Artificial Intelligence

AI Cognitive Services

Imagine for a moment that this article could be written automatically by simply specifying the main topic and the ideas that should be reflected in the article. Then an AI (Artificial Intelligence) application can be asked to create the article in seconds. Perhaps we are not that far away from this reality.

By Sergio Jimenez-Otero, Senior Data & Analytics Consultant

14 December 2022

We are experiencing an explosion of AI technologies that are rapidly evolving from research to mass technologies. One of the technologies that has attracted the most attention in the past year is in the area of generative AI technologies [5], [6]. More specifically, the technology that is able to change synthetic images, art or the style of images and is available in applications such as DALL-E [7], Nightcafe [8], Midjourney [9], [10]. A simple text description of our image, including the style, can instantly create synthetic images that closely resemble real objects or even artworks (text-to-image => [11], [12]). Not only that, it can also change the style of our real images and retouch certain parts of the image. This is only the case with generative AI technologies. AI applications such as the automatic generation of text, music, art, faces, animations, the restoration of old films and the ageing or rejuvenation of faces are other examples in this category. Other types of AI technologies include automatic communication through increasingly sophisticated chatbots (conversational AI [13]), automatic translation systems in a variety of language pairs, both written and spoken (machine translation [14]), recommendation systems [15], natural language systems for extracting key phrases or entities (NLP AI [16]), automatic recognition of objects, content and people in images or videos (computer vision AI [17]) and many more. These AI technologies are expected to fully take over tasks related to content creation and the generation of images, videos and audio. Such tasks may deal with the automation of processes, the search, recovery and synthesis of stored information, the recognition of people and writing, human-machine communication, but also more complex intellectual activities in the area of content and report creation and the generation of high-value insights for organisations.

"The future of AI has only just begun..."

But this is just the beginning. We will be amazed at what these technologies can do for us in the coming years. The most important thing is that these technologies will go from experimental to everyday use in our lives and businesses in the shortest possible time. These technologies are not here to replace us, but to help us with boring and repetitive tasks. Tasks in which humans perform cognitive activities that do not require any effort on our part, but are present in our daily lives and businesses. All this results in a multitude of opportunities for the complete digitalisation of companies and can consequently lead to great savings and optimisation of business processes [18], [19]. Recent studies show, for example, that AI-powered call agents will replace humans in 15 % of customer communications in the next three years, which corresponds to an estimated saving of 80 billion US dollars [20]. Many of the AI technologies mentioned above are already being used as cognitive services in the cloud, via the REST API or in other ways. They are quickly becoming commodities and can be used by ordinary people and companies.

But how can I integrate these AI technologies into my company? How can all these cognitive services and AI help me in my business, with new customer needs and process optimisation? The answer is that many companies already have the data and access to the necessary technologies to utilise these cognitive AI services to support their business. Many organisations started collecting data and using big data technologies a long time ago. Much of this data remains stored in data lakes or similar digital repositories without being explored or utilised. There is great potential in these sources when big data analytics and AI cognitive services technologies are used. A single source or a combination of several is sufficient to perform many use cases and optimise processes. For example, the same streaming video from cameras on site could serve different purposes. It can be reused for automatic intrusion detection, safety of people in hazardous environments, automatic reporting of industrial activities and inventory, production triage and fault detection. In these cases, the cognitive services of machine vision would perform most of the tasks. As another example, a single source of documents, texts or written communications can be used for automatic document classification, document categorisation, summarisation and extraction of important ideas or keywords. It can also be used as a source of information in intelligent searches. In all these cases, cognitive AI services for speech will be a great help in developing solutions.

And how well are companies doing today in terms of AI technologies and cognitive services? Are organisations using all these technologies to really improve their business and customer relationships? The reality is that many companies lack a consistent vision and strategy regarding the use of big data analytics, AI and cognitive services. Many companies have simply failed or are now struggling to sustain the use of such technologies in their business [21], [22], [23]. The reasons why many companies are not successful in using AI and big data technologies are manifold. One of the explanations is that many companies focus on solving individual cases or at most several similar cases. Companies develop data pipelines and AI models from scratch and in most cases using open source technologies. Therefore, the effort to adapt their framework to new AI technologies or new data, update AI models and maintain the infrastructure is so great that it is not sustainable in the long run. Other reasons have to do with the fact that in many cases neither the data nor the data pre-processing was appropriate for the business case. Recent studies also show that more than half of companies (58%) focus on solving the so-called "need-to-do" cases and only 46% of companies carry out the "must-dos", which bring great benefits with little complexity [24].

