printer friendly version

Surveillance & AI roundtable

SMART Security Solutions held an online roundtable with a few surveillance experts to delve into the intersection of surveillance and AI, gaining insight into the market and how control rooms are evolving. Are operators on the way out? Can AI do everything? Is AI just traditional video analytics on steroids?

Image credit: Gemini.

The focus was, understandably, on agentic AI, its emerging role and definition, and the trustworthiness of its decision-making abilities. Currently, AI agents’ decisions are still verified by humans, but for how much longer? To listen to or view the round table video:

Find the video at https://youtu.be/h_F636qXrMs

Find the audio at https://www.securitysa.com/ex/SSAI_RT_26.mp3

Joining us were Gerhard Furter, Jean-Vicente De Carvalho, Dr Jasper Horrell, and Dave Keating. SMART Security’s editor, Andrew Seldon, asked the panel to introduce themselves and their companies.

Gerhard Furter is from Iris AI, a company that has been working with AI for 25 years and currently serves large portions of the retail and residential markets. He notes that the company’s monitoring services are increasingly being engaged for operational monitoring in addition to security. Sometimes, he later said, operations is the main focus with security monitoring seen as a bonus.

Jean-Vicente De Carvalho, founder and CTO of Lytehouse, was up next, describing the company as a video intelligence platform including a rules-based analytics engine using deep learning models, a cloud VMS, and the Scout AI system, which uses agentic operators to triage alerts alongside human operators.

Jasper Horrell, founder and chief innovation officer at DeepAlert, stated that the business operates in southern Africa, the UK, Europe, Australia, and New Zealand. Its focus is primarily on false-alarm filtering in the security space, with increasing work on sophisticated use cases such as suspicious behaviour and PPE. DeepAlert is also actively developing agentic solutions.

Dave Keating, a co-founder and CEO of Refraime, explained that the company originated from Yellow Dog Software and aims to create a true African-based AI practice, noting that many first-world solutions fail in Africa due to requirements for sophisticated infrastructure and unlimited bandwidth. Refraime focuses on developing AI solutions in the African context, beginning with CCTV analytics as its first vertical, given the region’s security challenges.

What is real AI?

Keating raised an important point that there are still companies trying to pass traditional rules-based video surveillance off as AI. He proposes that potential buyers use a ‘litmus test’ to ask whether a system was programmed, taught, or learned a skill. Read more on this in his article here.

De Carvalho offered a slightly different view, suggesting that a rules-based engine does not necessarily exclude AI, provided it uses deep learning models. He notes that traditional computer vision, deep learning, and agentic AI each have their own valuable use cases, and that rules-based systems (using deep learning models) offer a lower-cost alternative compared to larger agentic models.

Horrell provided a perspective on different levels of AI, distinguishing rules-based or programmatic approaches from deep learning models and the new wave of LLM-based technology, which mimics reasoning and takes seemingly autonomous actions.

Furter supported this, noting the need to define AI and educate clients who mistakenly see “clever motion detection” as AI. He defines AI as “anything that you get from a camera that you did not put in in the first place,” distinguishing machine learning (like human or car detection) from “real AI,” which involves Vision-Language Models (VLMs) whose outputs can be more complex than the inputs.

Image credit: Gemini.

Kicking out the humans?

One of the pressing questions that has become all too real today is whether AI will replace humans. The general attitude in business is that there will be significant job losses - the IMF estimated that 300 million full-time jobs globally could be affected by AI-related automation (https://tinyurl.com/mr7xubw5). In our discussion, the jobs question was focused specifically on surveillance control rooms, and the risk to jobs is just as real. At the same time, agentic systems can add tremendous value.

Horrell highlighted that agentic systems can access and enrich data (including historical data), enabling them to potentially make better decisions than a human operator with limited access to information. De Carvalho emphasised that agents’ immediate value is absorbing “the 99% of garbage alerts” that operators currently handle, freeing up operators to manage a significantly higher number of cameras.

Keating raised a cautionary note, however, that for high-value activities like security, autonomy must involve collaboration “with a human in the loop to make final judgment calls”, especially given the current gap in trust. He also notes that some client requirements are simply not doable by AI alone.

While Furter agrees with human verification and the centrality of trust, he notes that on many of Iris AI’s sites, highly trusted detections (such as person detection) enable the AI to take autonomous steps. He estimates that about 40% of Iris AI’s sites run completely autonomously, with humans intervening only for complex or untrusted algorithms.

