In the past 6 months or so I’ve seen the creation of AI apps and the interest in AI increase at a massive level.
So I thought I’d take a slight diversion from my usual activitie, and create personal and business automation workflows using AI.
I started with ChatGPT for text research and generation, and Midjourney for image generation.
I discovered that AI actually is and could be very useful in saving time and costs, and in helping overcome creative blocks.
What I mean by ‘actually is ... very useful’ is that AI has saved massive time and costs in writing especially text in a real use case.
If the same output was briefed and generated fully by humans then I estimated that the time and cost savings would have been at least 50 times for time and the same for costs. Let me know in the comments below if you’d like me to write about the details.
I also say ‘could be very useful’ because I’ve found errors in the generation of text responses specifically by ChatGPT.
I had to prompt the AI to double check the response related to what I requested and to update the output. It responded with an apology (yes and apology) and that it had found some more information that was requested in the first prompt, and responded with a different and fuller output.
I was so surprised by the response, I didn’t fully check the response to find out if it was comprehensive or accurate.
I stopped with a genuine concern of both the response, and the bigger picture of the general uptake of AI, especially for researching textual information.
I'll describe my concerns and the bigger picture with from my experience.
The prompt to ChatGPT was to write a list summarising every reference to the Spanish Flu in the book ‘Virus Mania’. That was a few weeks before writing this post.
ChatGPT responded with a list of about 9 points. I read through the list and it seemed OK, but when I finished reading, it felt as if some information was missing. I felt this because I’m familiar with the author’s work.
So, I prompted ChatGPT to double check that it had included all the references to the Spanish flu and to update the list.
ChatGPT responded by apologising that it had missed some references, that it had found some more, and responded with a different and slightly longer list of references.
That’s when the realisation came that no one could be certain of prompting an AI to summarise a topic especially from a large amount of text, and the AI responding with an accurate summary.
The only way I could be certain of the accuracy of the response was to check it against the contents of the whole book. With the the book being about 560 pages in length, the time for me to to double check the output would probably take longer than creating the desired summary list another way.
I tried a similar prompt yesterday and this time ChatGPT stated started with "I'm sorry, but as an AI language model, I don't have direct access to specific books...". I tried a few times but the responses were similar. What happened to previously having access to the book?
The second concern is related and about the uptake of AI by many people to summarise whole documents, and topics within the documents.
If you believe, without actually knowing it’s true, that AI will output a comprehensive and accurate response to a prompt, then you may beleive the output to be true, and then share it as true to be used.
This creates obvious concerns especially in critical uses such as law, medicine, and other highly technical fields where both accurate and comprehensive information is required to make decisions.
Without such accurate and comprehensive responses from AI, both people and educational progress could be put at risk.
What’s the solution?
For now, both my experience and intuition suggest analysing each use case for AI responses, and having checking processes that involve people to ensure complete, accurate and usable responses are provided by AI. Then the responses can be corrected, or not used, for the use case.
You may feel there’s no point in using AI text responses, but from my other real-world use tests, AI does actually save time and costs. However, human checks and editing are still required to give AI responses correctness, completeness and a human touch for the use case. Let me know in the comments below if you’d like to know more about the real world use cases.
From a broader perspective, the solution seems to require raising your level consciousness and being aware that a response from AI, whether it be text, image, audio or video, may not be complete, accurate or reliable, and that a human touch may be required in part or in full for a final use case.
That's all about text prompts in this post, however as I said I’ve been using Midjourney for image outputs, and I’ll write about the real world AI image generation use cases in an upcoming post.
For now, I wanted to demonstrate the requirement for you to raise your level of consciousness with AI generated photographic images. especially before accepting the images as accurate or reliable.
This is why: I prompted Midjourney to create an image of Martin Luther King and Donald trump in a selfie. Below is one of the images that Midjourney responded with:
If someone didn't know that Martin didn't exist in human form when the image was created, and that Donald was here in human form, then the observer wouldn’t know that the photograph with both of them could actually not be taken.
Such a not knowing observer could therefore have discussions and decisions about Donald being ‘real’ because he's holding the phone/camera and looking at it in a way an actual selfie is taken, and that Martin's just holding his phone up for his own reasons.
Hence you or any observer of an AI photo (or in fact any photographic image) should first check the accuracy and correctness of every aspect of the photo before a decision is made from the general nature of the photo.
So the question is: how do we live a life with AI generated responses that may not be accurate or reliable?
Well, I don't have a general answer, but my experience and intuition suggest that AI is going to be a part of our day to day lives and that you and I have to become responsible, and increase our level of consciousness to create and refine reliable processes for integrating AI in our lives.
That’s why I created life with AI.
Let’s work together with AI.
Please subscribe, post, and meet and learn with me and others in a life with AI.