Generative AI: Retrospective and Road Ahead

Date: 30/09/2023

It’s been about a year since OpenAI launched ChatGPT. I think its launch brought AI to people’s fingertips and the forefront of their minds, and this sparked much debate about AI and its role in society, even though many AI-based products have been quietly busy solving problems for some time before the launch of ChatGPT.

Since the launch of ChatGPT, several new Generative AI products have been launched, with the industry valued at around USD40bn, according to Bloomberg. The generative AI industry may reach over 250 products in 2032, when the industry may be worth USD1.3Tn. AI platforms generate images, text, and video and are getting good at it - actors have held strikes, and writers are concerned that their livelihoods are at risk from producers. ChatGPT passes legal and medical exams.

So how exactly have ChatGPT and Generative AI, more generally, improved things? This is what they have done for me:

1.   ChatGPT has helped me learn. It helped to explain complex concepts in language which is easier to understand. ChatGPT was helpful when studying for and passing the AWS Machine Learning Certification exam. It has helped redefine concepts in other courses I have taken, too. Generally, it helps explain concepts, and its ability to change its writing style allows it to provide explanations that work for me.

2.   ChatGPT has helped me brainstorm. When asked to review or write documents like standards, it provides valuable suggestions about what to include. Of course, it cannot know all of the context of any situation, so you need to apply your mind to using the answers it provides. You can ignore it without offending if you use it like an advisor.

3.       Perplexity.ai has proved invaluable when researching and needing up-to-date information. Perplexity queries the Internet based on your prompts, finds relevant information and then provides a concise summary of the information, with references to each page it finds. Generally, I have found it works well.

4.       Scite.AI has also proved valuable for research. Scite looks for relevant academic papers and presents a summary of them. Paywalls somewhat hinder it, and sometimes the responses don’t match what I want, but they provide a fast way to find information.

5.   ChatGPT has proved helpful for doing creative tasks like writing Odes to Load Shedding – a deliberate terminating electrical power to suburbs at various times, designed to prevent a total blackout. The retelling of The Shining spellbound my Facebook friends as The Scratching features a white cat, an evil litter box, and a Meowtel. Creating the prompts to produce the screenplay or part of one was fun.

6.   Tools like Dall-E and Fodor produce images, some of which were good. These tools struggle to add actual, understandable text, but other than that, they make reasonable images.

7.   The biggest thing for me is that it provoked a lot of thought about AI, how it works and how to secure it. That helped me find new and exciting things to get excited about, which made for some interesting work.

Does Generative AI have a future? I believe it does. I think people will use these tools to be more productive, and it will change the struggle to plough through data to find reasonable answers. Of course, becoming dependent on such tools is a cause of concern, and accepting everything it says at face value might lead to unfortunate outcomes. Someone was chatting with me about blockchain and innovation. Blockchain has always felt like a solution without a problem, and its biggest claim to fame is Bitcoin and other cryptocurrencies, undermined by scandals and financial losses. ChatGPT and generative AI are different. They solve people’s problems at a more practical level. You don’t need a secure wallet to use ChatGPT. You need to ask questions and be mindful of the answers you get back. With API, you can request many questions and get many responses quickly. Big companies are bringing out copilot solutions aimed at increasing productivity. Github, Codex, Polycoders and CodeWhisperer copilot help developers, and Microsoft 365 copilot helps knowledge workers create documents and presentations. Perplexity found 23 copilot-type products, and I am sure there are more.

The downside of Generative AI, like Blockchain, Y2K, WiFi, Bluetooth, Wearables, tablets, drones, the Internet of Things and the World Wide Web before this, is that we have to listen to much hype as companies try to sell AI as a solution for something or other. I appreciate that they need to make money, and sometimes some good products are introduced, which we should know about. We will see if Generative AI lasts when the hype settles or some new hyped thing comes along. I think it will.

The other problem I think will be resolved is that the training data used in models needs to be much closer to real-time. ChatGPT is now trained to January 2022, which is better than September 2021, but is some way from October 2023. Some AI engines can query the Internet to augment their responses, but it would be better if AI providers applied the rigour required for loading data more quickly. Model drift is natural in this situation, and relying on old data limits decision-making.

What I like the most about these generative AI products is that they often don’t try to sell you something, which is a significant advantage over a regular search engine. Of course, you have to pay a subscription fee sometimes, but at least I don’t need to slog through many promoted and mostly useless products to get the answer I want.

Am I worried that Generative AI is coming for my job? Not right now. I see it as a tool to find stuff out and help get stuff done. It can help with things like threat modelling, standards development, and pen testing to some extent and can help brainstorm when no one is around to do that. I see it currently as a copilot rather than a threat. More importantly, it creates new jobs and ways to work, so learning remains a crucial habit for me. Of course, some people should be worried about their roles. I don’t think the US Actors and Writers Guilds were wrong to challenge the role of Generative AI in their industries. I hope that the genuine creativity of humans is never entirely replaced by the bounded invention of AI.

We should be concerned about the barriers to entry for AI – it requires costly hardware to train a model using data, which a few companies and governments can afford. We must avoid an oligarchy of generative AI providers whose world views and adverts dominate life.

I expect we will start to see regulations governing the proper use of generative AI in the future. Hopefully, these regulations will build confidence in AI technologies rather than ban them and will work to break AI oligopolies and monopolies from developing. Reasonable rules might also limit the use of AI platforms to subvert elections or create social upheaval. Companies, governments and anyone else using generative or any AI in their products should be required to declare its use and be clear about how they use AI in a moral, transparent, and socially beneficial way like they are needed to tell you how they will process your data. Some organisations and governments will use these technologies to promote their world views and spout propaganda, and we need to guard against this.