But maybe it’s not what you think, because when we consider diffusion of innovation theory, it doesn’t seem so much of a problem, as a misalignment between terms, industry and an opportunity to do things differently. So let’s look at some reports from surveys done this year on Generative AI and figure out where the problem(s) might be.
Generative Artificial Intelligence appears as if it’s being applied to everything it can, and it appears to be doing so at a rate which has many folks alarmed with the pace of technology. The Life Sciences is a historically slow-to-adopt industry for a variety of reasons, but Generative AI is the first post-COVID-19 (Health emergency) advancement which might buck that trend. it’s hard to measure why that is, at this moment but it could be investment-driven, first-to-market perceptions, workforce driven, market relevance, streamlining internal systems, there are countless benefits to the potential utilizations of the tool both internally and externally.
According to CRB’s Horizons’ report, Most (76% surveyed) Life Sciences businesses have a strategy to implement AI in the next two years. This indicates a high level of interest, however, anecdotally, most folks doing the day-to-day work, aren’t really sure what it means for them or even necessarily, how to use it, or even if they should. Considering that according to a recent PEW study only 14% of American adults have used ChatGPT, and only 19% of those who have heard of ChatGPT, it wasn’t surprising to see a similar (20%) number from anecdotal polls of life sciences audiences at Life Sciences events focused on AI. These data are at odds with the McKinsey report “The State of AI in 2023” which purports significantly higher use by industry, and also makes an important distinction of how the tool is being used. Important not to confuse this McKinsey report of the state of AI in 2023 with The State of AI Report 2023.
Image from The State of AI Report in 2023 McKinsey and Company
It should be no surprise that Technology, media and telecom, hold the highest percentage of users, but the meaningful metric here is actually the “have tried at least once” predominantly because these numbers seem significantly (10 percentage points higher) than both Pew and anecdotal evidence from AI-Interested industry professionals at conferences. One flaw in this question might be that there really isn’t any middle ground between use-scenarios like occasional use or where the respondent hasn’t used it, but is aware of it. (but let’s remember that in the 1980’s McKinsey & Company infamously predicted that by 1999 only 900k cell phones would be in use, so let’s take their data with a grain of salt.) The idea that 22% use it regularly for work generally aligns with other data, however the idea that this is just work related might be unbalanced between industries as this survey indicates.
Part of the discrepancy can be understood by looking at CRB’s Horizons report, as we notice C-suite use and inclination towards AI use is not only high among the CTOs and CIOs, but also CEO and other C-suite.
“What’s more, 40 percent of those reporting AI adoption at their organizations say their companies expect to invest more in AI overall thanks to generative AI, and 28 percent say generative AI use is already on their board’s agenda.” (The state of AI in
2023)
These data do not disagree with the Horizons report, seeing 76% of Business integrating AI in the next two years, as the Question in the state of AI in 2023 report is much more specific, stating an intention to invest enhanced by Generative AI, rather than that they likely already have a plan in execution around AI in general. Perhaps this nuance can be better aligned when we look at the financial aspects, after all Talk is cheap, but actions are more reliable data points. Investment should provide a more useful frame for all of this well informed so-far conjecture.
“It’s impossible to measure ROI for a digital transformation using the results from a single project—the value will be obscured by high infrastructure costs, as well as the inertia of changes to roles and business processes. Instead, companies should rely on a longer time horizon (three to five years) to assess the potential benefits, as well as measuring a variety of KPIs.” (Kulkarni, Thompson, Horizons Report 2023)
When we see the ROI reflection from the Horizons report, that’s where the sentiment from day-to-day operations professionals lines up with expectations. There are some essential notes here to consider: Artificial Intelligence is pretty much most software at this point, it’s a big blanket term, and anyone using software applications these days uses it, whether it’s to guide data management, take notes, retain customers, manage clients. What’s new here is Generative AI, or GenAI. GenAI is the consideration that is worth focusing on when we’re applying diffusion of innovation. Importantly, the manner by-which GenAI is used, is not only through new products and services that integrate an API or even use their private data libraries to forge their own corpus (great word right? It means body in Latin, but in this case: it’s basically a dataset of billions to trillions of pieces of language that help inform the explicit use for the model) for a Large Language Model and hone their Natural Language Processor (If the LLM is learning the Spanish language on Duolingo, an NLP might be going to Mexico for a week and speaking only Mexican Spanish while you’re there) with their semantics and hopefully well maintained data.
Other differences in the reports: Importantly the Pew study focuses on Americans, while the McKinsey report focuses on professionals worldwide, and the CRB Horizons report surveys C-suite leaders and does not disambiguate AI with Generative AI. Considering these differences, the three reports together paint a picture where we can take a closer look at Diffusion of Innovation.
Image from Cornell Blogs: https://blogs.cornell.edu/info2040/2020/12/16/the-diffusion-of-blockchain-technology-across-industries/
If we consider that at least 14%-22% of people have used ChatGPT/LLMs generally and professionally, we’re somewhere near the transition stage of diffusion of innovation, which some folks call the trench, because of how often things fail there. I also don’t appreciate the idea that laggards are tech-phobic, individuals who are tech-phobic do not engage and are not counted in the population/market considered.)
It’s worth noting Towards Data Science Author Clemens Mewald’s article purports to be of the stance that OpenAI/ChatGPT skipped the chasm entirely, however, based on the survey data above, Generative AI and ChatGPT, are likely beginning their position on the early majority sometime late summer or about now. It’s important to note he’s approaching it from a product-centric view for the developer market, whereas I’m looking at the market as a whole from a workforce perspective and claimed utilization as a marker for figuring out adoption. Looking at Workforce development can allow one to infer market readiness, which in my view is more important than a product’s perceived usefulness.
