So far, we have charted the problem-solving processes and the data inventory for a specific context. In lesson two, we said that one of the pillars of building a successful AI strategy is to ensure that your organization is well-placed to maximize the use of a systems. Thus, the last step to consider in designing your AI strategy is the organizational structure of your workplace and the existing competence. The goal of inspecting the existing configuration of work is to link the future AI systems with systems already in place. The question that will guide us here is what current activities are taking place that would enhance or hinder our ability to use AI techniques to solve a specific problem? What knowledge relevant to the implementation of AI technology exists in the organization? AI technology is systemic, in other words, it benefits from large amounts of connected information. Therefore, introducing a processes in one area of your business should be done while keeping in mind the fit with other ongoing or planned digital automation and transformation processes. For example, let us suppose you want to automatic is the decision regarding reimbursement by introducing an AI system, getting its data from the past and present expense reports from all members of the organization. Currently, the expenses are reported electronically, including the receipts via a software, let's call it expense. The future AI system will be connected by a documented API to expense to pull all the necessary input data. The questions that need to be asked in order for linking AI implementation with current and future digitalization plans is will the expense software continue to be the expense reporting system? Is it scheduled for any major updates? Are there any criticisms about the way expenses currently working? Asking these questions will help avoid the situation in which one group or section of the organization deploys an AI system while another group is about to change an element upon which the AI system depends. You can imagine the lack of coordination across teams would result in the failure of AI implementation. The last element that needs to be mapped out is the knowledge relevant to AI system that already exists in the organization. As with existing data and processes, the AI competence of the workforce may be dissipated outside the IT unit. Awareness about the existing hierarchies of knowledge is another requirement. We discussed in the first lesson of this week how disruption of its not only markets but also competence. Appending established hierarchies may result in resistance to the new technologies. To prevent this, chart the competence needs of the staff and addressing them head on via, for example, upscale or rescale programs supported financially by the organization. We will talk more about risks associated with competence disruption in week three of the course. For now, it is enough to know that knowledge just like data is a precious resource that needs to be fostered and rewarded. Can you think of the current knowledge hierarchies in your organization? Can you anticipate that the introduction of AI as a colleague in the workplace will change the value of existing competence? How much resistance do you expect to meet from the people embedded in the existing structures? [MUSIC] Some of the knowledge that can be AI relevant pertains to the techniques, capabilities, and applications of artificial intelligence. As we will soon see, these properties have to do with software development and hardware management. But also with the family with certain data types, and with logic, and reasoning. Someone may have AI relevant knowledge because they understand how photography, or design, or copyrighting works. These people can be the experts that train your future AI system, and thus their knowledge is directly relevant to AI implementation even if they are not computer programmers. I mentioned the techniques, capabilities, and applications of AI. I will briefly define some of them now so that you can identify people in your workplace that are familiar with one or several of these. AI techniques are ways to reason about the world that enable AI systems to either modern reality or to discover defining patterns that describe it. Examples of model driven techniques are logic based reasoning, knowledge representation, or planning. Some of these can just as much be learned in data science courses as in philosophy courses. Examples of data driven techniques are supervised and unsupervised learning or reinforcement learning. These techniques are derived from advanced statistics and mathematics. Modern driven techniques are the pioneers of digital technologies. They have been used for decades in building AI solutions to problems from manufacturing, to medical diagnosis, to internet searches. The techniques mentioned have to do with how to represent information, how to reason over it, and how to plan most effectively. Take knowledge representation, knowledge representation is about structuring information in an understandable way for computers. For example, organizing documents so that they are appropriately labeled with labels that describe the content, and fit it into categories, and the relationship between them is clearly marked. Knowledge representation is used for example to sort through large amounts of data. Data driven techniques have taken off more recently, even though they have been invented a long time ago. Supervised learning is one such technique wherein algorithms use annotated data, that means data with a correct desired answer already provided. In training, the output of the model is supervised, and the algorithm is informed on whether it got the right answer. This information is then used to adjust the model. The AI techniques are used to implement capabilities. Capabilities are specific systems that allow us to understand or change an aspect of the world. Some AI capabilities are computer vision, speech processing, natural language processing, robotics, and prediction. In other words, machines can make sense of visual and text data of physical movement of speech. They can also be built to predict trends on the basis of existing data. AI capabilities in the field of language are directly relevant to any organization because no matter with what data we work, we communicate and share ideas and results using words. AI systems have been built so as to understand and manipulate language, both voice and text, and even to generate language. Speech recognition, for example, is the result of interdisciplinary collaboration between linguists, mathematicians, and data scientists as well as sound engineers, and many other agents in the fields. Described in a very simplified way, the process of recognizing what a human has said begins by cleaning the noise from the audio data that has been collected through a microphone. This data is then broken down not into words, but into even smaller speech path called phonemes. Phonemes are language specific and have been labeled as such by expert linguists. Added to this are models for pronunciation and for language so that a computer can recognize that which is being said to it. An important caveat is that the speech recognition models are domain specific, they work best for a defined area of communication. Anyone who has attempted to dictate a professional text to either Siri or Google Assistant would know that it does not quite work. However, communicating simple sentences about everyday situations works practically without fail because this type of speech is the data that the algorithms have been trained on. Finally, applications can solve a real world problem by combining capabilities. Applications of AI are numerous but some of the most frequent can be named as project planning, legal case research, documents search, automated support, fraud detection, and fault prediction. Because the applications of AI in the workplace are so many and so context specific. In the next lesson, we'll be dedicated to giving you some examples of AI implementation. [MUSIC]
