Using AI to improve the efficiency of knowledge workers

New capabilities of AI, capable of recognizing the context, concepts and meaning of concepts, open up new, sometimes unexpected ways of working together knowledge workers and machines. Experts are able to make their contribution to training, quality control and fine-tuning the results of the work of AI. Machines can complement the knowledge of their fellow people and sometimes help bring up new experts. These systems, more likely to imitate the human mind, are more reliable than their predecessors, depending on the data. And they can have a significant impact on knowledge workers who make up 48% of the US workforce — and more than 230 million knowledge workers around the world. But in order to take full advantage of this smarter AI, companies will need to revise workflow and jobs.

Knowledge workers - people who make decisions at work, reason, create and apply ideas in cognitive processes not related to routine - for the most part agree with this. Of the more than 150 experts taken from a large international study about the use of AI in large companies, almost 60% say that their old job descriptions are quickly becoming obsolete due to their new collaboration with AI. About 70% say that they need retraining because of the new requirements associated with collaboration with AI. 85% agree that senior management should also be involved in general attempts to change the roles and processes of knowledge workers. And starting the task of rethinking the use of mental labor in conjunction with AI, they can apply some of the following principles:

Let people experts tell the AI ​​what is important to them. Take the medical diagnosis in which the AI ​​is likely to be everywhere. Often, when an AI issues a diagnosis, the logic of the algorithm is not clear to the doctor, who must somehow explain the decision to the patient - this is the problem of the black box. Now, Google Brain has developed a system that can open a black box and translate its work into human language. For example, a doctor evaluating the diagnosis of AI "cancer" would like to know to what extent the model took into account various important factors - the patient's age, previous chemotherapy, etc.
')
The Google tool also allows medical experts to introduce concepts that they consider important to the system and test their own hypotheses. For example, an expert may want to see if the introduction of a new factor, previously unaccounted by the system, will change the diagnosis. Bin Kim, the developer of this system, says: “In many cases, when working with applications on which a lot depends, experts in a particular field already have a list of concepts that are important to them. We at Google Brain are constantly confronted with this in the medical applications of AI. They do not need a set of concepts - they want to provide concept models that are interesting to them. ”

Make models that match your common sense. With the accumulation of cybersecurity concerns, organizations have become increasingly using tools to collect data at different points in the network for analyzing threats. However, many of these data-based technicians do not integrate data from multiple sources. They also do not include the common sense of cybersecurity experts, who imagine the spectrum and diversity of the motives of the attackers, understand the typical internal and external threats and the degree of risk for the enterprise.

Researchers from the Alan Turing Institute , a British state institute where they study data science and AI, are trying to change this situation. Their approach uses a Bayesian model — a probabilistic analysis method that takes into account the complex interdependence of risk factors and combines data with estimates. In cybersecurity networks of enterprises among these complex factors are a large number of devices connected to the network, and their types, and the knowledge of experts of the organization about hackers, risk, and much more. And although many AI-based cybersecurity systems include the ability for a person to make decisions at the final stage, researchers at the institute are looking for ways to incorporate expert knowledge at all levels of the system. For example, expert knowledge of motivation and behavior associated with an attack through IP theft — and how they differ, for example, from a DDOS attack — is directly programmed into the system from the very beginning. In the future, this knowledge, together with data sources from machines and networks, will be used to train more effective cybersecurity systems.

Use AI to help newcomers become recognized experts. AI is able to quickly turn novices into experts. HP demonstrated this by using a cognitive computer platform from the AI ​​lab to analyze data on customer calls over two years. The call center used the backbone queuing system to distribute calls, due to which customers had to wait a long time for an answer, and the quality of user support was low. The cognitive computer platform was able to identify the unique “micronautions” of each specialist - knowledge of specific types of user requests received from previous calls. Now it is used to redirect calls to agents who successfully handled similar situations earlier. As a result, the support center improved by 40% the indicators for resolving the situation on the first call, and reduced the number of call redirections by 50%.

As training for AI support specialists automatically updates their knowledge, eliminating the need to do it manually in their profile. Moreover, the more knowledge a specialist receives, the more complex tasks the software is redirected to him. In the meantime, the software is constantly improving its knowledge, and the AI's conclusions about micro-skills improve the efficiency with which the expert teaches the software. It is worth noting that several companies are working on this task of retraining; for example, an ASAPP startup provides real-time offerings to support service specialists.

Use effectively using these technologies of AI for the layout of workflow experts. Experts on many types of knowledge are quite rare, and do not give a large amount of data suitable for training. But deep learning and machine learning, on which many breakthroughs in the field of AI are based, require a huge amount of data to build bottom-up systems. In the future, we will see more top-down systems, which will require much less data to create and train, which will allow them to perceive and take into account the special knowledge of employees in their work.

Take the recent competition organized by the Medical Image Processing Laboratory at the University Hospital of Brest and the Faculty of Medicine and Telecommunications in Brittany. Participants competed in the greatest accuracy of medical imaging systems, which were supposed to tell which tools the surgeon used at each time point in a minimally invasive cataract operation. The winner was the machine vision system, which was trained for six weeks with just 50 video recordings of surgical operations, 48 ​​of which were conducted by experienced surgeons, one by a surgeon with annual experience, and one by an intern. Accurate instrument recognition systems allow doctors to analyze surgical operations in detail and look for ways to improve them. Such systems can potentially be used in generating reports, training surgeons and assisting surgeons in real time.

All these examples indicate that engineers and pioneers in various disciplines are developing AI, which could be easier to teach and evaluate, and also include extremely valuable and often rare expert experience. To begin to take advantage of these new features, organizations need to review their budgets for AI. And in order to extract the maximum from both these systems and knowledge workers, they need to reconsider the interaction of specialists and machines. As today's MO systems complement the capabilities of ordinary workers, so tomorrow's systems will raise the efficiency of knowledge workers to previously unattainable heights of perfection everywhere