What’s Possible with AI: Understanding Its Scope and Limitations

The hype surrounding artificial intelligence (AI) is tremendous. The media portrays AI as being better than humans in everything from playing chess and driving to writing poems. This notion often leads to misconceptions and confusion in the marketplace as to what AI is and what it can reasonably accomplish on the production line. The gap between what AI can do in theory and what it can deliver in practice is especially evident in manufacturing.  

“Demos are awesome. They show what is possible with AI in the future,” says Eric Danziger, co-founder and CEO at Invisible AI. “But there’s a large gap between a really cool demo and what is ready and scalable for the manufacturing line.”

Danziger works with automotive OEMs and tier-one automotive manufacturers to solve real-world challenges on the production line. He describes manufacturing as a “complicated beast” where it takes thousands of people and machines to build a vehicle. Everything and everyone must work together to achieve sufficient quality and efficiency while maintaining a profit margin. 

While it may seem like AI can do anything, AI is not magic. Danziger puts it this way: “Computer vision doesn’t have x-ray vision. If a human can’t see it, computer vision can’t see it either.” 

So, What is Possible with AI?

Flexibility

Traditional automation and robotics systems have long been powerful — but rigid. “Today, automation and robotics technology are not very flexible,” notes Torsten Kroeger, chief science officer at Intrinsic. “AI brings the flexibility to adapt faster.” That shift toward adaptability is one of the most significant transformations underway in modern industrial processes.

One emerging approach involves using AI-enabled development environments that allow manufacturers to design, simulate, and deploy robotic systems with far less friction than before. For example, some platforms now allow engineers to begin with a digital twin of the robotic work cell — an interactive simulation of the real production environment.

For example, with Intrinsic’s Flowstate platform, users can construct behavior trees made up of reusable “skills,” each representing a modular unit of robotic capability — instead of writing complex control code line by line. Companies with deeper technical expertise can still create custom skills, but the barrier to entry is dramatically lower.

Motion planning is a good example. In traditional automation, motion paths are manually programmed and rarely revisited. With AI-based motion planning, the system can automatically analyze how a robot moves and detect sources of inefficiency. Intrinsic says its tool, Flowstate, can improve cycle times up to 20% and lead to a rapid return on ROI.

Another major opportunity lies in simplifying perception. Instead of training custom vision models for every task, AI foundation models can now perform one-shot object and pose detection using only a CAD file — no lengthy training cycles required. Because these models generalize across lighting conditions, cameras, object types, and orientations, they dramatically reduce integration time and eliminate many of the traditional sources of friction in deploying robotic systems.

This efficiency means a single engineer can implement tasks that once required a team of specialists, making AI-driven automation more accessible and cost-effective across an organization.

To understand this shift, consider a typical pick-and-place application.

Historically, robots required a rigid, position-based approach: each object had to be located, identified, grasped, and moved following a carefully programmed sequence. A change in part geometry or orientation often meant rewriting large parts of the system.

AI-based skill frameworks enable a task-oriented approach instead. A perception skill identifies the part; a movement skill determines how to grasp and relocate it; and intelligent algorithms compute the motion automatically while avoiding collisions. If the part changes, simply uploading a new CAD model updates the behavior — no reprogramming required.

These capabilities are no longer limited to a few vendors. A growing ecosystem of companies now provide AI-powered motion planning, automated path optimization, and simulation tools for tasks such as pick-and-place, assembly, and material handling.

Bringing AI to Every Production Line

There’s a reason AI has typically been more likely to be implemented by larger manufacturers compared to small ones. First, they have the volume to make large investments worthwhile. Second, they tend to have sophisticated processes and they know, to a degree, where they are spending the most time and the most money in the manufacturing line. With this information, they can assess where automation will have the greatest positive impact AND they can estimate ROI. Being able to estimate ROI is critical to justifying the investment in automation.

