These AI trends will accompany us in 2021
This year everything will be better. Hopefully. Although many companies are facing major challenges as a result of the Corona crisis, technological trends are not standing still. In fact, the opposite is true. There is a digitization push that – more than ever – makes the use of artificial intelligence possible and valuable. Business processes are being digitized faster than expected. From the assembly line to strategic management of a business, data is emerging where there was none before – the requirements for AI have never been better. Find out which trends will become important in 2021 and how strategy work will benefit from them.
Now that many companies have already tested the application of machine learning and developed valuable solutions to use cases, it's time to take the step out of the lab. Bringing an ML system into practical use in a stable and secure way is challenging. It starts with resources: If a Data Scientist, an exported dataset and domain expertise are sufficient to realize a proof-of-concept project, the successful and sustainable implementation of ML solutions requires an entire data team. To build a reliable system, ML operations, data engineering, and data science have to collaborate. Due to the complexity, AI platforms are required that manage the necessary infrastructure and thus simplify topics such as cluster management and model lifecycle management.
We can also look forward to progress in the field of Auto-ML. In Automated Machine Learning, one AI is trained by another AI – sounds like science fiction, but it's already a reality. In late 2020, Google developed a fully automated time series forecasting tool that beats an average of 92 % of models created by Data Scientists in Competitions. Many companies use forecasts for sales, market volumes, etc. in strategic planning. With AI tools like Auto-ML, they can facilitate their forecasting work and easily detect outliers, input errors or eye-catching forecasts in the aggregation of many individual forecasts. AI can therefore reduce risk in strategic planning.
Robotic Process Automation (RPA) plays a similar role. RPA is a technology that makes repetitive processes within existing software systems automatable by "robots". These robots imitate human behavior. They log into software systems, move files, perform actions. The potential is huge and still far from being fully realized. 2021 will show whether expectations are fulfilled, and the potential can be tapped.
In 2020, there was a lot of buzz about GPT-3, an OpenAI trained language model that broke a barrier. Previously, it was common to train large language models on massive amounts of data and adapt them to a specific task through transfer learning. GPT-3 is different. It can program web pages, invent stories, discuss, translate, etc. without specific fine-tuning. With the massive size of the model (175 billion parameters ~ 700GB), it does not fit on any commercial server. Just maintaining the computing infrastructure is extremely expensive and demanding. It will be exciting to see how it will continue with superlatives on the way to artificial intelligence worthy of its name.
On the other hand, more and more language models are being used in practice. Google's search engine uses the BERT model to display semantically more appropriate results. The use of language models in strategy work is very valuable because a lot of content is collected via text, such as market trends or SWOT analyses. Texts from the entire company can thus be captured and highly aggregated bottom-up via text clustering. This makes it easier to maintain an overview of which trends, moods and opinions prevail in the company during strategy development.
The role of humans in the ML process is increasingly shifting. The more operational tasks are automated, the greater his (social) responsibility becomes.
Examples from the recent past show how important it is to be aware of the societal challenges posed by AI:
As artificial intelligence penetrates more and more areas of life, the issues of ethics, algorithmic fairness, and regulation of AI become more pressing.
Does an AI solution have to explain and justify its decision? That depends on the expected consequence of the decision: If I play against a chess computer, I don't expect an explanation of the move. If WhatsApp suggests the next word to me while I'm typing, I don't ask why. The situation is different when it comes to the future viability of a company or personal health. If the AI of the strategic management system detects a large deviation between budgeted and actually expected company development, this must be comprehensible.
The issue of explainability represents a fundamental weakness of machine learning and – due to the sheer increase in deployed AI tools – is moving more into the spotlight.
In addition to these trends, AI is increasingly becoming an integral part of many software products. Specialized functions are emerging that are oriented to the specific use cases of the respective products.
This is the case with Evolutionizer's Solyp 4.0, which covers all aspects of strategic management in the enterprise. In this context, AI functions are used to integrate the knowledge of many in the company in strategy development and to identify opportunities and risks at an early stage in strategy implementation.
Your Florian Dreher
