The magic of AI: the next leap
Success with Artificial Intelligence (AI) will increasingly depend on the ability of organizations to integrate knowledge by blending different AI techniques.
Although AI encompasses a range of technologies and capabilities, until now, much of its value within organizations has been generated from one category of AI called ‘supervized learning’. This can revolutionize speed and efficiency by automating a relatively simple mapping of A to B. So, for example, if A is a picture, then B is the name of the object shown in the picture (image recognition). Similarly, if A is an audio clip, then B is the text transcript (speech recognition); and if A is an email, then B is what kind of email it is (text recognition). These mappings are at the heart of successful AI applications such as predictive maintenance, IT servicing and support, customer services and marketing.
Toward unsupervized learning
To implement this kind of supervized learning, enterprises need access to plenty of computing power, which advances in computer architecture have brought in recent years. They also need large amounts of historical data in order to ‘train’ the AI to achieve high levels of accuracy. When it comes to industrializing AI, the challenge is that each sector – and each enterprise within a sector – has its own processes and terminologies. As a result, training AI to optimize a manufacturing or supply chain process, for example, can be challenging as the vast amounts of data required are not always available.
For this reason, to fully leverage the knowledge of an enterprise, it’s necessary to incorporate other types of AI, including what is sometimes called ‘old AI’ (based on symbols and logic programming), as well as deep learning, natural language processing, and other techniques. With richer datasets and technologies, AI becomes even smarter.
Enterprise Knowledge Integration
At Atos, we are pioneering Enterprise Knowledge Integration, which enables a dialogue between the different technologies in the AI family. The results from each technology are used to feed the others, creating a virtuous circle between supervised learning and symbolic AI. Progress with Enterprise Knowledge Integration requires strategic investment and focus in three domains:
- Optimizing the power of high performance computing and building the very large infrastructures capable of running more complex AI workloads.
- R&D programs with industry partners such as Siemens, to develop algorithms and models for highly demanding AI use cases (such as advanced video recognition and fraud management).
- Establishing a global network of AI labs, mixed teams of Atos specialists, partners and customers to collaborate on AI solutions, address key business issues, and add value to HR, the supply chain, marketing, contract management, finance, and other business areas.
Looking forward, via digital tools, AI will become more and more embedded, and invisible, within enterprises – just as we see with everyday consumer tools, particularly smartphones. At the same time, AI will become more knowledgeable about the data it is handling; for example, it currently may not have much intelligence about the value of the things it is recognizing (such as knowledge of the object in a picture). In the near future, we’ll see more and more semantic understanding of the objects that are processed by AI. In other words, AI will become more intelligent by merging datasets and continuously integrating its knowledge. These kinds of advances are already accelerating because while AI has been around for decades, we are, in many ways, just at the start of its story.
Digital Vision for AI
This article is part of the Atos Digital Vision for AI opinion paper. We explore the realities of AI and what’s ahead for organizations and society, as artificial intelligence advances fast as an enterprise solution.
