..png)
What is Generative AI and how does it work?
Generative AI does what it says on the tin. It enables users to generate a variety of new content based on low human input. And sure, this is a basic description of what is an incredibly complex technology but for most people this is the most important thing to know. So what does it do? Well, there is a simple answer to this which could seem mundane.
Generative AI at its foundation generates text through tools like Chat GPT, images through tools like Midjourney or Dall-E and now looks to be creating videos too (that aren’t something from an acid trip) with Sora. Although its true potential is the deeper use of this, the text it generates might be incredibly valuable insights, garnered from thousands of data points. Its ability to generate images means it also able to interpret them, making it capable of ingesting thousands of hours of footage, tagging and storing content. It is not its ability to simply generate that makes it exciting, but its ability to make connections and learn.
It’s important to note that AI in various forms has been around for some time. The rise in Generative AI, however, is down to its broad and widely accessible use. The fact that 90% of businesses believe it will give them a competitive advantage is both ironic and an example of its scope.1
Its ability to generate text, images and (almost) video is huge, simply because a significant amount of human output sits within these categories. Couple this with the ease of use and sprinkle on the vast amounts of investment being thrown at it,2 means that it is something that lots of people find useful and is only likely to improve. Currently however, for most, it stays there, as a useful tool.
The generation of this content (content i.e. images, code, text) obviously has huge implications for organisations, as just with human made content, there is huge value. At its simplest, used correctly, Gen AI can generate the same content, with the same value, just much faster, and worryingly for most, with far fewer people.3
How does Generative AI work?
But before we get into the good, the bad and the ugly of AI use cases that exist, having a top line understanding of how it works is incredibly important for organisations, purely because it could underpin how successful your use of it could be.
At the heart of Generative AI are machine learning algorithms (which includes deep learning and neural networks. These algorithms enable the system to learn from data. By looking at vast amounts of data the model can analyse patterns and generate outputs based on these. In essence, lots of data goes in which could include images, text and code, and AI can learn from these and then mimic them back in new ways.
The reason this is so important is that depending on what you put in, depends on what you get out (something all marketing agencies are aware of). Currently there are concerns that content will slowly become increasingly homogenous as AI continues to pump out imitations of the information that it learnt on. What is even more interesting, or scary is the idea of AI ‘eating its own tail.’4 Right now consumer facing models such as GPT or Midjourney are trained on the massive amounts of human creation that exists on the internet. But what happens when AI begins to train itself on content that it originally created? Will our content become similar, homogenous, and boring?
Thankfully, there is already a way around this, and this is why it's important for organisations to understand how Gen AI trains itself. If you feed in your own high-quality data, you can create a model that isn’t replicating nonsense, and a model that is hyper tuned to your needs. Now this starts a larger conversation around how your organisation currently stores and sorts its data… but there are also solutions to this too, courtesy of Gen AI. (phew).
Use cases of generative AI
Now… it's fine thinking about the magical (or dystopian) future that people are currently predicting. OpenAI CEO, Sam Altman has recently claimed that 95% of agency work will be completed by AI. 4
Whether this happens or not, what is important is employing AI for your business as there are certainly going to be quick wins from the start. Of course, cleaning up your data and teaching your own model is the holy grail, but we need to be realistic and recognise that proving its value is step 1. Luckily, this is easy to do. When we think of the use cases in Gen AI, we separate it into four categories:
1. Operational is the first and this constitutes the tasks that every professional service organisation must complete, these can be slightly different depending on the industry, but by and large they look the same. This is note taking, email writing, team management, project deadlines… all the things we must do to keep the day-to-day running. A super simple win early on is to review everyone's workflow from a bird’s eye view, understand how long it takes to complete the tasks and how they currently move between one person and the next. Once this is done it can be as simple as employing and connecting a variety of tools. By the end, every meeting is being captured, every deadline is being set and every team member knows exactly what they're doing and when… and if they don’t, they can simply ask the AI.
2. Point solutions are the next stage. These tend to be more specific to an individual business and the returns are easier to measure. In essence, what high value workflows are you currently completing, and can they be automated? Recently at Leapfrog, by [insert what you did] we have turned an eighty-hour task (per week) into an 8-minute job. This solution was hyper specific to that agency and your point solutions will be specific to yours., understanding exactly how that job is completed is key and then building custom Gen AI to complete it is the usual process. With no-code tools and the value that comes from owning IP this tends to be a lucrative investment.
