high number of publications on this topic, it is hard for us readers to judge in single instances if a machine or an application labeled intelligent really is intelligent. Just because something is called intelligent, doesn't mean it really is! The term intelligence is vastly overused and defined in so many ways, but there is no one universal definition. Likewise, AI applications vary widely in their cognitive capabilities. Ultimately, something is intelligent if it offers value.
Weak AI and rule-based systems already offer us considerable benefits and enormous potential in the future. They manage financial transactions, make forecasts, and simulate weather and economic developments. AI detects anomalies, for example, in the form of credit card fraud. It is an excellent means of making diagnoses and prognoses in medicine. Particularly notable is that artificial intelligence can first evaluate radiological images before the radiologist makes a final diagnosis. When it comes to recognizing patterns in texts, images, handwriting, materials and substances, AI is more advanced than humans. It is crucial for predictive maintenance and repair. Artificial intelligence is particularly useful when the best solution or the best possible decision must be made based on huge data volumes and a high number of options. AI has made important progress recently in the area of complex challenges, for instance, in language control and processing.
Humans love convenience. After all, they invented the wheel, because they preferred to drive rather than walk. Just like humans did back then, we are now asking ourselves: Does AI help us? Does it make our lives easier? Where there are benefits, there is acceptance. The more similar interactions with AI are to human behavior, the more accepted AI will be.
On the path to disruptive core technology
Artificial intelligence has great potential in the realm of economics and business. It not only relieves workers from having to do repetitive or even dangerous tasks, it is also much faster in analyzing data volumes, making decisions based on this, and completing tasks. What's more, robots will further automate production, which will open many new doors. For example, countries such as Germany will become a more attractive production location, thus increasing its competitiveness. There will no longer be any economic reasons for outsourcing production to low-wage countries. Whole new business areas will emerge as a result of AI joining up with connected products, processes and machines (Internet of Things; IoT).
AI is developing more and more into a disruptive core technology. It will revolutionize our working lives and current software applications! Let's keep our expectations realistic though: instead of drastic breakthroughs, progress in development will come gradually. For the time being, there won't be a do-it-all AI application that can combine human-like skills and capabilities.
Just like humans, machines are also capable of making mistakes. As long as human health, life and death are not at stake or people are not being assessed, mistakes are acceptable. Using a percentage tolerance level, we humans will define probabilities which allow us to decide if a computation is correct. We will no longer have to complete tasks or process steps ourselves, but will monitor and optimize machines while they handle them.
Information is the key
Information is the linchpin of artificial intelligence. When we use artificial intelligence, we learn from information and utilize this knowledge to make automated decisions for concrete outcomes. We teach AI systems to analyze data and give them algorithms to solve problems. To make this possible though, we have to prepare a basis of information for AI. Future AI systems will train themselves as well!
In contrast to humans, AI needs large data volumes to learn effectively. Again, different from humans, it can easily handle large data volumes. For example, to recognize an image of a dog, it needs hundreds of thousands to millions of images with and without a dog. A small child can recognize a dog at first glance, though. Algorithms have to analyze large data volumes to be able to solve a given problem. While machines today need mass amounts of data to get results, the good news here is that they are also capable of actually processing such a large amount – in comparison to the first wave of AI in the 1980s. This puts an end to one paradox that there is scarcity in plenty! Right now, too much information slows down decision-making.
Inundated with information? A blessing
Nothing happens without information. A barrage of information may be seen as a curse today, but it can be a help to us in the future. Our cognitive skills are overwhelmed though by the flood of information. We don't even use 80 percent of the information we gather! With every passing day and year, the information multiplies. Industry 4.0, the Internet of Things, etc. are all causing a dramatic increase in data: By 2020, the worldwide volume of data will increase tenfold, going from the current 4.4 zettabytes to 44 zettabytes.
