A colleague brought to my attention the ChatGPT natural language processing (NLP) model sharing his surprise at how well it emulates a real conversation and how good its answers seem to be. As the recommendation engines work these days, my Google Newsfeed showed me The New York Times article: A New Chat Bot Is a ‘Code Red’ for Google’s Search Business. Reading it, I immediately knew that ChatGPT would benefit a lot from a good monetization strategy. Researching the subject, I didn’t find one made public yet, so I thought this would be a nice challenge. My thoughts on it are below.
Curios by nature, I decided to look deeper into it and also gave it a try. With this being said, let us get started.
Disclaimer: What you are going to read below does not reflect the point of view of my current company, nor the point of view of OpenAI. I have no contact with OpenAI employees, and neither hold any kind of monetary interest in the company.
What is ChatGPT?
ChatGPT is a natural language processing tool based on AI technologies that aim to optimize natural language processing models for human-like dialogue. The language model can answer questions, and assist you with tasks such as composing emails, essays and code. It is developed by OpenAI, a leading AI research and deployment company. Per the company website, its mission is to ensure that artificial intelligence benefits all of humanity.
Good mission one would say, but the history of the company tells us a different story. It was founded as a nonprofit organization by Elon Musk and Sam Atman. Of course, in 2019 “The AI Nonprofit Elon Musk Founded and Quit Is Now For-Profit OpenAI LP is now a capped-profit company “, while the excentric celebrity still holds a share in it – did you think that guy does anything that does not bring him a profit? I don’t think so.
Long story short, I would define ChatGPT as a showcase of current capabilities for AI-powered chatbots and NLP models. Although still under testing, and fine-tuning, the fact that OpenAI released it as a demo, allows them to build a brand around it that later they could profitably monetize. Even from the technical perspective (the software development release cycle) the move makes sense, they get free testing, at scale, indeed at the cost of the Azure infrastructure, it runs on.
What could ChatGPT be used for?
You already have the answer – as a chatbot. I was looking into chatbot implementations since 2015. Back then, it was foreseen that contact centre services would be progressively replaced by chatbots and there were some companies which already did that (mostly based in Asia – yeah – the leading technology advancements happen first there nowadays). Even proposed in 2017-2018 a solution based on speech-to-text and text-to-speech to progressively develop an automated IVR based on an NLP chatbot. The technology was far away from being production ready back then, also proved by the below hype chart:
Source: Gartner’s Hype Cycle for Artificial Intelligence 2021
You might say “Look, chatbots are at the lowest point of ‘through of disillusionment” and you would be right, but as the article Chatbots at their lowest point according to Gartner (and why this is great!) argues you need to think of what comes next?
The answer is simple, the Slope of Enlightenment and the Plateau of Productivity. And this is where ChatGPT plays – in the Slope of Enlightenment. And guess what? ChatGPT gained 1 million users in under a week. Here’s why the AI chatbot is primed to disrupt search as we know it
If you are a visual person, perhaps the graph from this article will resonate with you: ChatGPT takes the internet by storm, bad poetry and all
All of this tells me the excitement is back, and that could only mean enlightenment.
How could ChatGPT be monetized?
I won’t make here an analysis of the benefits AI-based chatbots could bring to various companies from all industries and the cost reductions implementing an automated customer service solution would mean for them. Monetizing ChatGPT as a B2B solution that could be sold as SaaS through exposed APIs, is a no-brainer. You can already see this on the OpenAI site for other products they’ve built: https://openai.com/api/pricing/
As per the New York Times magazine: A New Chat Bot Is a ‘Code Red’ for Google’s Search Business e should consider the B2C cases too, or more exactly the multisided platform business case of Google
Although, I tend to agree with this research article Multi-Platform Strategy: A Case Study of Google, don’t worry we won’t go into this academic debate here. Let us keep it simple.
In my view, ChatGPT has the potential to play into the Conversational AI and Virtual Assistant emerging markets. Look at the below graph for predicted use cases for chatbots:
In the short tests I’ve done on ChatGPT I tried exactly these ‘Making a reservation’ and ‘Buying basic items’ starting from asking for the recipe for a specific traditional meal. I did this naturally before reading the above-quoted article.
Perhaps yes, with the right platform/multi-platform business model, ChatGPT could be free of use, we humans would be the product and at the price of convenience, we’ll give it more and more data that OpenAI could use. Yes, they would make a lot of money out of it, but not by selling the data, by exposing APIs that would allow B2C customers to channel their services through the personalized ‘Personal ChatGPT’ assistant might be the way they would be able to do that.
The actors of the platform business model would be:
- Producers: Developers, B2C businesses
- Consumers: People
- Providers: ChatGPT as a digital assistant, Customized ChatGPTs
- Owner: OpenAI
Producers could become Consumers and Consumers could become Producers too, this would drive adoption, wouldn’t it? Entrepreneurial people usually jump into new ways of creating income pretty quickly.
Multiple other actors could be added to the platform in various interchanging roles.
Of course, the convenience would need to be fueled. Taking a look into the IoT-related use cases of Alexa, Siri and Google might accomplish that. That could go even into robotized physical implementations of digital assistants – wouldn’t that be cool C-3PO and R2D2?
Reality check, getting our feet back on the ground
The technical complexities of such an implementation at scale would sky rock. Still, it’s 100% achievable with today’s technology. The problem would be funding the solutions, a combination between organic growth and VC funding might address this, though it might not fit the current legal entity as “capped” for-profit, with a profit cap set to 100X on any investment.
Per OpenAI, ChatGPT was trained using supervised learning, followed by reinforced learning to train a reward model and optimize the policy. Quite basic techniques if we look at the current research in this field. Have a look at the following article that showcases 14 Types of learning
I am not an expert in the field, but what about Statistical Interference or learning techniques such as Active Learning and/or Online Learning? They might be far reached, still, history (just a few years ago in this field) taught us some important lessons – one example being the late Microsoft’s AI chatbot Tay. It looks to me that as per now controlling the data the models train on are our only option for now to achieve an AI that consistently fakes intelligence for now.
Another major issue is the current stage of AI-based implementations. Reading ‘Real world AI’ you will find out how difficult is to make AI implementations production ready. Some are doing it, but most fail at it. If we look at the place of AI on the Ladder of causation – Judea Pearl – The book of why – we’ll better understand ChatGPT’s limitations. One can’t expect much from Rung 1:
t least the below quote seems down to earth:
ChatGPT is incredibly limited but good enough at some things to create a misleading impression of greatness. It’s a mistake to be relying on it for anything important, but a preview of progress. We have lots of work to do on robustness and truthfulness.— Sam Altman, CEO of OpenAI
In my tests on it, it looked to me that it mimics understanding pretty well, but fails quite easily for anything that would be the first step in a Turing test
Can’t expect the same down-to-earth approach from the businesspeople that would want to monetize it. They’ll just jump from buzzword to buzzword resulting in an engineer’s worst nightmare when it comes to implementation. Still, I believe in the platform business model sketched above. I could develop it a lot further, but that would mean a paid consultancy and some RSUs, at least.
I have no clue if this is what they are aiming at. As per Reuters: ChatGPT owner OpenAI projects $1 billion in revenue by 2024. But I’m pretty sure I can build a business case based on the above to justify the projections. I’m pretty sure Microsoft knows what it’s doing when it invested 1 billion and is looking at adding to its stake.
I would like an AI NLP model trained on the Encyclopedia Britannica. I would pay for it as my assistant as long as it provides me with the answers to my queries together with the references it based its answer on. Wouldn’t that be nice product libraries and editors could sell?