Will ChatGPT Change The Way We Work?
Firstly, I’d like to say, ChatGPT is not magic. It’s just an algorithm, but a very powerful one. Trained on massive amounts of data, it can generate human-like text in response to prompts. It can answer complex questions, but it’s not intelligent in itself, like a human. There is in fact no intelligence in Artificial Intelligence. There is mathematics and training. However, this is something other than the usual chatbot you meet when you’re contacting your bank or insurance company. Because it’s trained on so much data, it goes much deeper in its understanding, and then it answers. And it even has some degree of creativity, although it just mixes what it has learned.
There are already a lot of business use cases such as generating content for marketing and advertising, and web content in general. And you could imagine use cases in customer relationship management, giving precise and personalized answers to specific questions raised by customers.
Another interesting use case is code development and automation. It’s a bit like having a junior programmer at hand. It gives you something you can use as a draft, and that saves you some time. On the other hand, there are already examples of people trying to use it to develop malware. If you ask it to analyze a website from a security point of view, it gives you all the steps to test its defenses. So, if you’re a beginner in hacking, it can teach you how to attack a website.
Seen from a business perspective the crucial point in all this is return on investment. If you decide to invest in such a model for your specific business case, you must ask yourself, how much will it cost to train it and to run it afterwards, and does the cost make sense compared to your gain?
But even more importantly, when I look at our clients, conversational AI is far from the top of their priorities. That’s because the essential prerequisites for all this are digital continuity and clean data. Yes, you can do complex data science and analytics, and you can build powerful algorithms trained on large amounts of data. But only with good data, and getting good data is 80% of the work. Our clients have not yet reached sufficient maturity in their use of data. They need the right tools and data platforms, and they must find ways to connect their IT systems, their Internet of Things (IoT) environment, their machines. Let’s first break data silos, automate data flows and lean on data to improve the business processes, and bring the right information to the right users. Only once this vital work is done, will it start to make sense to implement more complex tools.
“ ChatGPT is not magic. It’s just an algorithm, but a very powerful one.
A lot of people are interested in ChatGPT, and I definitely share their interest. I can see the big companies flocking around it as well. Regarding the application itself, it’s important to split it in two halves. One part is the content it’s distributing, the other is the way it’s presenting it. I believe there are some issues around that combination.
Firstly, the content is not really fact-based. ChatGPT lies a lot, actually. It constructs stuff, so it’s not a reliable source of information. It has no relation at all to facts, and it’s not a curated source of information. It constructs information from what it has seen most frequently. The most popular phrases are being rephrased using a language model, which is actually quite good, as it can write about topics in a human-sounding way.
That’s part of the attraction, but it’s part of the problem as well, because it’s not necessarily fact-based and correct. And users could even perceive ChatGPT as more authoritative than other sources, because it produces text that reads as if it has been written by a human. Moreover, I think we’ll soon see some legal issues emerging, for instance when people use it as a medical or financial adviser.
However, it will be interesting to see how Microsoft, Google and other big players are going to monetize this type of conversational AI. In my opinion, the race for ChatGPT is probably going to be a monetizing race, following the well-known three phases of platform development. First you attract users, then you bring in advertisers. Reaching the third phase, the main goal of the platform will be not to serve users or advertisers, but to make money itself, and present the content that makes the most money. We’ll have to wait and see how that goes.
If you ask me which business cases I see in my own line of work, maybe the servicing of complex systems could benefit from the conversational approach. If the people servicing a highly automated factory, instead of browsing through manuals, could type their observations into a chatbot, it could guide them in the right direction to fix the problem. It could be a kind of documentation portal for a complex system, provided it has been trained on the relevant data. But actually, that’s always the problem with AI. It requires large amounts of data to work properly, and when you look at specific use cases and specific industries, you rarely have the data volume necessary. Furthermore, it’s not curated, so it could be a source of error for the AI system.
I understand why many people are fascinated by ChatGPT but to use it in business, it needs to be directed somehow, and I think you will struggle with getting the data you need to support your business case. So, although it looks promising it might fade away anyway. We’ll just have to see how it goes.
However, when we look at AI and machine learning in general, by far the most popular applications are camera- and image-based. It’s well-established in face recognition, surveillance, and crowd control, and it’s used extensively in industry for production and quality control. There, you don’t necessarily need a huge data set to build your application. A few thousand images may be sufficient.
In fact, I find AI-powered image creators far more interesting than their text-based counterparts, for instance, DALL-E, developed by the same people that are behind Chat GPT. That’s interesting because the system doesn’t just repeat what it has seen most often. It starts experimenting, and synthesizes stuff. For creative people, such a tool must be a great way to get new ideas.
I do share in the fascination over ChatGPT, not just from a technology perspective, although I under- stand the technology and how it works underneath the covers. What I find absolutely fascinating is that this is the first time I’ve engaged with an AI-type technology with responses so similar to what I would expect from a human. That is shattering the sense of reality most people have, because there’s that real fear of robots and what their impact on society will be. And of course, we don’t know, which makes it both scary and exciting.
Yes, I’m very interested in its impact seen from an employment perspective. In general, when new technologies come out, many people’s immediate reaction is fear of losing their jobs. But history tells us that’s a temporary thing. After a while we learn to understand and apply new technologies in certain ways, which leads to further developments and then more jobs are created. ChatGPT is going to be no exception. I think we’re going to see a blip in job losses in the initial phases as the world tries to work it out and early adopters bring the technology in. Then we will see a whole raft of employment opportunities coming.
As an IT professional, I see a significant impact on coding. I run a development team, and we’re experimenting with the capabilities of ChatGPT. One thing is that it’s like having a junior programmer on your team. You can ask it to give you a template that does a certain activity, and it provides you with a draft. But we don’t find that to be the most useful part. What we find really useful is code explanation. Sometimes ChatGPT will come back and highlight edge cases or bugs that we just didn’t think about because we became quite myopic when we were focusing on solving a particular problem. So that’s something we’ve found to be very useful and interesting.
The other thing that we get a bit of a time dividend on, is commenting and notes. Rather than spending a lot of time putting detailed commentary around why something is occurring or how something’s functioning, we can now use ChatGPT to provide some of that commentary for us. This is good code hygiene, and moreover it frees us up as programmers. Now we don’t need to explain what the code is doing, because the machine does that for us, but we can actually change our thinking and say, well, what was the intention behind me doing this with respect to the problem I’m trying to solve? It’s that shift of thinking that ChatGPT is affording us, because we don’t have all of the time in the world. So, when we’re focusing all of our time on solving the problem, getting the code written and trying to put as much comment as we can within a certain time frame, we’ll often forget maybe the most important part–what was the context and why was I solving this problem in the first place? Using these sorts of technologies means I can just focus on doing that. That causes a mental shift, and that I see as being very exciting in this space.