Elise Carmichael is the CTO of Lakeside Software, where she oversees the design and delivery of its digital employee experience platform.
getty
It’s been a little more than a year since OpenAI launched ChatGPT using the GPT 3.5 large language model (LLM), forever disrupting business as usual everywhere. Even though generative AI is very exciting for long-time computer science folks like me, it will take many organizations a while to get comfortable with using and/or building custom GenAI models and, more importantly, to trust them.
It is easy to trust a chatbot that seems to give credible answers, but these models are really just figuring out what the next word in a series should be. Although they are scarily accurate, they are known at times to “hallucinate” (15%-20% of the time in the case of the latest ChatGPT model). One infamous example: “Some species of dinosaurs even developed primitive forms of art, such as engravings on stones.”
Regardless, we can’t live in the dinosaur age of technology (a.k.a. the era from, say, the wheel to November 29, 2022). In fact, we’ve all heard that “AI is eating the world.” We don’t need to fear it; however, instead, why not grab a spoon and dig in? Only then can a company claim a primo seat at AI’s table, relishing its business benefits if the AI recipe is right.
In my world as the CTO of a software company that serves the IT community, from CIOs trying to mitigate digital risk to IT practitioners trying to deliver a better digital experience, I have five of my own predictions about AI for IT in 2024. Although the kind of AI models I envision may not be as cute as Disney Research robots (my husband helped build this one!), I still find them to be pretty darn fascinating.
1. The Importance Of Keeping The ‘Human In The Loop’
One of the biggest issues with LLMs comes down to trust. Whether you’re integrating existing LLMs or building your own (which only the largest enterprises are likely to do as I discussed with Zack Kass, formerly of OpenAI), can you and your customers trust the model to give you right and relevant answers? Explainability is a problem that most product companies have when implementing machine learning algorithms.
If you’re making business decisions off the GenAI model, it must be incredibly trustworthy. How do you get around the hallucinations? Well, you keep the “human in the loop” to validate as you go. Let’s say you want to query a large dataset, so you prompt, “Show me all employees and their corresponding time zones.” You want to ensure that the response is labeled and know exactly where it comes from. If there is a business decision to be made, the model’s output must be accurate. AI without human oversight in applications is dangerous where the right level of trust has not been established.
2. AI As An Easy Button
Let’s face it: Early-career employees have grown up with iPhones and laptops that just work. The digital employee experience of these employees is simply not the same as that of the generation that grew up with less sophisticated devices that required regular troubleshooting. Given that IT teams have to strategically accommodate these two different levels of experience in the workforce at the same time, I think that IT teams are going to look to GenAI to serve up quick answers or automated self-service options—with AI as an easy button of sorts.
If IT cannot accommodate easy fixes, digital natives are going to get frustrated very fast when their unfamiliar operating system is “running slowly.” It’s important to acknowledge, too, that being on the phone for IT support is the last thing early-career employees are expecting. That vision is as outdated as a landline phone, so IT teams will want to mature their automation and AI capabilities for faster responses and simpler self-help options.
3. Leaning On The Edge For Training AI Models
There has been this giant move from on-premise data centers to everything cloud. Because the cloud is very expensive, however, I foresee that more and more companies will turn to the edge, even for training foundational AI models. Not every company has a partnership with Microsoft/Google/AWS to get reduced-price compute power, but every company has an army of compute that mostly sits idle at night. My prediction is that we will see a move toward the distributed edge network for compute power.
4. The Great Divide Between Early And Late Adopters
I personally think that, because of GenAI, the world will be unrecognizable in 10 years in terms of tools and technology—with a big impact on how users work with computers and services. We may start to see the first fissures of a chasm between the early AI adopters and the late adopters.
Remember the nearly 20-year-old theory of “disruptive innovation” by Clayton Christensen, Michael Raynor and Rory McDonald? Gee, I wonder what Blockbuster thought about that? With GenAI gaining a steady foothold, we will face another wiping out of corporate mastodons if they don’t start innovating (today) and delivering (tomorrow) on all the possibilities of GenAI.
5. Data Is The Name Of The Generative AI Game
LLMs aren’t new, but how fast and accurate the models are is new–-even when it’s coming up with new content (zero-shot learning, or “the ability for AI to learn to do new tasks without any training”). Data is the main driving factor, making the latest models “scarily” accurate! They often create new, accurate content without ever having seen it in the wild.
So, my prediction for this year (and well beyond) is that data will be the name of the GenAI game. As my company’s CEO Dave Keil recently wrote, “The real breakthrough of generative AI is the data itself.” Indeed, the better the data, the better the AI. That said, I think that companies with the deepest data lakes for their given industry will win the AI long game. Since my company has been collecting IT endpoint data for more than 25 years, I certainly hope so!
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