Today's barrage of information is ideal for artificial intelligence applications. Still, ERP software like SAP is not capable of processing the bulk of this information. What's needed is a context-sensitive software that can efficiently manage and store data volumes and, if necessary, scale horizontally. This is and has always been the intrinsic purpose and capability of enterprise content management systems such as Doxis4 from SER. Just look at DHL Express for an example of this: 8.5 billion documents are currently being stored in the Doxis4 information repository. These documents are accessed by one million users per day on average. In peak times, 1,060 accesses per minute are registered.
Already 20 years ago, 80 percent of all information in a business context was unstructured and only 20 percent was structured. This remains the same today. In the information repositories of SER's Doxis4 ECM software, all of this information can be found – including SAP data (both current and archived data), emails, documents, social media content, websites, machine data, images and videos.
Cognitive services are home to AI
In the age of artificial intelligence, information is finally becoming a production factor. Information logistics will become one of the strongest influencing factors of value creation. If you already know the value of your information and store it in an ECM system, then good for you. The information repository, the core of Doxis4 ECM software, functions as a safe for the new currency in business: information. Used as a digital archive, this information repository stores empirical values and has the ability to remember. This omnipresent intelligence gives a new architectural layer to Doxis4 ECM: cognitive services. All AI technologies of deep content analytics (deep CA), ontology, and natural language processing (NLP) are covered in cognitive services and accessible on all ECM applications. Moreover, there is innovative potential in the ability of these services to link, say, statistical and semantic modeling.
Federated content integration service
Information management is technological and complex, which is a challenge for companies today. In addition to SAP, numerous other business applications are being used and their content is stored in separate databases and structures. This is already a source of woe for the productivity of knowledge workers. As long as this situation is not factored into decision-making, it will have negative consequences on the outcomes of AI in the future. AI needs data from various information sources to be able to learn and make forecasts. The integration of information silos spread out across companies is strategically more important than ever for IT teams.
Incorporating these sources of information into various metadata structures will be handled by a federated content integration service in Doxis4. The content of the information silos can be migrated, but it's not a must! Through the federated metadata platform of Doxis4, it is possible to access any databases, file systems, etc. from Doxis4. AI crawlers are used by the platform to track down information from the sources. To avoid an information overload, not all information is permanently archived. Just like the human hippocampus, it is also possible here to transfer the most relevant information from the short-term memory to the long-term memory.
Contact with new technology
The human-computer interface – a terrible technical term – is no longer limited to a keyboard, mouse, scanner and camera. Soon all types of devices, products and software applications are supposed to respond on demand. Instead of typing and pressing a mouse or touchscreen, we can simply speak a command. Not in a technical language, but in the same way we would communicate from person to person. The possibilities of natural language processing for ECM are currently the topic of a joint research project of the Austrian Institute of Technology (AIT) and SER.
Listening to the machine
Instead of reading for ourselves, the machine will tell us in a natural-sounding voice what our mail or documents say. It will be able to determine what is important or not. We will be able to have a dialog with the machine and ask questions about the content of documents and records, which the machine will answer. Future ECM software will feature language synthesis to find the right words in an understandable language and to translate what is spoken back into computer code.
This is a huge advantage for mobile use. Our ear essentially becomes another interface to the machine. In addition to getting the document we need, we can also get key data about business partners, business processes or a project. The next ECM generations will be able to utilize practically all of the IT innovations of the past several years. These include acoustic language recognition, context-based processing of spoken sentences, correct context assessments, and searches for the right answer.
No more user interfaces
In contrast to humans, virtual agents do not need user interfaces. In the future, user interfaces for capturing data and for searching, forwarding or filing information will not exist in the traditional sense. As evident already in financial transactions, humans will only get involved in the business process if the system registers an anomaly or is out of control. With such algorithm-based ECM systems, business processes and many decisions can be automated for the most part. These virtual agents can be integrated into one another simply per plug and play. Connecting virtual agents causes new syntheses.
The companies most likely to be impacted by this will be those of the financial services sector where the administrative staff is primarily processing information. Accounting departments will also benefit from AI. They have to deal with huge amounts of data and its growing complexity due to new legal regulations and stricter compliance requirements.