Here is a key way AI is disrupting the manufacturing line. Today, managing a line is filled with estimates. Estimates of how many people will be needed, how long a process will take, how much downtime to expect. Sometimes these estimates are based on bad information or assumptions.

In one sense, AI is about collecting data and using it to improve decision-making. With sensors, cameras, and computers, manufacturers can track real-world operations at scale. Knowing exactly how long a process will take, even down to the impact of an individual operator, enables manufacturers to plan better.

However, the information AI gathers and assesses can be used for much more than just tracking progress. When there is more data available to assess, not just what an inspector with a clipboard can gather, patterns emerge. For example, a pattern might arise where defect rates are higher at a certain time of day. Troubleshooters can now focus their efforts to identify the cause. Is it because of rising temperature during the day, a particular operator, or a dependency on another process? AI helps assess what is working and what is not working. AI can even improve worker safety.

Efficiency and Worker Safety

Efficiency is key to a healthy bottom line in manufacturing. So is worker safety. When workers get hurt, the consequences go beyond medical bills. There’s the pain and disruption of life, loss of experience and skill, and impact on the work environment. With hiring already a challenge, even one injury can slow production and strain the workforce. Thus, it is in a manufacturer’s best interest to maintain safety standards. This is one area where AI isn’t just a tool for efficiency, it can be a literal lifesaver.

Consider the challenge of identifying ergonomic issues such as repetitive motion injuries (common) and outright unsafe behavior (rare). Traditionally, safety teams are small and must support huge numbers of workers. They are limited in terms of the data they can collect and the scale with which they can propagate improvements in processes out to workers. 

Intelligent Cameras

Invisible AI produces intelligent AI edge devices with 3D cameras, effectively, intelligent computer vision (CV) devices, give manufacturers many more eyes on their operations. Currently, if a worker is hurt, the safety team must reactively investigate the accident to figure out what went wrong. This can be extremely difficult with limited data about what really happened.

AI+CV makes it possible to have eyes on workers in a scalable manner. The goal is not to force people to work faster at an uncomfortable or dangerous rate but rather help them do their jobs better, safer, and with less effort. Specifically, an intelligent camera can collect enough real-time data to positively impact:

ROI Calculator

Discover the potential cost savings of robotic automation over a 20-year system life

This calculator compares your current manual labor costs against the total cost of owning and operating a robotic system over its 20-year lifespan.

EXPLORE TODAY

Efficiency: Many tasks must be performed by a human because they are complex in nature. AI can assess the way that workers perform these tasks and identify inefficiencies such as unnecessary movement. This leads to improvements in task completion by making it easier for people to do the things they already do.

Endurance: Many skilled labor jobs require a degree of athleticism and energy. Helping workers conserve their energy and be safer creates a more sustainable work environment so workers don’t burn out in a few years. 

Preventative Safety: AI can track a person’s movements to assess whether they are moving in a safe manner. For example, Figure 2 shows how AI can identify when a person is performing a task outside safe parameters, i.e., leaning over too far, lifting with the back, etc. Workers can then be given direct feedback and training to teach them how to take better care of themselves. An intelligent camera can also provide real-time feedback to identify potential issues and prevent accidents before they can occur.

Similarly, AI may reveal that workers are struggling to meet aggressive production schedules, the perfect storm for an accident waiting to happen. With the right data, manufacturers can adjust timelines to be realistic and achievable, ultimately leading to higher productivity by avoiding costly injuries and downtime.

Figure 2: AI can identify when a person’s position or posture is outside safe parameters.
(Source: Invisible AI)

Forensic Safety: When an accident occurs on the line, one or more intelligent cameras will capture the event. Without cameras, the safety team must make assumptions and try to determine what happened. If they are wrong, changing processes could put workers at higher risk, not lower risk, of injury. With AI+CV, the safety team can quickly and accurately determine the cause of the accident. Awareness of real-world causes is critical to being able to address and prevent similar accidents from happening in the future. 

