There has been a recent renewed buzz about AI (artificial intelligence), including recommendations to small business owners that they should be considering using AI to keep up / get ahead.
The buzz is a result of the November 2022 launch of ChatGPT developed by Open AI, which uses the large language model (LLM) of machine learning and is an example of “generative AI” – meaning that you can ask it a question, or give it prompts, and it will generate information for you to fulfil your request. For example, you can ask ChatGPT to write an essay on a topic or explain a concept to you.
On the back of this buzz, the media is full of recommendations of use AI tools or be left behind. ChatGPT isn’t the only one of these types of tools — Google is developing their own tool called Bard; Baidu is planning to launch ERNIE Bot next month.
But are these tools *really* going to help your business?
To answer that question, this blog considers 4 risks of using publicly available generative AI tools.
NB: every aspect of this blog has been written by a human!
- Ownership issues
Who owns the content produced by generative AI? Does the generative AI tool own it? You could argue that you typed in a unique set of prompts for the tool to produce the outcome. But the AI tool will have a log of what was produced, and they have invested in the creation of the tool. But what about the information that the AI tool trained on? Should the authors of the training material rightly be the owner of the output? The content that is produced by generative AI tools might be visible within the tool by any user of that tool … so can anyone use it?
To help schools determine if the work of their students, and organisations determine if work of their employees, was actually written by them, as opposed to really being generated by an AI tool, another AI tool — Originality.ai — is being used to check.
Lawsuits have commenced against Microsoft, ChatGPT, MidJourney and others, due to their use of information on which these tools have trained, without consent, and when used in the generated content, without credit.
- Up to Date?
LLM AI tools such as ChatGPT are trained on very large data sources, and this training takes time. Therefore, depending on your use of the AI tool, the results may be out of date, or at least, not considering the latest data in a particular field of knowledge.
ChatGPT was trained on a dataset with an end date of October 2021, and therefore it cannot tell you anything about the invasion of Ukraine, or countless other events that have happened since then.
- Uniqueness
Publicly available tools such as ChatGPT generate results based on a text question or prompt you enter. There’s no stopping another person asking exactly the same question and getting a very similar result. If you are using the tool to generate content for your business – blogs, articles, proposal content – then you run the risk of publishing content that is not unique.
A more important question to ask is, do you really want AI to do your thinking for you? Wouldn’t it be better that your content is authentically created by your organisation?
- Accuracy
ChatGPT has been shown to be inaccurate for certain types of problems, including mathematics. ChatGPT doesn’t know what it doesn’t know, so the information it generates will still sound very authoritative, even when incorrect. So how will you know if it is accurate or not?
So How & Why Should My Business Be Using Generative AI?
The above are 4 very important risks that you need to be aware of, so then why are some suggesting that generative AI will benefit my business, and if I am not using it, I am going to miss out? Or asked another way, how can my business use generative AI safely?
Obviously, a specific answer will depend on your particular business and market sector, but in general terms, I can see uses where all 3 of the following are present:
– the results do not need to be unique, and ownership is irrelevant
– accuracy is unimportant e.g. you require purely creative content, OR you know what data the tool trained upon, and that is sufficient to answer your question
– you will save a lot of time e.g. you need to get specific information, fast, and doing so would normally take you a large amount of time
To give you a concrete example, I was asked by a Google representative to provide him with written feedback to give to his boss. I did that … I’m always honest when giving feedback, and perhaps terse, sticking to the point. I am Aussie, afterall. Anyhoo, after receiving my feedback, he asked me to give him written feedback again. Turns out, my feedback didn’t have enough information / superlatives in it. Frankly, I didn’t have the time to generate longwinded feedback that was going to impress his foreign employer. So, I turned to ChatGPT and asked, “please write feedback for a digital marketing consultant”. 30 seconds later, bingo, I had a 5 paragraph beautifully written feedback piece. I spent another couple of minutes adjusting it to have correct facts in it, and I was done. I figure that saved 15 minutes of my life.
This is an example of the use of an AI tool for creative content, which could be quickly adjusted for accuracy, saved significant time, and uniqueness & ownership were unimportant. I won’t use ChatGPT everytime I need to write feedback, obviously, but in this instance it was good.
