Generative AI Revolution – navigating the hype cycle

Having established an understanding of the capabilities of Generative AI in our earlier article, ‘The Generative AI revolution’, the agenda at leadership meetings will swing to ask questions about this technology – and rightly so.

“Can we be using generative AI to improve our business? In what areas? By when? Is this all just hype?”

These questions are increasingly on the agenda at leadership meetings of middle market companies. And they are not alone. Use of “ChatGPT” and other related names in publicly listed company transcripts and analyst presentations has skyrocketed since the launch of ChatGPT in late 2022.

Riding the wave of the hype-cycle

The technology is certainly going through the hype cycle, especially while we are all grappling with what the technology is, how it works and establishing its limitations. The graph below - courtesy of HFS - captures what’s called the Dunning-Kruger Effect in the case of ChatGPT - but the theory applies to any other generative AI tool at this point.

The graph illustrates how, at the outset, our confidence levels about what can be done with the technology are high. This is not due to our understanding, but rather due to the opposite – our limited understanding combined with the hype, which makes us over-confident.

As we understand more about the technology our confidence in its power to impact everything diminishes. But, as our understanding grows (moving right along the X axis) we begin to truly understand where it can be effectively used, with the result being that our confidence begins to rise again but this time grounded less on hype and more on reality.

Most of us are still in the left half of the curve.

A growing understanding of generative AI takes us from a state of hype to a one of realistic understanding of where it can be effective and useful

When will businesses be impacted by the AI Revolution? And in what areas?

Gartner’s latest tech impact radar (December 2022) suggests that Generative AI will have what it calls “High Mass” impact in 3-6 years. In 2021 it estimated this to be 6-8 years.

But in the meantime, some companies are already sitting on the right side of the curve, with use lying in these eight general application areas:

• Data generation
• Code generation
• Text generation
• Image generation
• Video generation
• Audio generation
• Model generation
• Avatar generation

The following are some specific examples for some of those areas:

1. Data Generation

In the mid-2010s, ground-breaking work was being done by firms like Washington DC-based OmniEarth (acquired in 2016 by Vista Equity-backed EagleView) in applying machine learning technology to satellite imagery for object recognition, e.g. to help property insurers quickly determine which of their clients’ buildings were damaged after a storm. However, these firms were hamstrung by learning datasets for their models. These sets were created by manually labelling undamaged and damaged buildings – an expensive and slow process that inhibited use cases and even made some unviable.

Generative AI’s core function of creating data using existing data makes it possible to synthetically generate these learning datasets. According to Gartner, “self-supervised learning…eliminates human-in-the-loop from model training by enabling labelled data to be created from the data itself. This is useful when available data volumes are limited, or when the benefits of the machine learning (ML) solution do not outweigh the costs of manual labelling or annotating of data.” 

Generative AI can also be used to clean, filter, classify and fill in missing data making them more commercially valuable.

2. Code Generation

Coders are increasingly using generative AI tools like GitHub CoPilot to bring speed and efficiency to the coding process. CoPilot uses OpenAI Codex and describes itself as being able to “suggest the code and entire functions in real time, right from the editor”. In other words, it lives in the code editor rather than needing to switch to a web browser or a separate application. It learns the context of the file the coder is working in and is able to make relevant suggestions and can generate complex algorithms that the coder is able to ask for in natural language rather than just being a component, a bit like what a puzzle piece is vs. the person putting the pieces together.

As a gaming developer who has only been using this technology since January this year described it, “it’s a bit like having another coder sat next to me to speed up and strengthen the final result, but without the personnel element, which can slow things down and cost much more.” The coder did however provide this caveat: “The outputs from the process must be checked!”.

3. Image & Text Generation: Rapid prototyping in industrial applications

Using the ability to distinguish between multiple images, process natural language, and rapidly work through complex requirements and other considerations e.g. health and safety regulations and specifications of raw materials, generative AI is able to rapidly create alternative designs for a given challenge. This speeds up prototyping for industrial products and provides teams with many options with slight variations in each case, which helps firms experiment more quickly and efficiently.

According to MIT, industrial applications are new with examples cited at Belgium-based engineering design software firm Bricsys (using generative AI to rapidly create 2D and 3D drawings and plans for buildings and engineering systems) and at Schneider Electric where it is “design[ing] clearer safety and repair instructions for their maintenance personnel responding to trouble tickets in the field”.

Commercial and industrial applications are still at a formative stage in many respects, with some applications being more advanced than others. The technology is still being hyped up, but the range of applications and the unending use case possibilities means that companies must take this technology seriously. At this point business leaders should be developing an understanding of the capabilities of generative AI and where it will impact business operations and addressable markets. A timeframe of much less than a decade will almost certainly require an AI strategy to be developed today.

Companies across a wide range of sectors should expect generative AI to materially impact them in the future. Whether it is a positive or negative impact may depend on strategies developed today coupled with the entity’s technology skillset, inquisitiveness and its ability to identify where the impact will be deepest in their business.