Automated inbound invoice processing is one means today of automatically processing or even of automatically posting a good portion of inbound invoices from all kinds of input channels. In all likelihood, this process will become almost completely automated. Furthermore, any mistakes or anomalies in transactions will be made transparent. As they have been in the past, the financial services sector and accounting departments will probably be early adopters of these AI-based ECM systems.
Digital mobility is making progress, and it is crucial that we don't ignore new developments. With every technological advance, a certain number of setbacks have to be suffered until the technology is ready for the masses. While laptops, smartphones and tablets are enough for us right now, they won't be in the mid-term. Apple, Google and others are already announcing the next era of devices: wearables. These mini-computers can be worn and operated directly on the user's body. They include smart glasses, smart watches, clothing items and much more. Not only do they provide information, they also use sensors to record information based on the context. Part of this information is directly processed on the device before being transmitted. Insurance providers – for example, health insurers in Germany – are experimenting with this right now as a component of their bonus programs. While ECM can store this kind of information, Doxis4 also factors in data protection aspects into the recording of who did what with which data and at what point in time.
Along with language control and keyboards, smart glasses (head-mounted display; HMD) are one of the most important ECM interfaces in communication between humans and artificial intelligence. They project information into the user's line of sight. In conjunction with augmented reality, they can also display virtual objects, thus melding the real environment with the virtual.
Smart glasses will be used in industry, retail, logistics, medicine or trade professions. They provide all the information a user needs right in the line of sight. For example, working hands-free through speech and movement, engineers, mechanics or technicians can view and read construction plans, maintenance manuals, etc. directly and without interrupting what they’re doing. Warehouse staff are shown how to get to goods in a warehouse thanks to augmented reality. Once the person is at the right location in the warehouse, there is no need for a hand scanner; the camera of the smart glasses reads the codes. During this process, the warehouse worker is given further details about the goods, the packaging unit, available warehouse quantity, a product description, etc.
Smart glasses will not be applicable to all use cases. And they don't have to be! It will also be possible to project documents, plans, etc. in high resolution images onto any surface, which can be operated by hand or speech. Even holograms seem feasible: Japanese researchers have managed to project a hologram with an ultrafast femtosecond laser. This femtosecond laser uses ionization to turn air molecules into plasma, which then emits energy in the form of light, creating points of light that display images floating in the air.
The Japanese researchers call this a voxel, a term originating from the words "volume" and "pixel". The Japanese team has successfully overcome the key problem of previous hologram attempts: the plasma field can be touched without painful burning. This technology is on the brink of a breakthrough. While traditional holograms have been simply too large to use on many electronic devices, researchers have now succeeded in making holograms ten times flatter than before using special quantum material. With these advances, holographic 3D projections could be feasible in business and in daily life.
Virtual reality (VR) in computer games is already a part of daily life, but it is still just a pipe dream in the working world. Given the immense potential of this technology, it is likely that VR will reach and change many work areas in the next couple of years. Practical use cases are imaginable for development and design processes, plant and mechanical engineering, building and landscape architecture, training, service and maintenance. Even before a machine and plant are complete, an adviser can already look at the object in the virtual world and provide relevant information.
Similar to a flight simulator, employees can learn to operate computers, machines and plants via virtual reality. Through VR, technicians can familiarize themselves with plants and buildings before performing service and maintenance. In all of these scenarios, ECM provides the necessary information on training, operation, etc. directly during work; there's no need to run back to the office. If there are questions regarding the maintenance of a design, construction documents and blueprints are visualized. Virtual reality and augmented reality are both steps towards paper-free maintenance.
VR will also join together our global world even more. Employees across various locations and from diverse companies can gather at a virtual location to not only work on a joint project, but also to explore a model within the project. Discussions at this virtual location require real information, which an ECM system can supply.
