Find out more about Calligo’s Machine Learning as a Service (MLaaS) capability here

One thing’s for sure: in 2023, Machine Learning will penetrate even more businesses in even more sectors, following tremendous growth last year. But when it comes to ML, what should organizations have their eyes on, to stay ahead of their competitors? Here Calligo’s Chief Data Scientist, Tessa Jones, reveals her five Machine Learning predictions for the year ahead, based on her experience at the front line of helping businesses use ML and AI to unleash the value of their data and boost ROI.

Top five Machine Learning trends

1. Predictive systems

The evolution of Machine Learning and Artificial Intelligence has reached a point where businesses need to adopt intelligent solutions to remain competitive – and therefore, I anticipate a significant rise in the use of predictive systems this year.

Predictive analytics encompasses a variety of statistical techniques – including Machine Learning, statistical modelling, and data mining. This combined power gives organizations the ability to extract precise, forward-looking intelligence from their data efficiently. Predictions can become rooted in business decisions, giving deep insight into the future. It’s about moving from being ‘data inspired’ – where decision-makers use data to feel better about doing what they were going to do anyway – to obtaining new information that allows you to make data-driven steps forward.

A competitor that’s adopting dynamic, sophisticated ML solutions is going to have an edge over a business that’s still relying on legacy systems – with a dose of gut feel.

Business consideration: How do we truly get value out of our data? Do we want to validate a decision we already know we want to make, or apply intelligence onto it and discover fresh actionable insight?

2. ML & AI regulation

More regulation and formalized audit processes will continue to emerge in response to increased attention to data privacy and ethical AI. Dip into Calligo’s latest ‘Data Privacy Periodic Table’ to see the constantly moving pieces of the world’s data legislation jigsaw. Not to mention whether data privacy and AI is even possible. For more on that, do read a thought-provoking blog on the topic by my colleague, ethics and governance expert, Sophie Chase Borthwick.

Clearly, ML and AI regulation will continue to be a minefield; one with shifting parameters depending on geographical location. On the face of it, the EU is further advanced in that it has regulation in place; this will continue to evolve. For example, how should businesses carry out ML and AI audits and risk assessments? Importantly, in a way that doesn’t become too much of a hindrance for organizations to adhere to.

Whereas the US is still grappling with what ML and AI regulation might look like. There has been some headway in the privacy space and there’s a new law in New York that requires recruiters using AI tools to go through a robust AI audit to rule out bias. But that’s just one niche, in just one state. It’s a step in the right direction, but there’s clearly a long way to go. And of course, in our global world, businesses don’t operate just within state borders or even countries – nor does technology. If there’s a deluge of different ethical AI regulations being introduced in different parts of the world, things could get even more foggy than they are now.

Business consideration: Have you considered how your ML and AI could be audited? And do you have the right privacy, governance, and ethical AI expertise in place to manage your risk?

3. Real-time vs batch

I predict a surge in the use of real-time prediction models this year, that are integrated into live systems, so they deliver immediate results. This can often be more valuable than batch predictions – and that’s why I anticipate greater adoption this year. In my mind, the businesses that are investing in real-time ML will be the movers and shakers.

Batch predictions have certainly played a significant role. A model gives businesses insights on a weekly or daily basis. This is useful when you want to generate predictions for a set of observations all at once. But many use cases suffer from lagging predictions; data will be left waiting until the next batch to be processed. And it involves a fair amount of human graft.

In contrast, real-time inference allows the model to make predictions at any time and trigger an immediate response. Historically, this has been the domain of the big tech companies, but I see this becoming more mainstream in the near future.

Business consideration: Is your organization ready to optimize ML systems when real-time predictions are superior to batch predictions? The natural progression from batch to real-time predictions can unlock huge rewards.

4. MLOps improvements

This brings me seamlessly on to my next (not real-time or batch) prediction. Businesses can’t start harnessing the power of these live systems until they have robust ML operations (MLOps) in place – and by that I mean processes and people. The uncomfortable truth is Machine Learning is difficult to make useful, carries a high risk of failure and is hugely costly when things go wrong.

Tech is all well and good, but to deploy predictive real-time models effectively, I believe there are at least six distinct roles needed on top of the heavy lifting technology. This can surprise businesses who think (hope) that investing in one data scientist will suffice. Read more here about the different MLOps roles needed – and how this de-risks ML for businesses, while boosting productivity and profitability.

There’s an exciting new trend emerging of employing an army of MLOps experts only for the time they’re needed. And it’s called Machine Learning as a Service – like a timeshare apartment that’s rented only for the weeks you want it, rather than paying for the luxury all year round.

Business consideration:
Are you aware that hiring one or two data scientists is not enough if you want ML solutions to boost productivity and profitability? You need an end-to-end data science team, consisting of at least six unique skill sets.

5. Edge AI & Federated Learning

As touched on in trend 2, the issue of privacy will become more and more prominent, as predictive models become more widespread – and this is where Edge AI and Federated Learning will increasingly take center stage.

As a quick reminder, Federated Learning is an ML technique that involves training an algorithm across several decentralized edge devices. Each trained model is then pooled centrally, modified (based on the specifications of the other models), then returned back to its local system. This effectively allows models to be informed by data that it never had access to, thereby supporting data privacy.

Edge AI meanwhile is the execution of already trained AI applications on decentralized devices, like a phone or off-line machine. This approach helps overcome issues related to latency, bandwidth, and privacy.

Business consideration: While Edge AI and Federated Learning are only relevant for select use cases and industries, it’s important for all organizations to be aware of new emerging tech that allows sophisticated systems to be built with higher levels of privacy.

Find out more about Calligo’s Machine Learning as a Service (MLaaS) capability.