ChatGPT, Machine Learning and the inevitable impact on Small Business
ChatGPT, Machine Learning and the inevitable impact on Small Business
Machine Learning is nothing new, but recent advances have turned a once interesting experiment into a cultural explosion. With the genie well and truly out of the bottle - what happens now?
Welcome to The Balance Sheet from The huuman layer - a monthly(ish) deep-dive into Small Business voices, opinion and futures. In this edition, Rob Boynes tackles a game-changing, if complex, tech-topic that has potentially huge implications for Small Business in the near future.
This edition of The Balance Sheet is brought to you by Xero - accounting software for everyday business. Try Xero for free today.
Deep Learning (DL), Machine Learning (ML), Large Language Models (LLMs) and Artificial Intelligence (AI) have been around for quite a long time. So long in fact that I've been avidly watching their steady progression for what seems like over a decade.
It’s hard not to admire the work that goes into building ever-new language databases aimed at trying to make sense of the mess that is daily human interactions. For a brief period in the late 2010’s it seemed everyone was working on, talking about or adopting Machine Learning (ML) in some way, shape or form. Since that early heyday, ML had, for me at least, moved into the background; downgraded in my list-of-stuff to ‘something of interest'; downgraded in the sense that the leaps and bounds that were made on a technical level were startling, sure, but not significant. Maybe it’s because ML was - is - everywhere, but it was hidden in everyday things - it had become so normal, so ubiquitous that it had become - dare I say it - mundane. You scan something and the text becomes magically editable? ML. Your iPhone Photos App recognizes faces? ML. You write an email and it tries to autocomplete it for you? ML. You make a Reel on Instagram and it automatically captions what you say? ML. A transaction gets instantly flagged as suspicious on your credit card? ML.
ML, it seemed, was in a hinterland of sorts, where things were progressing nicely and, sure, there were some really cool applications, but there there was no clear 'uncanny valley' moment, no ‘wow’ experience, no notable big leap which would result in a noticeable culture shift.
Last year, however, something did shift. The trajectory had changed. ML was back, baby - and this time it was significant in a way few imagined.
I'm not going to dig into the differences between Machine Learning / Large Language Models and Artificial Intelligence - life is too short for that. What I will dig into is why this new breed of ML is relevant to Small Business - because the current trajectory of these new ML products is so extreme that the things we consider normal now, won't be normal for much longer.
Is that statement pure hysterical hyperbole? A year ago I'd have said yes. Now it just feels like a statement of fact.
Here's some more facts. OpenAI, which is the Machine Learning research company founded by Elon Musk (Tesla, Twitter, Space X) and Sam Altman (Y Combinator) was established in 2015 with a seed funding round of $1 Billion USD. In 2018 Musk left and in 2019 Microsoft pledged a further $1 Billion USD. As of 2023, 8 years later, it is now raising money at a $29 Billion USD valuation. For context, Google bought the UK ML startup Deepmind (at the time the most defined and visible early stage Machine Learning research company) in 2014 for $500 Million USD.
Some big numbers and names for sure, but what this means is two things. The first is that - obviously - there is a lot of investor confidence in what ML is capable of (or will be capable of in the coming years). The second is that, with this much momentum, ML is finally having its ‘moment', which means culturally we're about to see some fireworks. But that's the investor story, and it's easy to be jaded by news stories of investors declaring something a 'big deal' - after all, we've all seen the tech boom-and-bust cycle - 'big deals' get downgraded, unadopted, and absorbed into future ideas. Tech is like that - it constantly eats its own young to create bigger and better things - and it can be hard to work out what is going to be impactful to everyday people as opposed to what is just going to make a bunch of already high-net worth individuals a bit better off in five years time.
Consider the Metaverse, and the billions ploughed into it over the past few years with little real outcome. The problem with the Metaverse (much like Augmented Reality before it) was that it was seeking to solve a problem that didn't really exist, to further a technology which produced, for the most part, a lack of useful or impactful products. The bar was too high to enter for mass adoption - purchasing a headset for $1000 USD to plug into a vertigo-inducing environment while tripping over your couch just to have a dull (if immersive) Zoom meeting that looked like Wii Tennis, was always a hard consumer ask. The same could be said of driverless cars - a problem again that didn't really exist that had too high a bar for entry - highly expensive, unreliable vehicles with a habit of collisions, an associated risk of injury, and the underlying unfortunate fact that cars (no matter how advanced) just can’t currently drive themselves when there’s non-logical humans on the road at the same time.
It’s worth noting that Apple have been quietly investing into these areas for a while, so fortunes may change. Apple do have a solid track record of slowly manipulating existing poor-fit products to fix actual people-problems with a it-just-works execution - so time (and the markets) will tell if their late-adoption curve approach will show riper fruit. Regardless however, both the Metaverse and driverless vehicles have suffered from the tech curse of being declared a 'big deal' early on, only to be perceived later as something of a bust.
So it's easy to dismiss ML products in the same breath. But that would be a mistake.
What's different about OpenAI is how useful and impactful its ML products have proven to be in real-world contexts. So useful that Microsoft are planning to invest an additional $10 Billion USD in OpenAI to power their headline products like Word, Excel, Powerpoint and Bing. So impactful that New York City Schools have already moved to ban OpenAI's flagship product ChatGPT from being used on its student networks.
