Aicadium’s AI-driven computer vision is building strong use cases for workplace safety, visual inspection, and productivity.

When we think about AI-powered industrial computer vision solutions for enterprises, Aicadium which is part of the Temasek portfolio ecosystem, may come to mind. Its technology leverages surveillance cameras that are already installed in work/industrial environments, and then layer on top Artificial Intelligence (AI) and computer vision capabilities, so that the organization can gather invaluable workplace insights that address domains like workplace safety, productivity enhancement, and visual inspection.

Now, the market may already know what surveillance cameras are capable of, but what happens when AI technology enters the mix?

Phoebe Poon, Product VP, explained, “So, when we consider the fusion of computer vision with AI, it’s going to bring an evolution in terms of capabilities and outcomes.”

Computer vision systems already have the objective of detecting things, patterns, anomalies, and so on, so she pointed out that when AI is infused to it, learning and adaptation happen.

Learning quality and compliance

She illustrated the use case of food production in industrial kitchens. “If you have a video camera looking at the food items, you need to detect what items are being packaged.”

Instead of having images ‘hardcoded’ into the software, AI would be able to quickly learn what it needs to detect, from just a few examples. This is because the cameras are integrated with image recognition algorithms that have been trained on pre-trained data sets about various “components” of the workflow process, for example personal safety, packaging variations, quality standards, and different food types.

In effect, these AI-powered cameras increase the accuracy, efficiency, and quality of the final food product. It will also monitor for compliance to safety protocols, by flagging any misplaced items on the kitchen floor, and alerting when foreign objects are detected in food.

AI-based computer vision is so powerful because it is able to learn from data, adapt, and continuously improve its performance over time. 

Operationalizing AI in an organization

Its flagship product, Aicadium View, is a software-based solution so Aicadium partners with hardware players and would deploy capabilities even up to the edge ie. in close proximity to where surveillance cameras are located. 

In essence, the company takes an end-to-end approach when it comes to deploying AI in an organization.

“We firmly believe that in order to operationalize AI, it’s not just about building that first model and handing it to a client,” Phoebe said.  Rather, it requires being with the client throughout the entire process, from the first algorithm model, and continuously improving upon it.

Going the mile with clients as Phoebe puts it, involves answering questions like how to collect additional data to improve algorithms, how to retrieve results and explain them, and even how to see what data went into the training of the model.

The objective is to enable clients to monitor and measure model performance, detect changes over time, and if required, upgrade models to newer technologies as they become available to ensure they continue running better and faster.

Aspects like explainability, data governance and transparency around model decisions, especially for mission-critical use cases with high-stakes, are very important elements that Aicadium are looking to “bake” into their product offering.

“So, if a client asks us to explain the model (the data that went into it), we will be able to.”

In fact, it is during these high-stakes mission critical use cases that Phoebe finds that humans cannot be taken out of the process. “So, the AI in this case is about how it helps humans make better decisions, because machines can crunch data much faster.”

Research and the future

Large vision language models (LVLMs) are definitely a trend in the industry, and Phoebe added, “I think in the last year we have seen what large language models can do and there is already a lot of active research going on which are very quickly making its way into large vision models (also).

AI-powered cameras increase the accuracy, efficiency, and quality of the final food product. It will also monitor for compliance to safety protocols, by flagging any misplaced items on the kitchen floor, and alerting when foreign objects are detected in food.

“We are also very excited about multi-modality, which is the intersection of different modes of language and vision, and how that can be very beneficial for our client’s needs.”

Phoebe recognizes the importance of data availability, so new technologies or new techniques that help clients get to their first algorithm models with less data, or that helps clients generate synthetic data to quickly create use cases, are welcomed.

These, as well as the use of language for model explainability, and also to help search through large volumes of data, have become key areas in Aicadium’s product roadmap.

Aicadium was founded as an AI centre of excellence (CoE) in 2021 by Temasek, and which Phoebe expressed wants to be able to do the right business. I think every company will find pockets where AI will work for them. But, we have to scope the use case.

“Ultimately, it is important to provide honest and transparent advisory to Temasek portfolio companies, and this could even mean advising against using artificial intelligence for certain use cases that are not suitable or which would have limitations.”