Auto AI: a new reality in building artificial intelligence
How to build AI with the help of ordinary employees, using modern solutions and the right methodology
Modern systems for process robotization, computer vision, voice analytics and the prediction of customer needs use mathematical machine learning models. People are disappearing from routine processes and from cognitive tasks whose solution is beyond their physical capabilities. The figures speak for themselves: according to Gartner, by 2021 the business effect of applying artificial intelligence (AI) will exceed $3 trillion. This was discussed at the IBM Data & AI Forum conference held in Miami at the end of October 2019.
The IBM Watson for Oncology system, for example, helps save human lives by finding the best ways to treat cancer. The Pandora music service analyzes musical works in order to recommend tracks that listeners will enjoy. A Tesla car is capable of being driven by a computer and, in Elon Musk’s view, artificial intelligence will soon surpass humans in the safety and reliability of driving.Neural networks were described by Warren McCulloch and Walter Pitts as far back as 1943. But the absence of digitized data on the areas being modelled and the low computing power of hardware made it impossible, until a certain point, to build high-quality systems driven by artificial intelligence.
Evgeny Scherbinin - CEO of Prime Source
Having acquired sufficiently powerful computers and a strong mathematical apparatus, humanity has run into a new problem — a shortage of people who know how to apply mathematical models to business tasks. They came to be called data scientists, an acknowledgment that these are not ordinary specialists but rare scholars with a special talent.
That talent lies in the fact that a data scientist describes their observations of certain processes with a set of mathematical formulas. Systems that perform functions that previously required human involvement are built on this principle. Examples of such systems include the automatic identification of a person by voice or face, the detection of fraudulent activity among tens of millions of banking transactions, determining the composition of raw material components to obtain the optimal alloy in manufacturing and many others.
The presentations given by participants of the IBM Data & AI Forum confirmed the conclusions of the Prime Source team about the role and functions of the data scientist, drawn from ten years of experience in building predictive analytics systems in banks, telecommunications companies, the oil sector and public administration. In practice, a data scientist is an ordinary role in a team building an AI system, in which the employee uses specialized software, together with standard tools and methods, to build and test mathematical models. The key points here are the composition and roles of the team and the composition and architecture of the software used. A little more on that below.
Besides the data scientist, a team building an artificial intelligence system includes a subject-matter expert, a business analyst, a developer, a data engineer, DevOps specialists and others. The team works to a set of rules that has proven its effectiveness in practice and that assumes the involvement of different roles at different stages of the process. It is worth noting that Kazakhstan certainly has brilliant data scientists, just as it has outstanding experts in many other areas of business.
As for software, modern teams use an integrated set of applications in their work — the so-called Auto AI. It includes tools for accessing data from various types of sources, automatically identifying predictor variables, building, testing and optimizing models, analyzing and visualizing modelling results, embedding models into business processes, working collectively on models and much more.
One example of the new architecture for working with AI is the project to implement IBM Watson Assistant and Watson Natural Language Understanding at Lufthansa Group in order to optimize service at the Service Help Centre (SHC). Just 22 SHC employees serve 15,000 Lufthansa agents at 180 Lufthansa sales sites around the world. They answer questions about service, check-in and boarding — around 100,000 enquiries a year. Thanks to the integration of IBM Watson AI services, previously disparate data sources are now combined and used to generate answers to customer queries. In addition to reducing customer waiting times at check-in, the quality of the information has been improved overall.
Evgeny Scherbinin, CEO of Prime Source