Once the Doxis4 ECM system has read and understood information, it has to decide what should happen with it. This is the job of classification. Parallel to image recognition, the system learns to differentiate between documents and information via learning sets. You could say that the machine bases its knowledge on past experiences. Different from machine learning, AI will use deep learning to train itself in the future. At the same time, classification will go well beyond document types. Based on assumptions made through classification, AI will develop classification suggestions and forecasts. This way, it will be possible to logically distribute information across input channels and to steer it to the next processing step. One of the fundamental attributes of AI is its ability to find hidden patterns or correlations in the chaos of unbelievably huge data sets and then to develop models that forecast behavior or outcomes.
Patterns also pertain to emotions. Sentiment analysis, often called sentiment detection, is applied as a subdivision of text mining to recognize positive and negative moods in text documents. It essentially tries to understand human emotion by reading texts written by humans and extracting their subjective opinion. This analytical method is interesting from an economic standpoint, because it identifies positive and negative opinions about a company and its products published in written correspondences and in the internet. This kind of knowledge can reveal, for instance, that a customer switched providers and products due to dissatisfaction. To run the most effective automated sentiment analysis, domain knowledge is required. Technology, application, user friendliness, etc. would be domains in software. The adjective "quick" would be praise for a technology; for user-friendliness, however, it has a negative connotation that might mean "quickly leads to errors".
Within business processes, digital assistants can run things, at first handling routine tasks, and, later on, making decisions in specific situations. As described earlier, decision-making is one of the biggest strengths of AI. Using past experiences, AI can make forecasts regarding upcoming business processes. With human support, AI forecasts can help determine capacity utilization and thereby enable faster processing. Automation would also mean that intellectual work no longer needs to be outsourced. Simple and complex tasks can be completed more efficiently by machines.
Proactive information management
If we need information, we need to find it. We have become accustomed to this. In fact, we spend 20 percent of our working hours looking for it, as a McKinsey study has found. This could also become a thing of the past. Instead of a search machine, AI paves the way to a finding machine; essentially, proactive information management. Predicting our needs, it gives us information in the context of our work, actions and decisions. The system knows its users, their information needs and work habits. It gives us a summary of all important information for an appointment already before it takes place; the system sees in our calendar what's up next. It suggests the right documents based on the business context, and we never have to search for a thing.
One of the most important fields of application for the industrial Internet of Things (IIoT) and Industry 4.0 is inarguably predictive maintenance and service based on predictive intelligence. Predictive maintenance utilizes algorithm-based data analyses to monitor the status of machines and plants. It identifies potential malfunctions already before damage incurs. Technicians have all relevant information for maintenance directly on their smart glasses. The glasses scan machine or assembly group codes and thereby supply precise information required for maintenance and troubleshooting. Wearables also warn and protect technicians from bodily harm, for example, through heat or radiation. They can also analyze the environment's conditions, while AI evaluates the recordings.
These kinds of use cases are still in the early stages though. Unlike in production, virtual assistants will take a while to emerge here. Human deficiency has a big advantage over AI: in contrast to machines, we humans are forgetful. If machines learn something incorrectly, it is difficult to fix this later on.
Digitalization at light speed
Despite all the hype, digitalization is not so far advanced in companies that we can consider it a given. Digitalization is a prerequisite for artificial intelligence. The topic of digitalization may be firmly established in the minds of C-levels, but to what extent do their companies actually practice it? Can and do we really want to change information management and all of its business models and processes?
Many firms have become complacent due to their past successes and don't feel the pressure to start changing anything. Which is a mistake, as the competition is definitely one-sided: disruptive companies are the children of digitalization; they have been masters of it since birth.
The digital revolution is forging ahead at light speed. It will not wait. We are only at the beginning of epic developments in intelligent machines and the end is nowhere in sight. Until AI is fully developed, we should take the time to push ahead with digitalization. Without a powerful deep learning repository, AI applications will lack big data "food". These and many more aspects are all reasons why ECM systems need to be at the top of the agenda for most companies. For already 20 years now, the ECM system Doxis4 has been using neural networks for classification and extraction. ECM systems are highly valuable and practical – they just need to be deployed.
It's the first step in the right direction!