The Great Generative AI Gold Rush
Written by All In on Data
The Great Gold Rush
While listening to a panel of experts discuss Gen AI, analytics, and responsible AI, I realized that there is a large and widening skills gap between people who really understand Generative AI and those who've recently anointed themselves experts. But there's a good reason that so many people are trying to get in on Generative AI.
ChatGPT has single-handedly transformed the future landscape of technology and business. If your project or company or startup is not in some way embracing the use of an LLM then you are not getting funded. Of course you must now add '.ai' to every domain name and consequently the small Caribbean island of Anguilla is making a boat load of money registering domains. Now ChatGPT has brought into the mainstream a technology that has existed for several years. Well to say it brought it into the mainstream is a small understatement. That's like saying Taylor Swift plays the guitar. I firmly believe that you can find a tribe in the middle of a Brazilian rainforest and they will know about ChatGPT. Suffice to say it's exploded into our collective imagination in a way that no other previous product has. Ever.
Those of us who have been following these technologies and reading the academic papers are not terribly surprised at these technological advances, but what OpenAI did was execute the greatest marketing feat ever. When your grandmother starts talking about ChatGPT and LLMs you know things are getting weird. At their most basic level, LLMs and Generative AI is just funky math with a lot of compute power thrown at it. But that funky math has produced an incredibly powerful tool that will have a broad, deep impact on all our lives and we are only at the very beginning of what has already been a wild and unpredictable journey.
But as history has shown us, when there is a gold rush everyone becomes an expert miner even if they've never before gotten off their couch or lifted a shovel. If you watch enough YouTube videos you can convince yourself you're an expert in the nuances of gold mining. And with your newly-discovered expert opinion you are free to opine on how one should mine for gold. In the AI gold rush, thousands of videos and millions of tweets (or X posts, or whatever they are called now) will be created by all these expert miners. Mind you none of these miners have any calluses from the hard work and since it is this easy we will, in no time, all become fabulously wealthy! (Because look how well that worked out for crypto mining and NFTs.)
I have been through the hype cycle of a new technology more than once and the pattern is usually the same. So-called experts rush in providing their guidance and eventually the market reverts to the mean and those that are interlopers fall by the wayside. Those that have developed the calluses and sore muscles from years of doing the hard work eventually get heard and people begin to appreciate the complexity of mining. These true experts are usually pushing a more realistic and cautious message, based on their years' of hard-won experience. Instead of unbridled enthusiasm and hype, they counsel moving deliberately and putting some guardrails in place. Not an exciting message for the get-rich-quick crowd, but eventually the inescapable truth is recognized: Doing hard stuff takes hard work and lots of experience.
Generative AI and LLMs are definitely “hard stuff.” They involve complex pieces of software (and sophisticated hardware) that require a lot of hard engineering effort, along with deep institutional and organizational knowledge to make them work effectively at scale.
Navigating The New Frontier of Gen AI
Corporate boards are increasingly recommending that CEOs consider the integration of Generative AI (Gen AI) into their business strategies. This directive is cascading down the organizational hierarchy, emphasizing the importance of adapting to Gen AI to maintain a competitive edge. While the urgency to adopt this technology is evident, it's crucial to approach it with a balanced perspective. Gen AI represents a significant advancement in technology, offering substantial benefits, but it is a tool to augment human decision-making (Augmented Intelligence), not replace it entirely. It's essential to understand its capabilities realistically and implement it thoughtfully to optimize its potential in enhancing business operations.
An organization comfortable with traditional analytics and AI will see both the promise and shortcomings of Gen AI. It has immense power to alter an organization and make it more efficient but it will also be highly disruptive, and often not in a good way. Understanding where and how to apply Gen AI and LLMs is going to take time, and education will be critical to success.
I've seen organizations fail miserably when deploying innovative technology. Frequently they will put engineering teams in charge who have no business context or, and more likely, misaligned incentives. Engineers live to solve hard problems, often the hardest and most complex way possible. This gives them a chance to prove their value, learn new skills, and even improve their resumes. The most complex solution is not always the best solution for the business. In the case of Gen AI engineers will try to solve what they think are business problems that business people would want to solve. Of course many engineers have zero business experience but why should that stop them from trying? The business, on the other side of the table, will often ask for the ocean to part and miracles happen. They fully expect that this new technology will make their jobs easier and help them earn more revenue.