There is also the problem that attempts are made to build use cases around data of analogue origin instead of redesigning systems to use purely digital data. For example, handwritten texts, scanned documents or handwritten forms. It is also important to bear in mind that all these AI technologies are developing very quickly, which requires highly specialised experts, up-to-date and technical knowledge and the continuous updating of AI frameworks. This in turn means very high costs for companies in relation to the benefits. As a result, some companies have abandoned AI projects whose benefits did not justify the huge investment. In short, many companies have hired large numbers of engineers, spent huge amounts of money on resources and developed their own systems from scratch without a solid big data strategy or ROI projections.

AI and its cognitive services are merely tools in a business toolbox. Tools that extract insights, find complex patterns or optimise processes. We cannot pretend to solve all cases with these technologies alone without knowing the data and the business. Don't try to solve isolated problems that are part of a much larger digital entity, nor use analogue information in a digital world. The potential of these high-techs is far from exhausted. Companies are still resisting the massive and systematic use of AI technologies where they can play a prominent role in optimising processes, gaining insights and continuously finding patterns (AI augmented analytics [25]). The new digital revolution will come from these AI technologies if they are used on a massive scale in our daily lives.

But how can I accelerate innovation and digital transformation in companies without the problems mentioned above? The answer lies in the use of cognitive AI services in the cloud [26], which are in constant development, are ready to use as tools, require minimal effort to set up and can be easily reused in similar cases. AI-cognitive services in the cloud make it possible to put the most advanced AI technology at the service of the digitalisation and optimisation of companies. They enable the industrialisation of this type of AI for the entire company and allow a profound digital transformation at all levels. It eliminates resistance and poor processes. They automate and simplify workflows and data processing. But mostly they perform repetitive and boring human cognitive tasks. The widespread use of these technologies in all areas will make companies more efficient and thus reduce their expenses. If there is a process that can be optimised or automated, companies will have no choice but to use these technologies if they want to achieve a complete digital transformation and be a data-driven company.

(Azure: https://azure.microsoft.com/en-us/products/cognitive-services/(opens in new tab))

In the case of Azure Cognitive Services [27], we can currently distinguish five main categories of services. Those related to speech, language and vision and making intelligent decisions, as well as an additional category currently in preview that will enable organisations to use the latest generation of large-scale AI models. These services use AI models pre-trained with big data, which in turn can be customised for our specific use by adding our own data with minimal effort. In addition, cognitive AI services can be combined with each other or with other cloud services to develop solutions that are fully customised to our needs. And most importantly, they enable the reuse of systems in other similar processes and the training of the same data source for other use cases.

The cognitive AI services currently available in Azure by category are:

Speech: Speech to text, text to speech, speech translation, speaker recognition
Language: entity recognition. Mood analysis. Answering questions. Understanding conversational language. Translator.
Vision: computer vision, customised vision, face API
Decision: Anomaly detector, content moderator, personaliser
OpenAI service

As already mentioned, these services are under constant development and expansion. (Further up-to-date information can be found in the official documentation: [28])

One of our recent cases of using cognitive AI services, which has been quite a success story, has to do with automatically reporting the use of construction machinery in the railway sector. The client stored video footage from all their surveillance cameras from more than 200 different locations. Video footage was only used for monitoring and security of the infrastructure

Thanks to storage technologies and cognitive AI services for machine vision in the Azure cloud, we were able to implement an automated system that not only recognises the machine type and operating hours, but can even report the model of the machine with an accuracy of over 95%.

The system avoids having to manually report the use of construction machinery, making it more efficient and precise compared to humans.

The same data source was later reused to warn of human presence in dangerous areas with minimal additional development effort.

In other cases, we have used these cognitive AI technologies to automatically classify legal documents. For another company, we were able to perform a PoC (proof of concept) that demonstrated that the identification and retrieval of manually filled fields in CC forms was possible. Even though the models were very accurate, it was recommended to avoid the use of analogue information in our proposed solutions where possible.

About Swisscom Data & Analytics

Swisscom Data & Analytics supports business customers in the consulting, design, integration and maintenance of analytical information systems such as data lakes, data warehouses, dashboards, reporting and ML/AI solutions based on selected technologies from Microsoft, AWS, SAP, Open Source and more. More than 50 dedicated data and analytics experts support our customers in various industries on a daily basis to turn them into true data-driven organisations.

About the author

Sergio Jimenez is a Senior Data & Analytics Consultant at Swisscom, specialising in Advanced Analytics. Since joining Swisscom in 2016, Sergio has worked on numerous projects for several clients ranging from Business Intelligence to AI/ML. He has successfully developed innovative solutions using the latest technologies.