an A.I. running on safenet would solve the issue of privacy and user data collection!! 
2 Likes
Keeping data private takes a lot more than anonymisation and sitting it security.
Firstly, de-anonymisation is largely doable if you have large enough datasets to cross reference with the anonymised data. So much of it data has already been collected and sold that Iām skeptical of any claims around anonymisation and this is regularly confirmed by privacy advocates. FYI @oetyng
I think medical data is a particular concern because it is both very sensitive and highly valuable commercially.
Secondly, at least with LLMs anything you input is stored and, so far has been shown to leak through the UI itself, even without other considerations.
Also note that those developing this tech have no interest in hampering it with privacy measures, and many will already have form in this area.
5 Likes
Yeah, it requires that actual medical data is not already out there in the public, with that in turn having ties to an identityā¦ Then it can of course be correlated, depending on how much that is.
Here in Sweden that sort of data is tightly kept. So it would have required security breach, or that the patients themselves leak it. (But itās actually not so easy to extract for a patient either, with the systems we have for it, so requires determination to leak it.)
1 Like
You donāt need other medical data to de-anonymise supposedly anonymised medical data. Many other data sets can be correlated.
Example: if your anonymised medical data contains the dates, even approximate, of your consultations or procedures etc. and somebody has access to location data it becomes trivial to focus in on a given individual.
1 Like
Yeah this is right. With actual medical data, I meant specifically just medical data, and with ties to an identity, there I meant exactly such things as information of times/dates that can be correlated with public transportation, face recognition cameras, social media etc. etcā¦ (Times/dates are essential for medical data as well though, so if one leaked then probably that as well.)
1 Like
To my thinking the real solution will be to have local systems to medical centres allowing the centre to fully control the access to individualās data. Added the laws preventing them from inappropriate use of the data.
Otherwise if its centralised like chatgpt is now then smart questions and posing scenarios will tease out information on individuals even though the medical AI is supposed made to not leak info.
Having it local then only the local medical staff has access and they are restricted by law.
Also to me the only way it can be done on the Safe network is for the personās medical data is stored in the personās storage. Only if the person gives permission can the data be stored elsewhere and potentially for other uses (eg studies, aggregation, whatever)
3 Likes
Maybe train an AI to fully and completely anonymize data, not be teased or tricked to give out anything, and then feed that result into the health analysing AI? 
2 Likes
Maybe use AI to stop yourself getting sick in the first place?
Problem with medical data is when something happens to you, you may really want to give access to the somebody (ambulance crew for example), but you may not be able to do that at the moment.
1 Like
Yeah, I donāt disagree with this sentiment.
One possibility is, if new jobs donāt replace those made redundant by AI, displaced humans could concentrate on being more social and just pursuing their own projects/hobbies/interests, rather than working. I myself think I could lead a pretty fulfilling life doing my own projects, but we would need infrastructure for people to not have to worry about having pay cheques, whether that be UBI or something else (I donāt know what the answer is).
4 Likes
Just to be clear btw, I didnāt really have ChatGPT in mind when I said that, but some other more specific AI that doesnāt exist yet, but was tailored much more toward healthcare.
The nature of healthcare could go very different in the future though. With use of nanotech, itās possible it could become very largely preventative in nature, where some little devices implanted in the body would just detect the onset of problems before they develop into anything serious. Who knows?
2 Likes
Given the experience of the last few years I have no desire to let anyone inject anything into me anymore.
I donāt even take an aspirin if I have a headache.
Pharmaceutical companies have no interest in making you healthy or preventing you from getting sick. They just want long term medicine dependence.
Iād rather eat well, move around a lot and lift heavy stuff.
5 Likes
Why would they? a patient cured is a customer lost.
3 Likes
Which is why ALL health care including pharmaceuticals MUST be removed from private hands.
This is where the Scottish and English NHS failed - we allowed the parasites to continue screwing the people for the price of the drugs. An other so-called āfailingsā of the NHS are due tolack of funding because OUR is money wasted on supporting an inefficient and top-heavy military, lazy molly-coddled civil servants and tax breaks for the rich. Plus the obscenity of the upcoming Royalist wankfest.
And if it requires a strong state to achieve that, then so be it - before the anarchists start their wailing.
5 Likes
There are also these dogs trained to smell cancer etc. Imagine animal skills and AI combinedā¦ Something that we have used to think below and above us now widening the scope.
1 Like
My ChatGPT experience thus far has been mixed. It is pretty bad at somethingās Iād like it to doā¦ like an undergrad level assistant that needs to be asked to do the same thing over and over, and then repeatedly makes the same mistakes again over time.
But then, for other things, I just love it, like:
3 Likes
Gpt-4 is much better at self reflection and knowing what it knows than gpt-3.5. In many cases the initial answers of gpt-3.5 and GPT-4 may be similar, but if you ask it to criticize its answers, reflect on whether it the fullfilled the prompt or if the answer can be improved, it does way better than gpt-3.5.
3 Likes
I mainly use GPT4, and yeah you can get it to correct itself, but then a few prompts later in the same thread it reverts, and it gets tedious to ask the same thing again and again.
I think a lot of this could be helped with an improved interface though.
5 Likes
Different interfaces will make things much better.
In Langchain for example, thereās this concept of agents and different agents are good at different things. It basically allows you to automate repetitive prompting, reasking, copy pasting. We should see plenty of this soon, so you get stuff like the option to select different agents from a dropdown menu and it will do lots of background prompting for each prompt to get a better answer. Plugins in ChatGPT also seems really interesting.
With the current ChatGPT interface weāre barely scratching the surface of whatās possible
We know a song about that, donāt we Billy?
and Hale and Pace - Billy & Johnny - Shopping in Soho - YouTube
can you lot tell I am desperately trying to find stuff to do to avoid digging the garden? Recent sad events have totally traumatised me about gardening.