Seven Reasons Why Generative AI Will Fall Short in 2024 — GigaOM Reports

Though generative AI is a huge innovation, there’s no guarantee it will deliver on promises right way.
The mark for 2024 is how bad early and rampant adoption of fully understood AI models is going to affect longer-term adoption.
Some CIOs may rush to say AI is going to change the world instantaneously. It won’t.

Generative AI is a thing. Let’s go further and say it’s a big thing, with lots of promise. But that doesn’t mean it will deliver out of the gate. We asked some of our analysts what will get in the way of generative AI in the short term. “The mark for 2024 is how bad early and rampant adoption of fully understood AI models is going to affect longer-term adoption,” says our CTO, Howard Holton. Agrees senior analyst Ron Williams, “Some CIOs may rush to say that AI is going to change the world instantaneously. It won’t.”

Why not, you may ask. Read on – forewarned is forearmed!

Badly formed answers will not reflect the business at hand, even if they appear to

Howard: Companies are absolutely going to ask badly formed questions about their business. They’re going to get a response that sounds reasonable, but will likely be wrong because they don’t know what the hell they’re doing.

Ron: AIs can hallucinate. Unless you have the background to understand that something is completely insane, you will believe it. Only because you have the knowledge can you evaluate the answers.

Model and algorithm selection will need more effort than perceived

Howard: Setting these models up is not trivial. Businesses are going to make some missteps, from small to huge.

Ron: Many in the press and the AI community have made it seem like training a model is something you do before breakfast, but it’s not. When you train a model, you have to address:

Which algorithm is going to be best for a particular question?
What bias is inherent in the way the learning model was created?
Is there a way to explain the answer that you’re getting?

The bias problem is huge. For example, in IT Ops, if you initially train all of your large language models on a lot of desktop information, when you ask it questions, it will be biased towards desktop. If you train it on, let’s say, infrastructure, it will be biased towards that.

Model training won’t take the business into account

Howard: Businesses will feed models tremendous amounts of business data and ask questions about the business itself and will get it wrong. We will have companies that think they’re training because they’re using one of the private GPTs that ChatGPT enables on the marketplace. This isn’t training at all; it’s manipulating a model. Early results are going to get them excited.

Ron: The business data that they’re going to be feeding this with, whether it’s coming from their salesforce or wherever, they’ve never done this type of thing before. Some of the answers will be massively wrong, and making decisions on those will be difficult to impossible.

Organizations will look to change their structures even before they are on top of it

Howard: 2024 will see companies grossly restrict their operations and hiring, thinking generative AI will help solve the problem. I don’t think we’ll see layoffs, but I think we will see like, hey, I don’t think we need to hire somebody for this. We can fill this role with AI or get enough of an offset with AI. And I think it’s going to go spectacularly, horribly wrong.

Organizations will go for low-hanging fruit but underestimate the higher branches

Ben Stanford, Head of Research: AI can enable teams to shortcut the menial stuff to add more value. But it feels like it might be a little bit like, oh, it made me write these emails a lot faster, and I could do these things really quickly, and then they start running out of steam a little bit because you have to be reasonably sophisticated to use it in a meaningful way and trust it.

There’s low-hanging fruit, but you must consider how you can implement it in a business to yield value. The question is, do businesses see it that way or say, we can cut headcount? Management in many structures are rewarded by how many people they can fire, and this looks like one of the perfect excuses to do that.

Organizational structures will not be set up to benefit

Jon Collins, VP of Engagement: It’s not about whether AI will be useful, but will people be able to drive it properly? Will people be able to put the right data into it properly? Will organizations be organized such that an output from some generative thing changes behaviors? If you get that kind of insight and automatically set up that new business line, that’s fair enough. But if you go, that’s interesting. Now we need to have ten committee meetings, then things are no further.

Howard: Knowledge is not information; information is not knowledge. Giving the information to a junior analyst doesn’t suddenly provide them with knowledge.

Ron: There is an assumption that junior people will be able to use the answers, and AI will provide them with the knowledge and the abilities of a senior person: no, not exactly; if you don’t understand the answer or ask the right question.

Vendors will focus on short-term gain

Howard: We can absolutely blame the big vendors for what they’re doing ‘selling’ their products. They don’t care if executives misinterpret the marketing, then turn around and buy solutions but find out later that, “Oops, we’re now in a three-year contract on something that doesn’t have the value they said it did.”

So, what to do about it?

In consequence, say our analysts, business leaders will hit a trough of confusion when they try to deal with the consequences of getting things not quite right. So, what to do? We would say:

Start anyway, but don’t assume everything is working well already. 2024 is a great year to experiment, build skills and learn lessons without giving away the farm.
Workshop what parts of the business can benefit, bringing in outside expertise potentially to really think outside the box – outside insights, productivity and experience, and into product design, process improvement, for example.
Rather than hoping you can trust models and data sources outside your control, think about the models and data that can be trusted today – for example, smaller data sets with clearer provenance.

Overall, be excited, but be careful and, above all, be pragmatic. There may be a first-mover advantage to generative AI, but beyond this point, there are also dragons, so keep your eyes open and your sword sharp. Even with AI, the first thing to train is yourself.