Continuous Improvement: With access to more, relevant real-time data, manufacturers can make better decisions. And when improvements are uncovered, AI can help implement modifications down the line. For example, consider a worker injury that reveals workers are leaning too far over their workspace for long periods of time. Intelligent cameras can be updated to assess when workers are leaning for an extended period and issue a reminder to correct their posture.

Upscaling Work: An important aspect of improvements through AI is that people are very much part of the process. AI can identify issues that need to be addressed, but it is the workers themselves who can provide the expertise needed to develop the actual improvements. “AI helps create a high performance environment,” said Danziger. “It upscales the work people can do, making them more productive by enabling each person to do more.”

Return on Investment

“It is technically possible to automate an entire production line,” says Intrinsic’s Kroeger. “The question is whether it is economical and feasible.” The reality is, what manufacturers care about is uptime, quality, cycle time, safety, cost, etc. If AI can achieve this, great. If it can’t, manufacturers will wait until it can.

The real consideration is whether automating certain tasks is worth the price. After all, where’s the payoff in using robots if it slows production down and/or raises costs. There may be other compromises to consider as well. For example, fully automating a car manufacturing line might require changes to the shape of the vehicle to accommodate the limitations of robots. The adaptability/flexibility of automation equipment is another factor for ROI. AI technology is changing significantly every six months, and systems that are too rigid or fixed in functionality may be outdated before they can earn back their investment.

For AI to make sense, it must add value. Consider the cost in terms of personnel and equipment to manually observe a worker to determine and then enforce proper ergonomics day-to-day. Intelligent cameras provide many more eyes on the manufacturing floor, enabling a person to effectively watch the entire production line so nothing is missed. With AI, even small safety teams can collect substantially more data, evaluate every worker, and then prioritize those issues which will yield the highest benefit. According to Invisible AI, manufacturers are seeing ROI for AI edge devices in less than six months.

Simplicity in training, deployment, and operation is crucial as well. For example, many AI systems require an AI expert to capture and train models with thousands of examples of a process. Furthermore, cells that work in conjunction with each other may also require updating.

Using AI, it becomes possible to update a production line much more quickly, depending upon the change being implemented. For example, Invisible AI devices are pre-trained on human and vehicle motion. Thus, they only require a single example of a process to be able to start analyzing work. 

Speed-to-update is critical since, as more data is collected, more improvements are possible. The faster data can be leveraged, the faster the ROI. If it takes three months to implement an update, this limits the rate of improvement and requires additional investment with each update.

With AI, improvements can be implemented as they are uncovered, resulting in faster and continuous improvement. Similarly, reconfiguring existing assets into new lines can be done faster, making it easier to repurpose a robot to perform a new task. Depending upon the change, it could be as little as a few hours to change a production line. This allows manufacturers to benefit from even small improvements immediately.

The biggest change that AI brings is how it can extend flexibility into a production line. Without AI, automation was limited to specific use cases that could justify the expense of design, training, and deployment. With the increasing simplicity of using AI, as well as the lower investment required, every month AI is feasible in an increasing number of use cases.

Possible vs. Feasible

So, what’s the role of AI in manufacturing? AI helps humans by performing certain repetitive tasks so people don’t have to. For complex jobs that require a person to perform, AI can help make a safer and more efficient workplace. This frees people up to perform more value-added and engaging work. 

Where people get into trouble with AI is when their expectations are unrealistic. The question to ask is not what is possible but what is feasible. AI can do many incredible things, just not always at an affordable cost. Building a super AI system that can do everything end-to-end is expensive and complex to design, deploy, and use. It’s also unrealistic to expect to push a button and have everything just work. Manufacturing can be complicated. People need to be trained to perform assembly jobs. So do robots. 

Ultimately, manufacturers don’t need a robot that can do everything. They need automation equipment that can handle a given set of tasks. Frankly, no one wants to pay for functionality they don’t need (i.e., right-sizing). With a task-focused approach, manufacturers can focus on investing in AI where it will have the greatest impact for them.