The Reality Behind Manufacturing’s AI Myths

As a former professor in artificial intelligence, one of my favorite – and surely one of the oldest – technological myths is found in the masterpiece, the Iliad. 

In Homer’s poem narrating the Trojan War, the God of metalworking, Hephaestus, engineers one of the first robots known to history, a handmaiden designed to assist him in his forge. Not happy with limiting himself to manufacturing, Hephaestus steps it up by designing Talos, an automated bronze giant whose purpose was to protect ancient Crete from pirates and invaders.  

While thousands of years have passed since Hephaestus’ mythical robots came to life, today’s intelligent machines – strong with skillful AI – are making headway in our own workplaces. Take the factories and warehouses adversely affected by the pandemic as an example. With fewer and fewer workers willing and able to assist our manufacturers and fulfilment centers, many are embracing AI and machine learning to  automate tasks such as quality control which are traditionally reliant on scores of human workers. 

Despite the vast amount written about AI, this technology is still, in a sense, in its “mythological” era, mostly due to the mismatch between what science fiction portrays AI to be capable of, and what today’s technology does (hint: it does a lot!). 

How to Get AI Right (or Wrong)

Every new technology has a learning curve, and AI is no different. While an estimated 80 percent of enterprises claiming to already be using AI in some form, research shows that 91 percent of them envision significant barriers to AI adoption, mainly due to a shortage of AI experts and poor IT infrastructure. 

To help illustrate the misconceptions and myths around AI, we will use a hypothetical manufacturer of packaged food, Food Production International (FPI). FPI is a global food manufacturer whose overall goal is to improve production quality and reduce returns/rejects that impact its reputation and bottom line. In addition, FPI is under pressure as it can’t keep up or improve quality control due to a lack of personnel.

After an initial investigation, FPI has launched a corporate-wide initiative aimed at the deployment of AI-powered quality control in its many production sites.

Myth No. 1: I need to go to a university and hire an AI PhD.  

The very first challenges FPI will bump into is understanding what AI is, what it can do, and having the personnel to implement it.

U.S. universities graduate around 3,000 PhDs in AI-related fields per year, with a median of 5.8 years to complete a PhD. This makes finding a ready-to-go PhD very difficult. This is the first myth to dispel: today, AI PhDs are not needed, neither to get started, nor to get deployed in a final solution.

Software platforms are finally available to simplify complex AI problems, providing the needed integration hooks, hardware flexibility, ease of use by non-experts, and, crucially, a very low-cost entry point to make this technology ubiquitously available to Manufacturers and System Integrators.

Myth No. 2: I need to collect millions of images to even know if using AI is possible.

The second challenge FPI will face is collecting data for training visual AI models for quality control. Neural networks and deep learning architectures rely on deriving a function to map input data to output data… which requires, well … data!

While the effectiveness of this mapping function hinges on both the quality and quantity of the data provided, the first fallacy and myth FPI might fall prey to is that today you need a ton to get started. Platforms exist that enable the use of inexpensive (or free) Proof of Concepts (POCs) with just dozens of images. What’s more important, in some cases these platforms do not even require hard-to-find product images, namely, all you need to do is to collect images of healthy products.

Myth No. 3: The test went great, now what?

If FPI has succeeded in using an off-the-shelf AI platform, collected data, and designed a working Proof of Concept (PoC), they may assume that they’re only steps away from deploying a feasible solution.

But truth is that AI adoption in a production workflow requires clear success criteria, and a multi-step approach. There have been countless setups that fall short of implementation for reasons that have little to do with AI, and much to do with the right planning. A simple benchmark such as, “If the PoC delivers X at Y functionality, then we’ll launch it here and here by this time,” would go a long way in terms of helping enterprises define an actual deployment scenario.

Once a clear ROI is defined, the next step is selecting the right infrastructure – both software, and hardware. One common myth is the need for “centralized, massive AI infrastructure with tons of GPUs” to get AI to work. In reality, small, flexible, cost-effective platforms that solve a very large number of vision inspection problems across multiple use cases do exist. Also, in many cases, AI can run as well in CPUs as GPUs. 

Myth No. 4: AI is deployed! I don’t need to touch it ever again!

Because products and processes are constantly evolving, there will never be AI that works off-the-shelf. Rather, FPI will need the ability to build, customize and continuously update AI autonomously, possibly without having to spend thousands of dollars on an expert to come and retrain when something changes.

If successful, FPI will need to pre-emotively grasp the operational conditions that AI requires. It will need to consider retraining costs/time, the skillsets to do it, and the overall AI lifecycle management tools required to make sure an AI project does not become a mess, or worse, ineffective.

What seems like a huge lift in an organization – embracing a new technology – is today as easy as adopting a consumer-grade software or device that anybody can afford and be familiar with in minutes. As new tools are enabling more organizations to take on AI adoption, this new, ‘myth-free’ AI is leading the fourth industrial revolution.