Engineer’s Guide to Generative Artificial Intelligence

Creating content via ChatGPT continues to spark a wave of interest and speculation. Despite the concerns of the popular culture, artificial intelligence (AI), also known as 'generative' AI is increasingly making its way into the commercial/industrial world.

 

Fueling the AI Revolution

As with any advanced technical innovation, human nature often leads people to overstate its upside potential while understating the downside, and vice-versa. For instance, many content creators, including those in mainstream media, have been quite worried lately whether ChatGPT will replace their traditional creative roles, including your author.

Investors, however, noting the adoption and application of AI across industry and society, look to leverage the technology’s upside potential to increase productivity and profits, while reducing costs and boosting the return on their investments.

One look at the stock market recently supports this assertion. For example, shares of NVIDIA have tripled in value since 2022 on the notion that high-performance, cost-effective graphics processing units (GPUs) will fuel the coming generative AI revolution. NVIDIA is counting on it. “A new computing era has begun,” said Jensen Huang, founder and CEO of NVIDIA. “Companies in every industry are racing to adopt generative AI. With our ecosystem of world-leading software and system partners, we are bringing generative AI to the world’s enterprises.”

 

Skynet? Not Yet, At Least for Now

Figure 1. Despite the hype, Skynet’s version of AI and its army of T-1 Hunter-Killers are not likely to become sentient any time soon. Image used courtesy of Market Watch

 

Seeking both audience and investors, news and financial institutions continue to throw around the term, but what does AI, and more specifically Generative AI, actually mean?

 

What Makes Intelligence Artificial?

Broadly, Artificial Intelligence is a term used to describe computers' ability to emulate human intelligence and logic including analyzing, reasoning, detecting, and learning patterns. This is not new and technologically it’s been around for a long time. One notable example is Alan Turing’s breaking of the German Enigma Code which the Germans used to facilitate the coded communications guiding Hitler’s U-Boat fleet during WWII. Not only did Turing decipher the code, but he also developed an early computer, the Bombe, that by searching all possible settings, could predict what settings were most likely to produce coded messages

There are two primary forms of AI: weak AI and strong AI. Weak AI, also referred to as narrow AI, encompasses all existing forms of AI used today. Strong AI refers to intelligence equal to or exceeding that of human capability. Skynet, though truly a work of fiction, is a good example of strong AI. As of present, thank goodness and Linda Hamilton that strong AI does not exist.

 

The First Layer of AI: Machine Learning

The most common subdiscipline of narrow AI is machine learning (ML). This discipline uses data to make informed decisions or predictions on an outcome or pattern. There are two main forms of machine learning: supervised and unsupervised.

A predictive algorithm is an example of supervised learning, where a computer takes a set of input data often labeled as X, and uses it to predict a respective output y for each data point in X. The sample data is often split into a training set used to educate the model, and a test set used to determine how accurately the model can predict y for a set of X inputs.

A classification algorithm is an example of the latter unsupervised model. Input data X is presented to the algorithm with no known labels y. The algorithm must evaluate the input data X and identify patterns in the data set. A good example of unsupervised learning is image classification.

 

Figure 2. A supervised learning model splits a dataset into training and test data. Generally, large amounts of training data are required for accurate models, although techniques such as transfer learning are changing this. Image used courtesy of v7

 

The Second Layer of AI: Deep Learning

Peeling back the ML onion to another layer, we’ll find deep learning. Deep learning is a subdiscipline of machine learning. It requires less human “tuning” for accuracy and is more capable of improving its accuracy over time autonomously. It does so through the use of artificial neural networks (ANN).

 

Figure 3. A representation of the layers within an ANN. Image used courtesy of IBM

 

Classifying Generative AI

Generative AI started out as a form of unsupervised learning within the machine learning layer. At a surface level, these models generate a unique or novel response similar in context to the training input. These models do not require deep learning. However, the benefits of artificial neural networks used in deep learning made generative AI much more capable. The combination of the two yielded highly capable large language models leveraged by platforms such as ChatGPT.

 

Benefits of Generative AI in Manufacturing

Now that we’ve dredged through explaining what generative AI is, let’s look at some use cases where it can provide value. Given that it ingests unlabeled datasets, these generative models can serve a variety of purposes in manufacturing. A notable area that control.com has touched upon before is predictive maintenance. Predicting and proactively resolving future failures from occurring is bound to improve the OEE of plant assets.

Another area of value in the realm of quality control is defect detection. Identifying defective components via image capture before assembly is likely to reduce costs associated with rejects and improve operational efficiency.

Generative AI can be applied to find patterns in demand, capacity, and production planning. Given a large dataset on customer buying patterns, a generative model might suggest ways to predict consumer demand and help decide which products they should make more of to meet it.

 

Figure 4. An example of a defect detection model being used in the automotive industry. Image used courtesy of Qualitas

 

A final example of generative AI is an iteration-based model for computer-aided design (CAD) programs. Fixed project parameters are defined, while other parameters are adjusted to optimize material usage and weight while maintaining strength along particular vectors and the center-of-mass location.

 

AI in Manufacturing

The capabilities and recent growth of artificial intelligence are exciting. The technology will undoubtedly drive many technological advancements in the coming decade, and provide value to society. I’m looking forward to seeing how generative models are adopted in manufacturing settings.