#Data language#data#AI

Forget the buzzwords: what will AI really look like in 2021?

AI is now a victim of its own hype - and when 2021 isn’t the year that AI goes mainstream after all, there will be plenty who are disappointed

Matt Shearer
|Dec 23|magazine10 min read

Machine learning and other artificial intelligence (AI) approaches are developing at pace, and we have seen rapid evolution and some impressive deployments over the last year or so. The problem for AI is that it’s now a victim of its own hype - and when 2021 isn’t the year that AI goes mainstream after all, there will be plenty who are disappointed. 

That’s not to say that there aren’t many exciting developments happening in this space - there absolutely are. But if you want to use AI and machine learning to make a real difference to your business, then there are two key things to keep an eye on. 

If you want to read about how intelligent humanoid robots will be replacing the entire workforce by 2025, then this isn’t the article for you. But, if you’re interested in realistic, pragmatic predictions for AI over the next 12 months, based on the biggest practical hurdles to rolling out AI in your business, then read on…

Prediction #1 - Organisations’ data architecture will come to the fore as the essential foundation for AI

Understanding will grow across the business community that in order for AI implementations to be successful, good knowledge management is a prerequisite. AI works only when it’s built on good information management practices. It cannot fix the absence of them. This is not a new concept, but until now it has been under-appreciated, particularly by senior decision makers without a technical background. 

Good knowledge management means having a clear model for your domain, and structuring data in a consistent and detailed way. Data that is well structured and well described is the only kind of data that can carry its context and meaning across different business functions and applications.The growing awareness of the importance of data architecture can be seen in the two following trends. 

First, the rise (again!) in the popularity of knowledge graph technology and the parallel rise in job openings in progressive organisations, such as Apple, that are seeking to leverage machine learning and AI. 

Second, the use of buzzwords like ‘data fabrics’. This may be a new term, but it’s not a new concept. Data fabrics frameworks are simply introducing more ways to emphasise that in order to manage distributed data functions effectively, good data architecture is a fundamental requirement. 

Prediction #2 - The emphasis on software engineering as an AI success factor will dramatically increase

Slowly, companies are starting to accept that simply hiring a PhD in data science is not enough to develop successful, AI-driven commercial software. Over the last three years, there has been a surge in machine learning ops (MLOps) that promise to automate software engineering tasks - removing the need for companies to hire software engineers. This should, in theory, have made these processes more efficient for businesses, reducing human error. As is so often the case in the intersection of business and technology, though, ‘in theory’ and ‘in practice’ have not yet aligned. 

In reality, MLOps often make projects more expensive and difficult for companies. This is because MLOps are developed for use by data scientists, and as such still do not address the challenges of productising machine learning code. Ultimately, data scientists cannot do everything - and delivering large-scale, AI-driven platforms too often falls outside of their area of expertise. In 2020, many enterprise-scale companies found themselves down this rabbit hole, and have had to throw money at the problem in an attempt to resolve it. 

In 2021, businesses will come to realise that they do, in fact, need to invest in software engineering to develop AI programmes that scale. As acquiring the skills needed to understand machine learning is significantly easier than learning software engineering, this will change the balance of hiring. Instead of searching for data scientists that have some limited coding skills, they will move towards hiring software engineers with some limited knowledge of AI and machine learning. 

Looking forward

There are many other trends ebbing and flowing in the AI space, around topics such as GANs, GPT3, no-code and others. These definitely add value if the challenge requires them, and it can’t be denied that they often add a ‘wow factor’ for investors, but organisations will increasingly realise that they cannot start their AI projects from a ‘wow-factor’ inspiration. This approach will be rightly recognised as “technology for technology’s sake”. 

In terms of practical delivery of innovative AI implementations that make a commercial difference, there are but two important tectonic trends in play: increased understanding of the importance of both strong data architecture and strong software engineering. 

By Matt Shearer, Product Director, Data Language 

Find out more
Read here