By far, the greatest danger of artificial intelligence is that people conclude too early that they understand it.
– Eliezer Yudkowsky, the Machine Intelligence Research Institute
Man’s quest for knowledge has ended with the smartphone – well, mostly. There’s still stuff for humanity to discover but for everything else, there’s ChatGPT. Those of us with faded memories of encyclopedias as the most convenient summary of the world’s information are still amazed by how easy it is to get information so quickly from Google. But generative AI models have taken this to a new level, as though each of us carries an all-knowing, always-accessible wizard in our pockets who can instantly answer nearly any question.
These tools are also incredibly creative. If you’ve asked Bard or ChatGPT to compose an article or story for you, you can understand why the writers in Hollywood are striking. And, in my opinion, why they’re destined to lose the battle vs. AI. The first recorded incident of humans protesting against the introduction of job-threatening technologies was the 1675 English battle over the use of “stocking frames” in the textile industry. The machines have prevailed every time and this round will have the same outcome.
A Very Brief Primer on What Makes AI Different
Artificial intelligence is not a synonym for advanced technology. AI learns and improves on its own, making it unlike any other technology. Here’s an example: My wife’s SUV came from the factory with radar to keep her safely behind the car she’s following and lane-keeping to stay between the white lines. But it doesn’t learn from experience and get better at these tasks over time.
A fully autonomous car, however, is powered by AI. That means it learns to drive better based on its own experience and it’s likely that someday, self-driving cars will learn from each other. Soon, these cars will communicate with each other and everything else, too, meaning the many foibles human drivers suffer from (road rage, drunk driving, texting) will be moderated or eliminated by AI systems that operate with better judgement than people.
AI models learn the fastest with lots of data, and humans now generate more data than ever: Statista predicts we will generate 181 zettabytes of data globally in 2025 compared to just 2ZB in 2010. (A zettabyte is a billion terabytes. The Library of Congress’s entire printed collection could be stored in about 10 terabytes).
Additionally, computer processing speed is still following Moore’s Law, essentially doubling every two years. And just as we approach the physical limits of semiconductors, quantum computers are emerging that will improve processing speeds by orders of magnitude.
So, for the first time in history, we have technologies that can learn over time, get better at what they do automatically, and they’re powered by an increasingly capable infrastructure and a fast-growing amount of data.
The Tantalizing Reality of AI
Large language models (LLMs) like ChatGPT and Bard do much more than answer questions, solve math problems and find recipes. You can use either system to:
- Write code. I asked ChatGPT to write an API in Python to connect White Cup’s analytics module to Epicor ERP, which it completed in about 20 seconds. FYI, I haven’t tested this.
- Produce business documentation. In 10 seconds, ChatGPT produced a requirements planning document for a distributor looking for ecommerce software.
- Provide stock analysis. Google Bard wrote a nice summary of how Grainger has performed recently compared with stock-market expectations.
- Write a short story. I asked ChatGPT to write a story beginning with the words, “It was the best of times, it was the worst of times.” The resulting 641-word essay was remarkable.
- Evaluate disruptors. Google Bard did a great job responding to the question, “Is Amazon Business a threat to distributors?”
- Analyze acquisitions. ChatGPT wrote a thoughtful analysis of Distribution Solutions Group’s acquisition of Hisco.
DALL-E, from the developers of ChatGPT, uses AI to create images based on your specifications. I asked for “an interior shot of a busy warehouse” and got these four choices:
These aren’t perfect; for one thing, they’re not “busy.” However, they certainly look like real photos, right down to the gleam of the lights on the floor. But they’re not; these were entirely computer-generated, and I bought the rights to all of them for 13 cents. Having spent many hours searching for images only to pay top dollar at stock photo sites, you can bet I’m going to use this service sometimes.
But is that fair? Certainly DALL-E studied warehouse photos by the thousands to make these pictures, meaning they’re derivative of the work of real photographers, none of whom will be compensated for this image.
This affects my company, too. LLM’s could make it much easier for publishers like us to generate content for our website. But those same models are going to use our content to produce answers for people seeking information about distribution. When executives ask ChatGPT and Bard for information about wholesale distributors, some of that content could be derived from what we’ve produced, but we’d get no compensation or even attribution.
Types of Artificial Intelligence
Let’s organize our thinking about artificial intelligence in business in general and for distributors specifically. If, like me, you’ve spent a lot of time researching this, you’ll discover that AI technologies sometimes get organized into these four categories:
- Reactive machines. This is an AI system with a very narrow task. It has no memory and simply evaluates many possibilities to choose the optimal answer for whatever it’s assigned to do. IBM’s Deep Blue computer (the one that beat Chess Champion Garry Kasparov) is usually given as an example.