Another purpose I can see for our business is the creation of images to accompany blog posts. I often have in mind a particular image that I think would perfectly reflect the content. I also like all images used in a particular Contactpoint News to match each other, both in style and colour scheme. Creating or sourcing images can take hours for each Contactpoint News edition. Tools such as MidJourney may be a great assistance in this regard – in fact, I used MidJourney to create all the “pastel painting” images contained within our latest Contactpoint News, in which this blog post will feature. MidJourney provides its users with ownership of the generated images, but they can also use those images without restriction, which doesn’t worry me as we are not selling the images as works of art! Although, I’ll be watching carefully what happens with the lawsuit against MidJourney.
Using MidJourney for this purpose demonstrated that whilst the promise of AI is attractive, you will need to spend time, like with any new tool, in getting to know how to use it well before you can reap the benefits.
A programmer could utilise ChatGPT when searching for very specific code command e.g. using language ‘x’ how do I achieve ‘y’? Before using ChatGPT for such a purpose, the programmer will need to be confident that ChatGPT was trained on best practices for that language. In such a case, receiving a single result may be much quicker than numerous pages of Google results that need to be browsed through (perhaps including ads) to find a useful result.
Speaking of Google, you may have heard that the Alphabet Inc (their parent company) share price fell by 100 billion a few weeks ago, after a failed demonstration of Bard to their shareholders. There is a suggestion that Google can see the writing on the wall for their search engine and want to sure up their future role in search with their own AI powered search tool. Microsoft is a major shareholder of OpenAI (the creators of ChatGPT) and on the 7th February, launched a new version of Bing which is powered by ChatGPT which apparently has access to more recent data than that used to train ChatGPT. Watch this space!
[Update 27.02.2023 – criticism of generative AI has now also been levelled at Microsoft’s implementation, which you can read more about in the ACS article “The AI Revolution isn’t going so well. Microsoft forced to rein in a lying, angry Bing.“]
There has been plenty of conversation in recent years about generative AI being used to write blogs. News articles, particularly in the area of sports journalism, are being written by AI even now. We certainly won’t be using AI to write our blogs. I can easily imagine a world where AI generated content will lead to even less trust of online material. Surely journalism is under pressure enough already without using bots to write the news. Or perhaps, they need to use AI to get high quality news out more quickly than the competitor newsroom?
Other types of AI tools
We have been talking specifically about tools in the category of generative AI.
Thinking more generally about AI, within your business you might already be using AI without realising it. Applications increasingly use AI techniques such as machine learning, to provide better features. We use Hubdoc, recently purchased by Xero, which uses AI to perform data capture of scanned / PDF invoices and receipts to extract key information from these documents and re-create those within your Xero financials, greatly reducing manual data entry. Most smart phones messaging apps now suggest answers to your latest SMS or chat message, to help you reply quickly and without needing to type your answer, and these suggestions learn from the way you usually respond.
Do you have a need to publish your blog posts as audio? There are low-cost AI tools such as DupDub and Murf.ai that very quickly produce voice-to-text of text you supply. You can give them instructions for where to pause and where to give emphasis, and you can choose from a range of voices. Some of these AI tools allow you to upload your own voice to train it, and it will then produce voice for new material that sounds very much like you. This will be a huge time saver compared to manually creating your voice recording.
Conclusion
In summary, low-cost generative AI tools may prove very useful for your business, but you need to carefully consider the risks described above.
If you have an idea that could solve a real-world problem by quickly analysing vast amounts of data in reference to information supplied by a user, then an LLM type of AI solution which has been specifically trained on the right data for your purpose, will be worth considering. Creating such a tool will likely require significant investment to source the training data, train, test and commercialise the tool.
Do you have an idea for a custom technology solution, that you would like to discuss? We’re here to help.
Further Reading:
- “Generative AI won’t Revolutionize Search Yet” – HBR Feb 23, 2023.
- WeTrade Group plans to access Baidu ChatGPT-Style product “ERNIE Bot” to improve the efficiency of community precision marketing – Yahoo News, Feb 10, 2023.
- “Alphabet shares dive after Google AI chatbot Bard flubs answer in ad” – Reuters, Feb 9, 2023.