To understand what is making Microsoft so excited - and NYC Schools so worried, let's take a closer look at what that flagship product, ChatGPT, actually is.
ChatGPT (currently in its third version, ChatGPT3) is an optimizing language model for dialogue or 'Generative Pre-trained Transformer'. What this means in reality, is it's a web interface where you can ask questions and ChatGPT responds and interacts in a conversational way - like a human would. You've probably had an annoying conversation with a 'bot' online before, in place of a human customer service agent, so you probably know how this all goes down. The real ‘big deal’ here is that with ChatGPT, OpenAI have created a ML model which responds and interacts in a way which is near-impossible - if not impossible in some cases - to distinguish between machine and human. A response from ChatGPT reads and interacts like a human would, and it's so human-like that some are talking about the need to create products that can detect whether the text you are reading is human-crafted or simply an export from a ChatGPT interface.
It doesn't take much imagination to realize that this immediately opens up a massive range of possibilities. Creating website copy. Writing articles. Answering technical questions. Customer support. Internal knowledge sharing. Sales. Speeches. Pitches. Essays. Research. ChatGPT is so advanced in it’s current version that it can generate useable code to build web products, provide deep insights into highly specialist topics, pass written medical exams and even adopt tones of voice (ask it to provide you with an explanation of how something works using the voice of a pirate and you'll be rewarded with something no-one should ever have to read...but it will be shockingly accurate in detail). The reason that NYC Schools are so worried - as are an increasing number of educators - is that ChatGPT is capable of producing highly articulate and accurate articles in minutes based on simple questions and prompts. It's sibling, visual model DALL-E2, which can generate imagery based on combined terms ("Create a digital photograph of downtown New York where Triassic period dinosaurs roam the streets"), also opens up a massive range of additional visual possibilities. Create any illustration, painting or photograph for free, on-demand, in any style, by simply typing a combination of words into your web browser.
Suddenly building a website for your company seems easier. Suddenly writing good communications becomes available to everyone, regardless of education and first language. Suddenly deep knowledge on any topic becomes immediately accessible.
But with this immediate access to generated dialogue and imagery comes bigger things and inevitable knock-on effects.
Think about how you search for something online. Typically you go to Google, type in a query, and get presented with ten blue links in response. Each of those links is for a website. If you've queried something knowledge based ("Is it better to lease or buy a vehicle if you're self employed") then you'll likely get ten links, each with slightly different takes, each likely linked to content, a business, or knowledge resource. You'll browse through those results, pulling together a picture of a growing grey area - are you incorporated or not? What percentage of the vehicle use will be for business? If you're a more seasoned ‘re-searcher’, you'll know how to comb through a range of results and identify what is true or relevant, taking in differing opinions and factual variances.
If you've queried something around a product or service ("Which is the best landscaping company near me") then you'll likely get ten links, each to a poll, review, business, blog. Those links will be ordered through several opaque processes internal to Google - is the link sponsored? How much traffic does this link get (i.e its rank)? How accessible is this website (i.e can it be viewed on cell phone browsers)? Google's search algorithms are complex and often difficult to decode, but for Small Businesses who thrive by search or knowledge-based content, it's something they spend serious time, money and effort on. The difference between being in the first three results on Google and being the eleventh result is huge - both in terms of brand visibility and eventual profit. Even those Small Businesses who don't sweat their search rankings daily, usually have some time-investment in being more 'discoverable' to potential customers. Being found in search is essential to finding customers, which is essential to making sales, which is essential to making money.
So how does ChatGPT change this? Well, ChatGPT3 is currently what's called an 'offline' model. This means it doesn't react to the constant flux of the internet, rather it is 'fed' information that filters from the internet and wider human knowledge using ‘supervised and reinforced learning techniques’. But placing ChatGPT into an 'online' model, or at least connecting it to live search environments is something rumoured for later this year. In the ChatGPT search world, a user will type a query and ChatGPT will respond in kind with the 'correct' answer. What this 'correct' answer will be, will be related to how the query was asked, sure, but it will be based more importantly on what the model has been fed via those 'supervised and reinforced learning techniques'. This is opaque, but it's as opaque as some of Google's ranking criteria. The difference, however, with ChatGPT, is that because ChatGPT displays responsive conversation rather than 'links', we face a search future with definitive, conversational, human-like answers. This is a search which is the equivalent of asking a friend for a recommendation or a definition - it inherently comes across as more truthful or reliable as it is delivered conversationally. Google doesn't present itself to be 'human' or ‘truthful’ while ChatGPT effectively does. How does that affect how we trust the results or answers we get? How does that change the way that we search for items, answers or companies and services? What is the impact of moving from a model where we ask "Which is the best landscaping company near me" to see ten links and pages of more - as opposed to seeing one single answer in a human-like conversational response?
The one certainty in all of this is - this is the year things will begin to change, and the one unknown (among many unknowns) is - will we still be seeing pages of links when we search for something?
It's hard to predict the near-term impact of Machine Learning products on how we access and process information online, and it's even harder to hypothesize how this impact will directly affect the way Small Business will approach search, the knowledge economy and customer acquisition in the coming years. But with ChatGPT4 slated for release later in the year, significant change is coming - you can be sure of that.
If you found this interesting, why not share it with the other small business owners in your network? They’ll join hundreds of subscribers in accessing the best Small Business insights in Canada - for free!
 | A guest post by
|