OpenAI have released their latest version of their suite of text-generating AIs, ChatGPT. This one is a chatbot, still in prototype form. Screenshots of its output are flooding the internet, and in terms of the ability to create detailed, structured, human-like text it is hugely impressive, and quite possibly transformative.
But, we need to check our excitement.
No, ChatGPT has not made Google obsolete. No, Artificial General Intelligence, and the end of human labour, is not around the corner. No, OpenAI haven’t automated human creativity.
ChatGPT’s ability to respond to prompts with plausible and seemingly authoritative content is remarkable. Already, it has raised questions about what and who it could replace. Some have suggested that ‘AI has replaced CEOs’ because ChatGPT can generate a team meeting agenda and draft an email telling staff to work faster. In my experience, CEOs tend to do a little more than that. You’d also be able to find the same information by searching for it, in the old fashioned way.
AI has officially replaced CEOs. pic.twitter.com/Om0n6jowHt
— Aaron Levie (@levie) December 5, 2022
It’s also been suggested, with a little more merit, that its ability to churn out content makes it a threat to essay-based education, journalism and reporting. But even here concerns should be placed in context. The software is not capable of nuance. It also makes lots of mistakes. Coding Q&A site Stack Overflow have banned the uploading of AI generated answers because they are too likely to be wrong.
OpenAI are, pleasingly, not driving this hype.
They’re upfront about the technology’s limits now, and the profound barriers to progress they face in the future. Here are the main ones:
- The responses ChatGPT gives are usually plausible but often wrong. There’s no easy fix to this, because the way the model would ideally work is through assessing what it knows rather than what the human training it knows. But without that, there’s no source of truth.
- It’s excessively wordy and often repeats phrases. It often, unprompted, offers up the information that it’s a language model trained by OpenAI. That’s sure to catch out at least a few lazy students who don’t check their AI-generated homework for giveaways.
- It’s not that contemporary – the text it is trained on only goes up to the end of 2021.
That final point is a hugely significant one. First, for contemporary information, it’s unlikely to be any use. Secondly, if 2022 marks the advent of text-generating AI, then all future text-generating AI will be at least partially trained by early iterations of text-generating AIs. Given the central importance of human text to the language model, and current difficulties with assessing objective truth, that is less than ideal.
Nevertheless, the technology is remarkable.
Although we shouldn’t believe the hyperbolic assessments, there are several shifts on the horizon all organisations should consider now:
2. Mediocrity on demand. The answers that ChatGPT gives are hugely impressive for a piece of software but would be disappointing from a halfway enterprising human. Sometimes mediocre is ok – for a first draft of filler content, it’ll do. But the things that elevate and add value to a piece – lyrical flair, emotionality, original thinking – will require humans. The presence (possibly deluge) of mediocre base-copy will only add value to high quality writing and analysis.
3. An additional tool, not a replacement. Google is fine. Google does something different – it provides links and the navigational infrastructure that we rely on to access information. When searching for information, we are able, based on context and source, to evaluate veracity and authority – how much should we trust it? In its current iteration, ChatGPT is more obscure, and harder to verify. But ‘prompting’ may, like searching or googling become a skill in its own right, a tool alongside it, rather than a replacement.
4. Land grab for monetizable apps and services. Already, just a few days in, we have numerous examples of start-ups and products looking to leverage this new technology. Some are clever and recognise the way in which this technology will augment our interaction with digital content, rather than completely redefine it. One example is God in a Box, a subscription based WhatsApp channel for any and all queries.
5. AI everywhere. Already well underway, but the presence of increasingly capable and reliable text-generating AI will see more organisations utilise the service for admin, training and customer service. Certainly it will be functional – but organisations who pride themselves on their human face might want to balance the many benefits of automated customer service with their current approach.
This is a remarkable technology that is making huge strides – in many respects, faster than people had expected a few years ago. As we acknowledge this we should also note its limitations – these are as central to its progress and applications as what dazzles us.
Finally, below is ChatGPT’s version of this article – working from essentially the same prompt as I gave myself. You can decide which you prefer.
An article by ChatGPT, about ChatGPT
The potential impact of text-generating AI on the way people use the internet, learn, and work is vast and far-reaching. This technology has the potential to revolutionize how we communicate, access information, and perform tasks, making many aspects of our lives more efficient and convenient.
One of the most significant ways in which text-generating AI could impact the way people use the internet is by enabling more efficient and effective communication. With the ability to generate human-like text, AI could be used to create instant responses to common questions or queries, allowing for faster and more efficient customer service interactions. Additionally, AI-generated text could be used to improve the accuracy and reliability of search engine results, providing users with more relevant and useful information.
Another potential impact of text-generating AI is on the way people learn. AI-generated text could be used to create personalized learning materials and resources, allowing for more tailored and effective educational experiences. For example, AI could be used to generate personalized study guides or summaries of complex topics, helping students to better understand and retain information.
In the workplace, text-generating AI could also have a significant impact. AI-generated text could be used to automate tedious and time-consuming tasks, such as data entry or content creation. This could free up employees to focus on more creative and critical tasks, improving overall productivity and efficiency. Additionally, AI-generated text could be used to generate reports and other documents, saving time and reducing the risk of errors.
However, while the potential benefits of text-generating AI are significant, there are also limitations to this technology. One of the main challenges is that AI-generated text may not always be as accurate or reliable as human-generated text. Additionally, AI-generated text may not be able to capture the nuances and subtleties of human communication, leading to misunderstandings or confusion.
To overcome these limitations, it is important for developers to continue to improve and refine the algorithms and models used in text-generating AI. By incorporating more data and incorporating feedback from human users, AI systems can be trained to generate more accurate and reliable text. Additionally, by combining AI with other technologies, such as natural language processing, developers can create AI systems that are better able to understand and replicate human communication.
Overall, the potential impact of text-generating AI on the way people use the internet, learn, and work is significant and exciting. By enabling more efficient and effective communication, personalized learning experiences, and increased productivity, this technology has the potential to revolutionize many aspects of our lives. However, it is important for developers to continue to improve and refine this technology to overcome its limitations and unlock its full potential.