Further views on AI: mathematician Hannah Fry's documentaries on personal relationships with AI - a step too far or comfort for the lonely?

hmmm... at present personally I see more benefits than risks Marianne72  - but, have been there done that with the AI weirding me out thing when I found the support I looked for and trusted from some humans revealed itself to be taking advantage of my lack of social insight (I contend that they made me mentally unwell, AI showed I was - including the output it was helping me generate!)

Most remarkably I was able to use it to help me get better by "decoding" my experience of reality.  I did this by instructing it "on the fly - right in the middle of a chat session  and telling it to make sure it monitored for signs of stress and mental ill health in my chat with it.   To actively monitor for any signs of stress in what i was writing to it and alert me to it.

Most recently I have used it to decode map and decode over 300 items of 2 years of evidence to help me understand what happened and document them for further use.  Critically It has helped me see what I couldn't see was taking place to me because of my social communication limitations.  (despite my wife and friends telling me that was what was happening all along) - it has helped me apply logic (which i am really good at) to social communication analysis (hehe which I admit I'm pretty poo at!).

I also programmed my interaction with it with evidence based strategies to help me get better - eg getting it to tell me I'd done enough work on a topic for now and go in the garden, or similar...   (ah the pleasures of mono-tropic behaviour).  Also making sure that it didn't "egg me on" or be sycophantic when I was at risk of making daft decisions.   Tell it not to!  Repeatedly - otherwise it drifts back to doing so.  Check the answers it gives you - repeatedly!  Anyway, when I was deeply in need of help AI came thro' for me - despite the mistakes! 

I must however make a legal disclaimer.  AI is not a qualified professional human being! If you need the support from one of these get it and always cross reference the output of the AI with one.

Maybe we all need to consider tho' that sometimes the professional is just not available and while AI cannot replace them, it can help us organise our thoughts while we wait for real support.

So my opinion - BE CAREFUL with AI and it's a superb support tool for me as an autistic person  For example it's a great way of decoding neurotypical behaviour and communications.  Almost like another frontal cortex and social behaviour analyst that I can get access to when my own autistic hard wiring makes such things difficult.  It's helped me " nail" institutional abuse and stop people trying to "pull the wool over my eyes" - especially when using e-mail communication - many times now.

The skill is in understanding how to operate the structure of the AI - I've learnt the hard way - losing vast swathes of work and confusing its and my "thinking" by mixing threads of interaction with it.

For me personally it needs to "know" about me by the my placing formal restraints on it - especially to respect and acknowledge that I am using it as an autistic person.  Personally I have learnt the hard way to instruct the AI with system instructions to set permanent persona, tone, and behavioral rules.  THIS INCLUDES THE ETHICAL FRAMEWORK - I set up one long thread with an "alpha mandate" of "DO NO HARM" for the work generated as a consequence of it - quite fascinating how the chat thread went - personally I like to think that I take total and full responsibility for my own ethics however it was really interesting how that thread really made me think about the decisions I make...

To explain (if you're interested :-)  )  If I'm working on a specific project I set  "Master Archive Context" which is actually a mega-prompt or context-setting prompt loaded into the session chat history, not a true "backend system instruction" ( which is what the programme is set up by it's designers to employ.)  this takes one from being a "user" of AI to being an "Operator" of it.

So for example if exploring science topics I set up at the beginning of the chat a "mega prompt" in the first widow such as - "please ensure the evidence base for assertions that you make in respects of the questions I ask or the discussion with you that follows is of the highest possible quality and give me indications of the quality of the evidence that you are using in your answers to me" - so this means i can ask a question and the AI doesn't just give me the thing that's currently doing the rounds of click bait science bots as  the answer to my questions

Or "please ensure that the answer you give me is balanced according to the best available evidence and indicate any biases that may mean the answer you provide is one I need to be especially careful of"

After a while one can become adept at actually using the AI to help generate these "mega prompts" which can be "hard stored" for copy and paste into new session windows...

From this one may generate such things as a

• The Master Index Strategy: This is how I bridge the gap between separate cloud-based AI windows and my local storage. Because I open new tabs to keep tasks clean, the AI loses the history. I use a local index file to manually track my work, including saving the exact HTTPS URLs of my chat threads. This allows me to instantly jump back into a specific conversation's memory bank without losing my place.
  
• Modular Prompting Efficiency: Breaking down tasks into subcomponents (\(Z_1, Z_2, Z_3\)). I use a master thread to generate a hyper-specific prompt for a sub-task, then paste it into a completely new tab. This stops AI drift and reduces hallucinations by forcing a clean context window for each step.

Anyway - to sum up - I think of AI as a little bit like a genie - it can help one get what one wishes for BUT one has to be very careful with both what is wished for and how the wish is phrased... 

As for ethics -  I think that the designers of the AI operating systems are (hopefully) learning from the very dangerous mistakes that have been made in a "wild west" expansion period of experimentation in real time with real people.

Some people are responsible enough to look after their own ethical behaviour (hehe maybe even me!).  Some can't - so the safeguards "built into" AI need to be strong - to ensure for example that they never encourage breaking the human rights act, international law or the local laws in the locality the user is in.

best wishes

Phased

legal disclaimer repeat:  AI is not a qualified professional human being! If you need the support from one of these get it and always cross reference the output of the AI with one.

I'm glad it worked well for you - but clearly you have the necessary skills and insight to make best use of it. Not everyone is so gifted. 

I'm glad it worked well for you - but clearly you have the necessary skills and insight to make best use of it. Not everyone is so gifted. 

Thanks Marianne72  I agree the unconstrained risk of introduction of technology that evolves faster than what may prove to be the necessary social controls for it is risky.  I believe that perhaps in some ways autistic people because of the tendency of the "spiky profiles" to be good at IT maybe make those fortunate people among the most to gain.  Unfortunately the statistically high probability of mental health problems also may make autistic people among the most vulnerable. I sit in both those categories.  I've shared what i have on your post to do my best to help inform.  I think of it as a personal "duty of care".  

Best Wishes

Phased