The leaves are turning yellow, the temperatures are cooler now, and my favourite season, fall, has finally arrived.
I have come across some interesting AI use cases that I wanted to share with you.
-AI see you
-AI generated images and copyright
-Business: AI generated content and human preference
-Leadership: Why you cannot divide and conquer in pharma
AI see you
“You must leave your shopping trolley in the centre aisle” the security guard said as I entered the pharmacy. When I asked why he said “It’s to help prevent theft. People used to steal items by walking out with unpaid items in their trolleys. When we stopped them they expressed surprise and said the items must have fallen into the trolley as they brushed past them while walking past the shelf”.
“We have a new system in place now though” he said, pulling out his phone and indicating the cameras on the ceiling. He continued “the camera feed is monitored by AI, when there is suspicious activity, I receive a video clip”. He pulled up some clips to demonstrate. In one a man took a product out of its packaging and slipped it into his pocket leaving the empty box on the shelf. In another a couple leaving the pharmacy with purchased items exchanged the package contents with more expensive products they had placed close to the exit.
He said, “the system is good, but it is also learning all the time, I validate every clip I get to identify false positives, for example when someone puts their phone in their bag”, adding “of course I can’t personally stop everyone, but as the camera feeds from the shop, the mall and the parking lot are integrated, we can track people to their cars and get their number plates, at that point we involve the police and they take it from there”.
Key takeaways: 1) Everyone you meet can teach you something if you listen 2) The lower the margins the faster AI is adopted 3) Have a good business case for AI adoption and you will likely get funding.
AI generated images and copyright
Needing an image to illustrate a post I thought I would try text to image system Dall E3. I had a clear image in my mind and after providing many prompts and failing to get the quality I was hoping for I finally resorted to adding “generate an image in the style of Magritte and Dali”. Many images were provided, but they either fell short of my expectations, or looked like collages made using other people’s work which had me worried about copyright infringements.
When using ChatGPT I ask for source documents, which I check to validate content veracity and origin. This isn’t possible with text to image systems which are typically trained using millions of images that may or may not be in the public domain. While trying to identify the training data set for Dall-E I couldn’t find the desired information on the providers website, but I did find some text telling me that any images I generate are mine to use as I wish.
However, in 2023 several companies using AI to generate art have been sued for copyright infringement, in one case by visual artists in another by Getty images for using images to train AI models without permission or compensation (Ref 1, 2). And while I am not sure what this means for the end user I prefer to use content that I know I can reuse without any issues at all.
Further interesting reading can be found at the Verge – the scary truth about AI copyright is nobody knows what will happen next (3).
Key takeaways: The field is moving quickly, whenever you use online generative AI tools with a view to sharing the content, consider carefully, and check multiple sources for guidance on use. Also for business use get guidance from your legal team and other internal experts.
1) Lawsuits accuse AI content creators of misusing copyrighted work, Blake Brittain, 17 Jan 2023 Reuters 2) Getty Image, 2023s AI art generator Stable Diffusion in the US for copyright infringement; James Vincent, 6 Feb 2023, the Verge 3)The scary truth about AI copyright is nobody knows what will happen next 15 Nov 2022; James Vincent, the Verge
Business: AI generated content and human preference
There is widespread excitement about the potential to improve business efficiencies by using generative AI for example when writing scientific responses for customers. However, whenever optimisation is looked at it is important to take the human element into account.
A recent article by researchers at the Massachusetts Institute of Technology (MIT) did just that, exploring people’s perceptions, and bias, toward generative AI in the article “Human Favoritism, Not AI Aversion: People’s Perceptions (and Bias) Toward Generative AI, Human Experts, and Human-GAI Collaboration in Persuasive Content Generation” by Yunhao Zhang, Renee Gosline, published in 2023 (link). An article on the MIT website by Dylan Walsh posted in October 2023 outlines the key points (link), I have put together a short summary for your convenience below:
The authors Zhang and Gosline performed the study with the goal of identifying how people perceive content depending on whether it was generated by AI, humans or a combination of both, eliminating bias in some of the assessors by blinding them to how the content they were evaluating had been created.
The content was generated in one of four ways
1) Professional human authors only
2) GPT-4 generated ideas shaped into final content by professional human authors
3) Human generated initial content completed by GPT-4
4) GPT-4 only generated content.
The content was assessed by three groups: Group 1 was unaware of different content generation approaches; Group 2 was told about the four different approaches and the Group 3 knew which approach was responsible for the content they reviewed.
When reviewers didn’t know how content had been generated they preferred AI generated content. However, assessments of content improved when reviewers were told that a human had been involved in its generation, showing what the study authors called “human favoritism”, however, knowing a text had been generated by AI only did not diminish reviewer’s initial assessments.
From Dylan Walsh’s article: “The most direct implication is that consumers really don’t mind content that’s produced by AI. They’re generally OK with it,” Zhang said. “At the same time, there’s great benefit in knowing that humans are involved somewhere along the line — that their fingerprint is present. Companies shouldn’t be looking to fully automate people out of the process.”
Key takeaway: Generative AI is set to revolutionise content generation. Consider how you can balance process improvements with customer preference in your specific area as well as how to assess customer satisfaction objectively.
Leadership: Why you cannot divide and conquer when engaging with customers in pharma
A while back I was caught in the rain as I biked to the recycling plant. Stopping at a tram shelter I passed the time by separating my disintegrated paper bag from my recycling bottles and throwing the bits of paper into the trash. A tram came to a stop, and far ahead, the driver got out of his cabin. He walked up to me and handing me a large plastic bag said, “it looks like this might come in handy”.
I was touched by that simple act of human kindness from an employee of the tram company.
In many professions I have worked in there has been an us versus them mentality. The belief that one team has the customers best interests at heart, while another team does not. For example, when I was a physician, the nurses said “we truly care for patients, whereas you doctors just come and go”.
In pharma, medical affairs teams may feel commercial just cares about numbers, while commercial team members have been know to think that medical affairs colleagues slow them down and lack creativity and customer centricity.
While an employee’s specific department is significant to them, most customers are primarily concerned with resolving their issues. A patient who departs the hospital in good health typically appreciates all the staff they’ve encountered. Similarly, a healthcare professional’s perception of a pharmaceutical company is shaped by her interactions with its employees, regardless of whether they work in sales, medical, or clinical development.
Case in point, I feel positively disposed towards the entire tram company because of a single positive interaction with an employee that made a huge difference for me.
So, while I have seen leaders build strong teams using an “us versus them” dynamic, I think instead of fighting for the “customers’ favour” it makes more sense to identify customer needs and then to work together across functions to meet those needs.
Key takeaway: Customers perceive a company as a whole and company employees as company brand ambassadors, regardless of the individual employee’s function.
Thank you for reading, I enjoy sharing my thoughts and I love hearing what piqued your interest or any feedback. If you are currently working on a project in the fields of medical, digital, systems, analytics, channels, or facing any team or personal challenges, feel free to reach out to me for a chat. I am always happy to explore how I might be able to support you.
Isabelle C. Widmer MD
Image credit: Alex Knight @unsplash