So here is one problem we’re all seeing: an unclear outlook on the application of these technology advancements, in part because of the broad implications, the hype around the topic, and the rate of adoption vs investment.
This informs the second problem: who is doing the adoption.
As a person in Workforce Development, I’m keen on understanding the potential impact to the economics here, so the data above actually has something hidden here. The worker adoptions vs the organization adoptions of Generative AI. At PICLLC we focus on both elements, education and consulting, because we want to ensure appropriate adoption.
If organizations only or primarily adopt these tools, workers might lose a competitive advantage for wage negotiations, have risky use-cases for both organizations, clients and workers themselves, or use for nefarious purposes might dominate the smaller individual utilization.
If primarily individuals use these tools, organizations could cede market share to the few organizations able to capitalize on it, while individuals could create entirely new markets and systems which fundamentally alter the marketplace, accelerate products and services in a way that regulation has difficulty responding to because of the decentralized nature, this might fuel an extreme backlash codifying new regulation by lobbying from the few organizations who want to keep their advantage. This doesn’t even consider the small vs large organization adoption or other elements like accessibility, relevance vs. perceived relevance, but there’s a key takeaway from this, irrespective.
So the key for this tool’s adoption is balance!
Image via Dalle-3
So long as workers understand how to use Generative AI, Prompting, and Prompt Engineering, and all sizes of Organizations are willing to embrace it at a somewhat even pace, as a society, we could see less tumult and more prosperity overall.
Workers empowering their day-to-day use was a key take-away from the state of AI in 2023, because using it at home, and using it at work and home was far larger than using it for work. This indicates intentional experimentation, which can advance the understanding of the tool and expand the use-cases for the worker in both settings. It could even pique the interest of workers simplifying their job/role with new products they’ve created in their off time, inadvertently innovating and potentially creating the basis for a new company. Small business owners who are savvy enough to use the tool can not only find an edge that allows them to compete with larger marketing budgets, but might allow them to streamline a key workflow or pipeline. Back to diffusion of innovation: awareness is looming large here, adoption is growing, but there is currently a lag in continued adoption, likely due to its complexity. ChatGPT appears to hold a relative advantage, despite not having a product to really replace, even with new competitors, like Claude 2 and Bard. Its compatibility likely has some variance between industries, as “The state of AI in 2023” McKinsey report indicates. Its trialability and observability are understandably robust given its free nature (ChatGPT3.5) and instant results are designed to make you believe they’re good. But complexity is observably a problematic barrier, because not only does Prompting and Prompt engineering require a different approach to thinking, but it requires a degree of understanding the tools themselves and their goals. So cheers to the early majority! Hopefully at PICLLC can help folks break down that complexity, because if I go to another webinar and spend an hour waiting for a Scavenger Hunt on how AI can help my business I’m gonna scream.
Lets break some things down, because *maybe* acronyms are kinda the worst. Everything above is really about the effect these tools will have, but for the uninitiated, what exactly is The ChatGPT and is it like The Google?
For those who aren’t familiar, the thing Large Language Models/Generative Pretrained Transformers do is take big piles of data, and essentially the LLM is born of its training on this data, once enough training( aka machine learning aka labeling data according to value: useful or not), fine-tuning( providing direct guidance, often through a prompt on how to consider a path or workflow) and workflows (yea, that workflow you’re thinking of, it’s the same, but lets call it a neural network because its tech now) exist, it can generate things. This process collectively is an example of deep learning. Think of it like autocomplete uses a dictionary and how you type over time, now just imagine autocomplete just keeps on going, assuming the sentence, paragraph and more! That’s generally the gist of Generative AI. It is a lot more than just fancy autocomplete though. ChatGPT does have a few things on autocomplete, being its stronger pre-training, fine-tuning, the flexibility of prompting, and other constraints/enhancements like GPT-4’s Custom Instructions and Advanced Data Analysis (previously known as Code Interpreter) in addition to its new vision and audio integration(future blog topics!). “The Google”, as my grandma never referred to it as, is now more than a search engine as they “integrate ” generative AI guided results, but it’s not a Prompting or Prompt Engineering Tool. ChatGPT is a weird choice to use if one intends to use it as a search engine, because the benefit of the search engine is options of sources and accountability for the information. So if you do engage in zero-shot-prompting (that’s when you just say things to the model in one prompt with no other guidance but that prompt, it’s mostly awful, and requires prompt engineering to really perfect in the first place, but GenAI graphic interfaces like Adobe/Canva products often present it as the main option.)
A note from the Author:
We should all be disappointed that this wasn’t written with AI, given we’re a Generative AI education and consulting organization. But I’m sure you’ve by now no doubt noticed that our blog must be handwritten in cursive and transcribed by bald eagles to ensure it has that “Made In America” feel. Our Social content you’ve probably noticed a similar tone, but a different voice. That’s because as good as Generative AI is, while it can hone a brand voice for Socials, it’s a voice that feels like it belongs there. What you’re experiencing right now is just good-old fashioned man-dictates to a computer then edits it because he has arthritis. Plus every time I ask it to use a Gonzo style, it tells the reader to “Stay weird” or “Stay Pro” and I just can’t with this.
Thank you again for reading our blog, we appreciate your readership, and hope that you gained something valuable from each read.
Next week we’ll be delving into our consulting offerings, some interesting use-cases in research, manufacturing, and marketing, naturally we’ll even be sharing some Chat Histories. Some folks might not know what these are yet, but you want them, they’re super useful, we’ll have a blog on this one function at some point and the wild stuff you can do!
Prompting Integration and Consulting LLC 