3. Connected solutions are a build on the last stage. If point solutions are automating a single task, then connected solutions look at that task within their wider network. What does the entire process look like? How is the information received? What is done with the output of the job? Once you have a point solution built you can begin to look at the stages that exist before and after that job, from brief writing to client delivery, and look to build them into your new AI process.
4. Agency OS (operating system) is a new model and is the most time consuming, but also the most valuable long term. Some recent reports recognise that to truly be an AI organisation means you must do this at some point.6 In essence it’s about taking your business processes from decks to insight tools to campaign results and teaching your model all of it. By the end of this process you have a smarter, more efficient business and have created the framework to build AI processes across the organisation, you have created an internal brain (think Jarvis from the Marvel movies). What is key is what once would have been a costly process i.e. labelling + organising your data has been completely changed through AI. An organisation's unstructured data, all those briefs, decks and charts sit in your folders, can be accessed, and learnt from in relative ease.
Integrating Generative AI into Business Processes
Now for the all-important question… how do you get started? There are a variety of ways to do this and of course starting with Agency OS is the ideal route. If our experience teaches us anything, it’s that this is very unlikely to happen.
Aiming for a quick win early on should be the goal. Whilst there tend to be some in every agency who see the opportunity, there are also others that need persuading as it is a big undertaking.
Operational efficiencies and point solutions can be found within relatively short timescales. Our suggestion would be to start with a point solution as these tend to be clear revenue drivers i.e. this billable job that took 50 hours now takes us 2 hours. It's extremely hard for anybody to argue with this. And of course you can hire a specialist to help with this process but pinpointing these jobs can be simple and starts with the simple question…
Does the job involve repetitive tasks?
Our advice would be to always start with an easy point solution win and then look to build. If however you are part of the few agencies that have full management/shareholder buy in, then dive in and begin to create your Agency OS. The agencies that do this (and all will eventually) will be the ones to enjoy the biggest wins early on.
Our process at Leapfrog takes the full journey into account and is centred around delivering value as fast as possible, so if you are interested in adapting with AI reach out and get started with a workflow review call.
FAQ
1. What are the differences between Machine Learning & Generative AI?
Machine Learning (ML) and Generative AI (GAI) are both branches of artificial intelligence, but they serve different purposes and operate in distinct ways.
Machine Learning is a broader AI technique focused on teaching computers to learn from and make decisions based on data. It involves algorithms that can analyse patterns in data, learn from them, and make predictions or decisions without being explicitly programmed for specific tasks. ML is used across various applications, from spam detection in emails to recommendation systems on streaming services.
Generative AI, on the other hand, is a subset of machine learning that specifically aims to create new content or data that resembles the original training data. It uses advanced ML models, like Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), to generate new images, texts, music, and more, that can be indistinguishable from human-created content. The key difference lies in its creative capability; while ML is about understanding and analysing data, GAI is about using that understanding to generate new, original outputs.
2. Can Generative AI replace human creativity?
The answer is yes and no, but it is a question that concerns a lot of people. Some of what we currently create will be completed by Gen AI and some of it won't. Human relationships, high level creative work, long term strategic thought are all tasks that currently still need people, but who knows, these two may soon be completed by AI… What is important is to understand that we can only use Gen AI in a positive way if we accept that things are going to change, and these changes are not going to be good for everyone.
3. Will Generative AI replace jobs?
Generative AI is going to replace jobs. It will also create new jobs and it will also make some peoples jobs considerably better. Like any technological revolution before there will be changes.
4. Is Generative AI deep learning?
Generative AI often uses deep learning techniques to generate new data that is similar to the data it has been trained on. Deep learning, a subset of machine learning, involves neural networks with multiple layers that can learn and make intelligent decisions on their own. In the context of generative AI, deep learning enables the creation of complex models capable of generating highly realistic and sophisticated outputs, such as text, images, and sounds, by learning from large datasets. So, yes, generative AI can be and often is deeply intertwined with deep learning technologies.
4 https://arxiv.org/abs/2306.06130
5 https://www.marketingaiinstitute.com/blog/sam-altman-ai-agi-marketing
Related articles
When humanity creates a new way to grow we tend to seize it with both hands, driving it forward, with scale and efficiency being the ultimate goal. The industrial revolution paints this picture clearly and as we enter a similar period of drastic and rapid change it is worth taking stock of the lessons from the past. Basically, how do we not f**k up AI?
May 2, 2024