I will say it again and again and again. Deploying innovative technology is brutally hard. Innovation is hard. Full stop. The path to deploying Gen AI and LLMs is fraught with landmines. Big, large, coming at you from all angles, kind of landmines.
Now that I've scared you, what should you do?!! Take a deep breath. Start simple. You'll have plenty of chances to learn more as you progress, so don't feel like you need to tackle the most pressing use case first, and certainly don't get talked into a “boil the ocean” approach. You may have to remind your stakeholders periodically that this is technology, not magic, and it's still difficult.
Here are some things to keep in mind as you plan your first Generative AI project:
Select the
right leader
You may have some “hot shot” AI experts who are anxious to show what they can do. If so, that's great; put them on the team. But don't put them in charge yet. Make sure you've got a leader who has institutional memory, knows how to get things done, and knows where the pitfalls are. These are complex technical projects, and you'll still have to get through all of the technical approval gates to put a system into production.
Don't overlook the organizational challenges
Gen AI is likely to produce considerable organizational disruption and uncertainty. Virtually every part of the organization can be impacted by this technology, and many people will be feeling uncertain about what it means for them and their roles. Software developers may be anxious about being replaced, or wondering how they can obtain the new skills needed to work with Gen AI. People in other roles may be excited to use these tools, or they may be fearful that they'll lose their independence or even their jobs. You'll need to be aware of these challenges, and manage them from the outset.
Pick the right
use case
You may be tempted, or even pressured, to launch an amazing and transformative new product based on the latest large language models. Don't do that. Start with some smaller scale experiments that will give your team an opportunity to learn and develop the skills required. In selecting your use case, try to balance the utility of the use case against its complexity and risk. Pick something that has value, but that seems doable. If you can deliver on that, you can start ramping up the complexity (and the value) of future use cases.
Make sure you
have the data
The foundation models that power all Generative AI have been trained on data that's…not yours. This isn't necessarily bad, but it means that you can't fully understand what they were trained on, and you can't expect the models to fully understand your business. So you'll need to introduce your own data into the equation in order to create anything that's truly differentiated. As with most analytics projects, making sure your data is accessible and suitable can be a significant challenge.
Engineering
still matters
It can be tempting to think that LLMs and other AI tools mean that we've truly entered a no-code world. After all, LLMs can even generate code. But while there may be less traditional code-writing in some of these Gen AI projects, there is still a lot of engineering required. So get the best engineers you can find, and give them the training they need to work with Gen AI tools.
Start with
Responsible AI
While much of the public discourse about AI taking over the world and ending humanity is overblown, there are some real risks that go along with AI. You need to consider the safety, ethics, and potential harms of your project. And you need to develop a framework for managing those risks, starting with the first days of your project. You'll need a Responsible AI approach, to make sure you can actually deliver your project into the world, and to make sure it doesn't harm your customers, your company, your colleagues, or even broader swathes of society. Done correctly, this kind of approach actually increases the likelihood of your creation seeing the light of day. Done wrong, and well….good thing you brushed up your resume.
Looking Ahead
The journey to successfully integrating Gen AI into all aspects of business processes is going to happen but it will take time. For many organizations it will be a long painful process that will result first in many failures before they get it right. Those that don't well we won't be talking about them since they won't be around.
There are many what I call 'low hanging fruit' processes that are easily optimized with today's current state of tools without too much organizational disruption. And in many cases they are probably happening already without you even knowing about it. They are the processes where 'Augmented Intelligence' works best. Tools that facilitate and make someone's job easier. The software engineer using software that generates code based on requirements which will increase their productivity significantly. Tools that generate presentations efficiently and better. Design tools that facilitate and enable a much richer and better design process. The list goes on.
We are in very exciting times. The advent of this technology has already had a broad impact on our culture and it will only go deeper as it matures.
I will leave off with this one final comment - proper talent matters most. Without the right advice and the right people to smooth the way the move to Gen AI will be extraordinarily difficult. Make the right choices and hire the right people.