HypeCycle for AI 2021

HypeCycle for AI 2021

Every year, I look forward to the updated HypeCycle from Gartner. The Gartner team surveys a large number of companies about what they are using and what they plan to use, from which they can discern certain trends. Predicting the future is exceptionally challenging. When I attended their conference a few years ago, they proudly predicted that today people would communicate more with chatbots than with real humans. Fortunately, that did not come to pass. Chatbots today find themselves in the so-called valley of disillusionment, and practice has already revealed the weaknesses of traditional tree-based chatbots.

What trends do the Gartner analysts see in AI today?

1. Operationalisation of AI – today, it takes the average organisation 8 months to deploy a finished model into a production environment. This is far too slow, and thus most companies are contemplating how to shorten this time through better architecture. One solution is ModelOps – which reduces this time and can also include a system for managing and overseeing the entire AI lifecycle.

2. Efficient use of data, models, and computations. This group includes my favourite composite (hybrid) models. That is, the combination of powerful models, typically deep neural networks, with something that is easily explainable, such as an expert system. This also encompasses generative AI, which enhances datasets with generated synthetic data.

3. Responsible AI – AI is increasingly assisting people in decision-making, and with that comes a growing emphasis on reducing bias. There is frequent discussion about discrimination based on race, gender, age, the neighbourhood in which you live, and so on. In the EU and the USA, there is an ever-stronger focus on fairness, transparency, security, and privacy.

4. Data – increasing attention is being directed towards new analytical techniques known as "small data" and "wide data." This concerns how to utilise the data we have more effectively. For example, how can we predict the course of an epidemic when we only have a short time series (small data)? How can we extract more information from various unstructured and diverse datasets (wide data)?

Sources:

• https://www.gartner.com/en/articles/the-4-trends-that-prevail-on-the-gartner-hype-cycle-for-ai-2021
• https://signum.ai/blog/small-and-wide-data-is-important-and--relevant-is-the-era-of-big-data-coming-to-an-end/
• https://en.wikipedia.org/wiki/ModelOps

Originally published on Facebook — link to post