AI and I: Two AI Tales from the Ton

Gentle Readers … the newest Bridgerton season is upon us and in this newsletter, I took the liberty of mimicking Lady Whistledown’s tone in talking about AI, which seems to be on the lips and minds of everyone at the Ton.

I begin with a sincere apology for my tardiness in sharing my latest Society Paper. The minutiae of my vocation consumed my being and left little time for my avocation (or should I say AIvocation). ?However, as the Ton sets into its summer routine, this author is again putting pen to paper about the glittering and ghastly world of AI.

As always, the intriguing news first and then the main feature, which in this newsletter highlights two exceptional AI projects that students in my graduate Generative AI class completed.? I could not fit more here, so I will continue to include others in future letters.

AI News

I begin with three sensational developments that have emerged this season …

First, there is a new debutant in the world of AI with a most scandalous name if ever there was one, ChatGPT-4o. But the plot thickens.? The O was supposed to represent “omni” for omni-modal, but interestingly, when I asked how it differed from its much sought after cousin, ChatGPT-4, it said the O was for Office (implying some nefarious connection with Microsoft)! ??Methinks the new entrant from the same family is a better suitor to the likes of the Ton, what with its array of charms: Faster processing of prompts; more versatile in terms of languages (I asked for the essence of two world religions—Hinduism in Tamil and Christianity in German (two languages I know besides the Queen's English)—and ChatGPT-4o gave me both before I could say Lady Danbury!)

… it is rumored to soon-be-available to non-paying members of the Ton; plus, it is conversational and even a bit coquettish! Now you know why the Ton is in love with this new entrant even as envy spreads among its rivals. ?Not all is well with OpenAI though; they did upset the incomparable Scarlett Johansen by mimicking her voice. Even Cressida Cowper would not stoop so low!

Speaking of rivals brings me to my second juicy bit of news.? The biggest scandal in the Ton was not just the sibling rivalry between the ChatGPTs, but the timing of OpenAI’s announcement about ChatGPT-4o—a day before Google’s own I/O developer conference. ?Ah, the intrigue! But as you, my dearest readers know, Google is a worthy adversary.? Sundar—my thambi from another thai (“my brother from another mother,” in Tamil, our common native language)—spoke softly but carried a big stick, Vertex AI that includes a collection of AI models including Gemini 1.5 Flash and PaliGemma, a vision language model.? Even Queen Charlotte would be impressed with this compelling collection of contenders at this season’s AI cotillion.? I certainly was when I tried Vertex AI’s multimodal capabilities in its Model Garden.?

But all is not well with Google’s AI forays. ?Their AI Overview responses to search queries have caused much consternation among the Ton for the occasional nonsensical output (Glue on pizza! Good Heavens … first pineapple, and now this … Raffaele Esposito will be tossing in his grave!).

Finally, gentle reader, an entrant from an unexpected source offers intriguing possibilities, particularly for education. Khanmigo, an AI Tutor from Khan Academy, is built on the ChatGPT model with additional training from its own content. ?It offers a customized tutoring experience on a wide range of subjects from Math, Science and the Humanities.? What I liked best about this well-mannered tutor is it helps identify holes in your understanding of a topic.

I chose Generative AI as my topic to be tutored on and Khanmigo asked me ten questions, with each getting progressively harder.? It can also give you feedback on your writing, help you focus, prepare for a debate and even allow you to chat with a historical or literary character.? I asked Alice in Wonderland why the gardeners were painting the roses red and got the right response!? (I should have also asked what would happen between Colin and Penelope, but that would have given the game away.) ?Truth be told, I was very impressed, although it is still in beta version. ?Also, Khan Academy just announced that with its partnership with Microsoft, Khanmigo will be free to all educators in the U.S.? Even the snobbish Mr. Darcy might consider this development “tolerable” (apologies for mixing up my British period pieces).

Now, for the main feature (and I reverted to modern day lingo for this … it's tiring to maintain the Regency-era tone).

With my students’ permission, I have profiled two projects (from a longer list I hope to include in future newsletters) completed by a couple of brilliant graduate students in my Generative AI class this spring.

While each project represents a different use case, the students’ work experiences in their respective industries—manufacturing and pharmaceutical & biotech consulting—made the application of GenAI tools to solving business problems in these industries that much richer and more relevant.

AI and Manufacturing: The Use Case of Predictive Maintenance

The first project was developed by Shaji George, MBA , a professional with significant experience in manufacturing, where he used ChatGPT-4 to analyze machine performance, forecast equipment failures and develop a protocol for predictive maintenance as an alternative to the widely prevalent preventive maintenance model.? Predictive maintenance involves collecting real-time data from sensors embedded in machines—including vibration levels, temperatures and energy consumption—to help predict potential failures. This project used GenAI to show how manufacturing facilities can move towards such proactive maintenance, targeting specific components, and away from time-based methods.