- Limited Memory. These systems store data and use it to make accurate predictions. Pretty much every writer cites self-driving cars as a “limited memory” system.
- Theory of Mind. Definitions vary here, but the most common idea I’ve found is that theory-of-mind AI systems will be able to discern the feelings, emotions and beliefs of the entities it interacts with. There are many examples, but for a recent video, check out this segment from 60 Minutes Australia. A reporter interacted with an AI robot that arguably possesses these capabilities. It’s exciting or eerie, depending on your point of view.
- Self-Aware AI. This is where AI systems become aware of their own emotions. In other words, this is AI with its own consciousness.
I actually don’t care for this categorization because it’s inaccurate for at least two reasons:
- Only technologies that learn and improve on their own qualify as AI. This disqualifies reactive machines entirely. Deep Blue – the darling of this category – didn’t learn how to play chess; it was pre-programmed to pick the best move based on the configuration of the pieces on the board. Joe Hoane, one of its developers, said, “It is not an artificial intelligence project in any way … we play chess through sheer speed of calculation, and we just shift through the possibilities and we just pick one line.”
- We don’t know what consciousness is, and we can’t even agree on a definition of it. How can we define a whole category of AI on something so nebulous?
I think a more useful categorization of AI for business executives is:
- Generative AI. To paraphrase Wikipedia, generative models of AI create text, images, video, audio, etc. by learning the patterns and structure of “training data” and then generating new data that has similar characteristics. Google Bard and ChatGPT are the two most popular AGIs (Artificial General Intelligence) in the world right now. The capabilities of these models are jaw-dropping.
- Enterprise AI. This is AI that is trained not on internet data but on your own company’s data. EAI is designed to answer any question about your company, customers, transactions, products, services, etc. with great precision, faster than any human could. Anything that exists in your databases would be immediately accessible by anyone in your company who would get instant answers, even if it required the model to search across datasets to find them.
- Applied AI. This is AI that is designed to accomplish a specific task. Almost every technology company claims to be using machine learning for one or more of its key capabilities: ERP, pricing, inventory management, CRM, marketing automation, AR management, rebate management – you name it, there’s a system that is using AI today to help make your company better, more profitable, more efficient and more customer friendly.
My view is that Applied AI holds the most short-term promise for distribution companies and yet most executives don’t know what’s available or how it might benefit them.
AI Can Help Distributors Right Now
As AI becomes more mainstream, business executives will regularly be surprised at the capabilities it offers and how they can exploit emerging technologies. These are nascent technologies, but if you’ve tried them out, I assume you’re gob-smacked. And remember – AI gets smarter and it does so at an accelerating rate. That means these systems will improve their capabilities faster and faster, so what will they be able to do in a month, six months, a year or five years?
Predictions are all over the place, and many technologists are scrambling to catch up because they thought the capabilities we’re seeing in these new LLMs were years away. After decades of AI capabilities lagging behind their magical promises, they’ve leapfrogged beyond what most experts thought was possible at this stage.
Hundreds of AI solutions are available right now for distributors to plug into their tech stacks and databases to improve productivity, get immediate and better answers to business problems, and enhance customer satisfaction. If you aren’t evaluating and testing some of these technologies right now, you are already behind industry leaders.
We’re producing a conference on AI for distributors in October and registrations are going quickly. The attendees are a great blend of C-suite executives and technology leaders of various types. Some major distributors already have dedicated AI leaders and they’re sending them to the conference.
A great way to get started or to speed up your AI understanding and adoption is to attend the conference yourself. It’s relatively inexpensive ($995 if you sign up before the end of August), and you’ll gain access to technologists from legacy software companies and start-ups who already offer AI capabilities designed for distributors. Additionally, we have an impressive group of panelists and keynote speakers, including Zack Kass, the Head of Go-to-Market for OpenAI, the makers of ChatGPT.
But the best reason to attend the conference is to meet with other thought leaders in distribution who are already working on strategies to improve their business results through evaluating and adopting AI. This is the first-ever dedicated live conference on AI for the distribution industry and the energy, information sharing and networking will be fantastic. Don’t miss it.
Get more information about the conference and register. See you in Chicago in October.
Ian Heller is the Founder and Chief Strategist for Distribution Strategy Group. He has more than 30 years of experience executing marketing and e-business strategy in the wholesale distribution industry, starting as a truck unloader at a Grainger branch while in college. He’s since held executive roles at GE Capital, Corporate Express, Newark Electronics and HD Supply. Ian has written and spoken extensively on the impact of digital disruption on distributors, and would love to start that conversation with you, your team or group. Reach out today at iheller@distributionstrategy.com.