Donato Montanari, Vice President and General Manager, Machine Vision, Zebra Technologies, believes that there is simply no more time for hype when it comes to AI. 
When people talk about “AI hype,” we really need to ask what sort of AI they’re talking about – generative AI (Gen AI) is new, while other types have been around for decades.
Some types of AI are familiar while others obscure or in the future, like artificial general intelligence – AI systems with broad, generalised intellectual capacities (abstract thinking, common sense) exceeding those of a human, although definitions vary.
A decade ago, certain types of AI were a curious technology that researchers and enterprises were using to play and win games against humans as well as achieving some important advances along the way. With new Gen AI technologies such as ChatGPT, GitHub Copilot for developers, Sora and Midjourney, everyone is now talking about AI and enterprises are utilising these new capabilities.
But there is also a world of difference when it comes to consumer versus enterprise use of AI. Budgets, security, use cases, return on investment and expertise needed to reach new levels in business, leaving little room or time to admire the hype. Measurable productivity, process efficiency, workforce optimisation, cost savings and revenue growth are top of mind.
Business executives rank AI and Gen AI among the top three tech priorities for 2024, yet 66% of leaders are ambivalent or dissatisfied with their progress on AI and Gen AI—and only 6% have begun upskilling in a meaningful way, says a new report by BCG. Yet 54% of leaders expect AI to deliver cost savings in 2024. Of those, roughly half anticipate cost savings in excess of 10%, primarily through productivity gains in operations, customer service, and IT.
We can dig deeper into specific industries too. For example, 43% of automotive business leaders surveyed in Germany and 56% in the UK are currently using some form of AI such as deep learning in their machine vision projects, says this Zebra report. However, 34% in Germany and 24% in the UK say there are not using any form of AI, such as deep learning, in their machine vision projects and don’t see the relevance. The picture is mixed, and there are others using AI but want it to do more and do better.
Advances in machine vision is one example where we see manufacturers, like other industries, at different levels of maturity when it comes to getting results from AI. Modern machine vision unlocks new levels of analysis, accuracy, compliance and quality in production processes, and it provides engineers with new tools to work more efficiently – the sorts of things that cut through hype and show the true value of technology.
For example, at one production site, the Bosch Group develops solutions for diesel engine injection systems for the automotive industry. The injection nozzles are an important component that transports the diesel fuel into the combustion chamber of the engine. Bosch needed an imaging solution to further automate reading and verification processes, improve the traceability of injection nozzles, and reduce the number of machined parts needing to be checked manually.
With its machine vision system, the plant achieves a production volume of 7,000 parts per day. The proportion of incorrect rejects has been reduced to less than 5%, which is a significant improvement. The system runs on machine vision software, which controls the entire system and has enabled the team to reduce costs and set-up time and simplify installation.
AI (and Users) Need Training
Deep learning neural networks are powerful, advanced AI tools that mimic the human brain (specifically, convolutional neural networks in the case of machine vision, where connectivity is inspired by the brain’s visual cortex which processes images), but they are not magical. Sometimes an engineering team expects them to perform flawlessly.
It’s important to educate stakeholders on the capabilities and limitations of neural networks. Neural networks can achieve remarkable results, but they need to be applied in an educated way.
Realistic expectations should be based on the areas where neural networks excel (compared to human performance and conventional rules-based machine vision) such as detection of surface defects, detecting or counting objects, reading difficult characters or detecting unexpected deviations from previously seen objects (anomalies).
Selecting the appropriate evaluation metrics is essential for accurately assessing model performance. The most basic metric is accuracy (number of correct classifications divided by the number of all classifications), but it may not be suitable for unbalanced datasets. Instead, metrics like F1 score for classification or average precision (area under precision-recall curve – AUPRC) for detection tasks.
Metrics like area under the receiver operating characteristic (AUROC) that rely on true negatives should be avoided, as the numbers they produce may be misleading (too optimistic), especially when the number of true negatives is very high.
There are also many data issues that need to be addressed in order for a business to reap the benefits of AI. Mixing training and testing datasets, inadequate and unbalanced same sizes, ambiguous and inconsistent data annotation, and environmental factors need to be taken into account to ensure deep learning solutions work properly.
Approach AI With a Plan
Businesses also need guidance to cut through the hype to secure AI-driven value for their operations following the passing of the EU’s AI Act. The EU AI Act sets out a common framework for the use and supply of AI systems in the EU, along with a classification for AI systems with different requirements and obligations tailored on a risk-based approach.
The act provides a new catalyst for manufacturers to invest in the partnerships and technology needed to make digital factories and smart manufacturing operations a reality. More automated and autonomous workflows, better supported workers, and predictive and prescriptive analytics can be leveraged with AI and the mountains of valuable manufacturing data available.
Which manufacturing process is in need of automation and would benefit from AI? Which type of AI would be most appropriate? How is legal compliance ensured and recorded, who do you need in terms of staff and partners to make it happen? These are the sorts of questions that hype does not answer, but they must be addressed.
Advances Without the Hype
Right now, it’s not so much AI simply creating and eliminating jobs, although many headlines may suggest this. As with the car, telephone, and internet, hosts of new jobs and industries will spring up thanks to the growth of AI. What we see now are manufacturers equipping their engineers, programmers and data scientists with new and better tools with AI to do what they’re doing but faster, more efficiently, and handing over certain tasks to AI-driven automation.
Manufacturers, like other industries, can be challenged by labour hiring, training, and retaining. In these cases, business leaders are turning to automation to fill labour gaps, train workers faster, and assist the workforce they currently have. Workers with AI capabilities will set themselves apart, as they’ll have the knowledge and expertise manufacturers want in their plants.
Companies will also prioritise the democratisation of AI and machine learning as a strategic priority. Whether an engineer, data scientist or developer, workforces will be upskilled and given learning resources and readymade, easier to use AI tools that take on some tasks and support workers in other areas of their role. Some tools, like deep learning OCR – can be low/no code, meaning they’re ready out-of-the-box and don’t require specialist training.
Other tools are higher end and operate more like readymade environments for programmers and data scientists to create solutions using the platform, tools and libraries provided.
Eventually, this approach will be standard rather than a differentiator in the battle for talent, skilling the workforce, and optimising the front line with new ways of working. Those who can introduce and use new AI tools without falling for the hype today will put themselves and their customers at an advantage tomorrow.