Cluster Analysis

Shaji began by using the GPT to merge multiple data sets, representing different data streams from manufacturing equipment, clean the merged data and visualize it. Then he examined equipment failures by having the AI conduct two different cluster analyses—the first based on torque and air temperature and the second based on torque and rotational speed—using K-means clustering (where k=3). The results of the first analysis are here:

Classification

Next, to classify machine failures, he had the AI split the data into training and testing sets and identify the most accurate model. The Random Forest Model, with an accuracy rate of 98.95%, was chosen to classify the machine failures.? The graph below illustrates the importance of different features used by the classification model.

Predicting the Next Immediate Failure

Then came a key purpose of the project as embedded in this prompt (underlining added), “By considering the uploaded dataset as the latest available data, can you predict the next immediate failure and then recommend a specific maintenance activity to stop the immediate failure from happening?”

The Alert Message

And finally, the pièce de resistance, the generation of an alert message to the maintenance team with accompanying visualization of the predicted failure condition:

?This project highlights the role of generative AI in helping to build a predictive maintenance model in a manufacturing context by: combining data from different streams associated with different aspects of equipment operations; helping to visualize how failures are clustered; building a predictive maintenance model; and providing an alert message when the combination of failure conditions becomes critical.? Shaji’s project illustrates how Generative AI can be used in manufacturing contexts to build a predictive maintenance model as the field embraces Industry 4.0 that includes the Internet of Things.

AI and Pharmaceutical and Biotech Consulting: The Use Case of Recruiting

?The second project was developed by Jesus Arzate, MSBA who used GenAI to improve the recruiting process for pharmaceutical and biotech consulting—a field in which he has significant experience.? Overcoming the key challenges of reducing time-to-hire and ensuring fit-for-the-position are critical to successful recruiting in any field, and especially so in this specialized area of consulting.? Jesus used ChatGPT-4 and a specialized GPT, Human Resource Job Matching and Screening GPT, to improve the recruiting process.? Interestingly, he was also able to compare some of the outcomes that would have resulted from the AI-supported model with those from the original process that was used without AI support.

Parsing Resumes and Job Descriptions

The project began with the collection of resumes and job descriptions taken from an Applicant Tracking System.? The data sets were combined, anonymized and converted into JSON format for NLP processing. The detailed prompt below provides an excellent roadmap of the initial steps that the AI was asked to complete.

An excerpt of the output from parsing the anonymized candidate resumes and scoring them is shown below:

Assisting with Interview Questions

The AI was then asked to generate interview questions on five topics: Technical Skills and Knowledge; Problem Solving and Project Management; Communication and Team Collaboration; Compliance and Regulatory Understanding; Strategic Thinking and Future Orientation; and Cultural Fit and Motivation. An excerpt of the output is shown below:

Estimating the Impact of AI on Recruiting

Finally, Jesus was able to measure the estimated impact of AI had it been used during the recruiting process for this group of candidates across three different functional areas—R&D, Manufacturing and Finance—and the efficiency gains averaged about 45% (see below).? What is equally interesting is the distribution of candidate matching scores, which seemed to follow a Normal Distribution, that would have allowed the organization to focus on the above average candidates (about 25) who best fit the job requirements.?

Even after hiring, this AI-integrated approach can help with strategic workforce development in such areas as employee retention and skill gap identification. However, applying AI to HR issues broadly needs to be done carefully.? As Jesus notes, “Protecting data privacy and reducing biases within AI algorithms requires a strong commitment to security practices and adherence to regulatory standards. Using diverse training datasets and performing regular audits are essential to mitigate these issues.”?

As mentioned earlier, both projects benefitted greatly from the expertise of the two students, who were domain experts in their fields—a fact that underscores the importance of this eponymous newsletter’s theme, AI and I.? Understanding the context in which you can apply AI, interacting with it as you would with an assistant and verifying outputs are vital to ensuring successful outcomes. ?Unlike in software deployment, where implementation results in consistent outcomes (good or bad), working with AI is much less deterministic and requires constant engagement (at the start and throughout) and oversight (of the steps and outcomes).??

Dearest Readers ... as we wrap up this edition of our newsletter, it is evident that the innovations in the world of AI are as thrilling as any debutante's first season in the Ton. We are in a new era of manufacturing far from the steam engines I grew up with, which I heard the Duke of Hastings refer to as Industry 1.0! ?In these modern times, Shaji George's project highlights how AI can predict failures before they occur—much like preventing a scandal before it reaches Lady Whistledown's ears—to help keep the machines of industry running smoothly. And then, in the matter of hiring people, particularly in specialized industries, Jesus Arzate's efforts offer some sparkling insights about the use of AI. ?Just as a skilled matchmaker (ahem Lady Danbury) pairs suitable partners, AI can sift through hundreds of candidates and jobs ensuring the perfect fit for each role, which is amazing and alarming at the same time!

As always, gentle readers, stay tuned for more riveting tales of AI in forthcoming editions (where I hope to include some more student projects). Until then, may your endeavors be as creative and your projects as interesting as the two we highlighted here.

#AIandI #AIandManufacturing #AIandPharmaBiotechConsultingHR?