Since Refraime focuses on African solutions tailored to the continent’s realities, Keating noted that change management in the security sector is a significant issue, with some clients fearing job loss and reports of sabotage in large-scale deployments. A risk South African companies are well aware of.

It is not only about agents taking over jobs, however. De Carvalho stated that agents can increase profitability by maximising the number of cameras per operator, enabling control rooms to expand their scope. A goal of Lytehouse is to have the same number of operators in a control room managing 10x the number of cameras through the efficient deployment of AI agents.

Horrell provided a forward-looking perspective, suggesting that in the short term, technology will supercharge control rooms, allowing entrepreneurial ones to grow and develop specialised businesses. However, in the long term, technology will deliver value directly to the end customer, simplifying operations and health checks through agents, potentially leading to the shrinking and perhaps even the disappearance of many traditional control rooms.

Changing roles

This led to a discussion on the ethical and/or moral obligations of companies, with Keating emphasising the moral obligation of AI providers to consider job creation, especially in non-first-world countries. He mused about creating qualified communities for micro-tasks and decision-making, rather than traditional control room operators. However, the question remains whether the number of operators will be reduced, requiring a higher skill set to move from an operator to an analyst role. Education is key here, an area in which the South African government has failed dismally.

Furter notes that his company has a policy that AI will never cost someone their job. The focus is always to find a way to repurpose people, a principle the business has applied successfully for over 10 years.

Horrell agrees with the principles and risks, but says the problem is much broader, as AI can automate any job that involves following a workflow. The participants acknowledged that AI presents a significant challenge for society, extending beyond their industry to areas such as call centres and factories. “Wise and creative deployment” is a necessity.

Managing the expectation of miracles

The topic of educating clients was highlighted again, with participants noting the infeasibility of some client expectations. For example, Horrell noted the problematic market expectation of 100% accuracy at very low cost, which is unrealistic given environmental factors such as camera quality and network glitches. He suggested potentially offering customers different cost tiers that align with varying levels of accuracy, while explaining that perfect accuracy is not possible.

Furter agreed that client education helps manage expectations regarding AI results, but the problem of clients wanting an expensive solution for a low price point, often using old technology, persists. Keating compared this to the AI industry hitting the “trough of despair” in the hype cycle and slowly ascending the “slope of enlightenment.” However, the belief that AI can do everything at a low price point is not going away anytime soon.

As a result of expectations, the participants tend to focus on B2B markets rather than consumer clients, with some also engaging distributors to sell their services. The real market is still in supposedly simple areas like intrusion detection, but as noted, operations monitoring is becoming an important focus for everyone.

Looking ahead

When asked for their insights into the future of AI and agents in the surveillance market, nobody predicted an end-of-the-world scenario, but all expect the next couple of years to be a time of great change.

Keating predicted the emergence of niche AI agents controlled by an orchestrator, allowing for complex, multi-layered intelligence in a single alert, such as detecting violence, smoking, and safety threats simultaneously. Furter predicted that agents would be running sites, with human operators serving as verification, as agents become more autonomous and smarter.

Horrell anticipates rapid change, predicting that in two years there will be near fully accurate agentic offerings for various use cases, fundamentally changing the control room market and resulting in small, highly automated internal teams.

De Carvalho anticipates seeing the first two or three “super control rooms” within two years, managing 10 000 to 15 000 cameras with only three or four operator stations. He also expects a significant rise in video analytics providers, noting that new providers may lack the necessary domain experience, potentially leading to them getting ‘burnt’ due to the difficulty of maintaining cameras and managing high-risk events.

This brief summary of the round table does not do justice to the discussion held, but the full video and audio are available. Find the video at https://youtu.be/h_F636qXrMs, and the audio at https://www.securitysa.com/ex/SSAI_RT_26.mp3. SMART Security Solutions thanks our participants for their time in joining our round table.

For more information, contact:

• DeepAlert, +27 21 201 7111, [email protected], www.deepalert.ai

• Iris AI, +27 73 740 5900, [email protected], www.irisai.co.za

• Lytehouse, +27 66 113 5452, [email protected], www.lytehouse.io

• Refraime, +27 68 646 0492, [email protected], www.refraime.co.za

Share this article:

Further reading: