BLOG PODS #21 - AI: Its Use in Social Care
INTRODUCTION
Today, a good mate of mine-an experienced and capable social worker-sent me a link to an article. It was all about an Australian social worker using AI to produce a child protection court report - and getting sacked for it.
This set me thinking.
Of course, we know that AI is already being used for all kids of sordid and distasteful purposes from deep fakes of political figures through to the production of child abuse images. But this was a new idea to me, at least in as much as I hadn’t thought we’d be looking at a ban on using AI at work.
But here we are; someone writing a report for a child protection court hearing used AI and got it badly wrong.
A caveat
I need to preface the rest of this post with the clear statement that I’m no AI expert. Not even close. I’m pretty savvy with my gadgets to get stuff done, but I’ve never written a line of code in my life never mind even begun to understand how an AI works.
That said, I’ve been using AI for about a year or so now and have learned a few things on the way. So what I share here is a reflection of that - a metaphorical dipping of the digital toe in the vast waters of modern technological wizardry. Nothing more.
A fundamental misunderstanding…
My first reaction to the Guardian piece about the child protection worker using AI for a court report was, ‘they clearly didn’t use it very well.’
It sounds like the person in question fed confidential client information into the AI, asked it to write the report they wanted, hit ‘generate’ and then put the AI-generated text in a report that then went to the court.
This, I think, illustrates one of the misunderstandings about AI that needs de-bunking - that ‘it’ can ‘just produce’ what you want with minimal effort on your part and most people won’t twig that it’s written by a ‘machine.’
At best, this is a cursory and shallow take on what is a vastly complex and capable technology.
Mistakes this worker seems to have made
A more detailed, practical understanding of how to apply AI capabilities in everyday working scenarios, needs a grasp of one or two guiding principles:
* While AI can research, collate and organise huge amounts of information in frighteningly quick time, this is a very broad notion. The specifics of what it produces may or may not be accurate for the use-case the human user has in mind.
* Meaning, when it comes to words, phrases, contexts, cultures and professional fora (to name but a few) is hugely broad. The AI’s use of any particular word or phrase may or may not be suitable to the human purposes it was generated to serve.
* AIs produce information that is inaccurate, out of date, false and unsuitable in connection with the question/s it was asked. This brings us on to another really key point…
* Humans need to check, edit and fine tune pretty much everything they get from an AI, if it’s to serve any useful purpose in the real world.
Based on these very pedestrian ideas about AI, we have to exercise caution in how it’s used. Remember, artificial intelligences learn what humans teach them (for now at least), so those same human intelligences need to think carefully, be very wise and use the utmost discretion in what they use it for.
Lessons from Oz
Some thoughts about where the child protection worker in the Guardian story might have gone wrong:
* There’s no ‘might’ about the first one: no-one should be putting confidential information into an AI without the explicit permission of the ‘data owner.’ That’s just a massive own-goal. In the UK, adherence to this very basic rule of GDPR is a non-negotiable.
* The AI cannot make or replace professional human judgments in respect of detailed, nuanced and discretionary information about, for example, whether something does or doesn’t put a child at risk. Humans have to do that (at least at the moment).
* Presenting an AI-generated report to the court without thoroughly checking it first and adding the kind of detailed human insight mentioned above, is just daft. On a good day.
* It’s deeply disrespectful to the service user (the subject/s of the report) not to read and re-read each word, humanise it (in every sense) and add in the nuanced and detailed aspects that only a human professional who knows the family can possibly know.
So, this story is more an example of how not to use an AI, rather than a comment on the usefulness or otherwise of the AI itself. It’s the person that’s at fault here, not the AI.
The good stuff
Leaving aside the shortcomings highlighted in this particular example, there is a case to be made for employing AI in any work context where sifting information, organising text and conducting research are needed (to mention just a few).
Emails
Every week, we all write dozens and dozens of emails. AI can absolutely help with this.
Often, these kinds of things can be done with absolutely no threat to confidentiality by observing a couple of basic rules:
* Never use someone’s name or include any identifying details - insert a letter or a pseudonym instead (e.g. ‘X’ or ‘Snow White). Or just don’t use a name at all - often it’s not necessary.
* Think less about the AI producing what you want and more about it drafting something you can fine tune to use. Here’s an example:
* Prompt the AI about the purpose and meaning of the email only. For example:
‘Write me an email to a health professional of senior standing asking firmly, but politely, for a response to my email. Point out that it’s now over a week since I wrote and I need a response urgently. Use a tone which communicates politeness, respect and urgency.’
* Here’s what ChatGPT 4.0 came up with using exactly that prompt:
* Pretty good, eh? And definitely a timer saver. No confidentialities breached, no-one’s data passed on without authority; just a really efficient use of the technology to get something done.
And it took 5.5 seconds! In fact it took me a lot longer to prompt the AI than the AI took to generate the answer.
But that’s how it should be - the effort we put in pays off, but we have to do the work at the front end and then again afterwards, to make sure we get the best out of it, keep everyone (and their info) safe and don’t break any laws.
Self-help
If you’re new to this, or even if you’re not, we can all learn a lot about writing good prompts by asking the AI to help us improve. So draft your prompt and then ask something like this:
You are a prompt engineer. Review the following prompt, optimise it to make it better and ask me any questions you have before generating your answer. [then insert your draft prompt]
The AI will then fine tune the prompt make any adjustments to improve it and you can use that to generate what you need.
This short extra step in the process can help us become better prompters, save time on the front end and generate more targeted responses - win win win!
A myriad
This is just scratching the surface of how generative AI can help. There are a myriad others, too. Here are some of them:
* Automate Repetitive Tasks: Handle data entry, scheduling, and other routine tasks.
* Answer Emails Quickly: Draft responses or even respond to common queries automatically (see above).
* Personalise Communication: Tailor emails or messages based on the recipient's preferences.
* Summarise Long Documents: Create short summaries of reports, articles or meeting notes.
* Generate Ideas: Help brainstorm ideas for projects, presentations or content.
* Analyse Data: Quickly find patterns and trends in large datasets.
* Write Reports: Draft reports based on available information or data (but not in the way our Ozzy colleague did!).
* Create Presentations: Generate slide content based on topics or data provided (I’m itching to try this!).
* Customer Support: Answer FAQs or chat with customers through AI-powered chatbots.
* Schedule Meetings: Coordinate calendars and suggest optimal meeting times.
* Translate Languages: Translate documents or emails quickly and accurately.
* Edit and Proofread: Check for grammar, spelling, and tone in written communication.
* Research Topics: Gather quick information or insights on topics for reports or planning.
* Manage Tasks: Track to-dos, remind you of deadlines, and organise tasks.
* Generate Graphics or Visuals: Create charts, infographics, and simple images (like the one at the top of this page).
FINAL THOUGHTS
So there you have it - I just about managed to resist getting in too far out of my depth!
AI has tremendous potential to offer when it comes to improving efficiency, taking care of laborious repetitive tasks, processing mass datasets AND saving us time. It can do things more quickly, thoroughly and consistently than we can. But not everything. Not yet, anyway.
It can’t meet someone, build trust, form a human connection and use that connection to bring healing, companionship and recovery. It can’t inject an assessment or court report with the uniquely person-to-person empathic dimension needed to really understand someone’s needs. Only we can do this.
So IF we decide to use AI at work, and if we manage to do so without breaking any laws or betraying any trusts, let’s remember that we have to front load it with effort so that it’ll produce something helpful; then let’s check it, change it and improve it by bringing some humanity, professionalism and kindness to it so that we retain the thing that matters most - the human dimension.
See you in the next one!
More information:
* WEB ARTICLE: ‘AI ban ordered after child protection worker used ChatGPT in court case.’ (link)
* WEB ARTICLE: Internet Watch Foundation - use of AI to produce CSA imagery - (link)
* WEB RESOURCE: Artificially Generated Child Sexual Abuse Images - update from the Centre of Expertise on Child Sexual Abuse (link)
* BOOK: The Coming Wave - great book by one of the father’s of AI (link)
* BOOK: Nexus by Yuval Noah Harari - just brilliant! (link)
Subscribe & Follow?
* You can join Jonny’s mailing list here. Your information is safe and you can unsubscribe anytime very easily.
* If you want these posts sent straight to your inbox, click the blue subscribe button below.
* You can also “Like” this site on Facebook and “Follow” on Twitter, Pinterest or connect with me on LinkedIn.
©️ Jonny Matthew 2024
