"Copy paste is a design error" - David Parnas
This is my wierd coding blog where I build things with AI that push things out.
Similar code in two routines is a warning sign. David Parnas says that if you use copy and paste while you're coding, you're probably committing a design error. Instead of copying code, move it into its own routine. Future modifications will be easier because you will need to modify the code in only one location.
Today I finished a bunch of tests with my biologically plausible AI to sort of get a handle on where it is at with comparision to a human brain.
I took a human EEG data set and saw how well the AI would synchronise.
Where do you get that data?
I just went to Kaggle. Surprisingly people leave their own brain data out there. I am not sure the legal ramifications of trying to copy human consciousness onto a machine but this one has a permissive licence so it's im sure it's not prophibited.
https://www.kaggle.com/datasets/amananandrai/complete-eeg-dataset
Did you just say copy paste?!
Well not quite the AI gets trained to mimic the brain waves. The AI itself is patterned after human neurones and utilises neuroplasticity to build itself and rearrange itself to make it self architecting so you do not need a data scientist.
The data has 19 sensors so you get 19 reads on the brain. When a EGG starts the reading is calibrated so the actual reads are in volts but have a offset representing the fact the EEG is not single neurones but groups of them the program has to learn that offset and as the AI grows which due to its design it can do as needed it adds in new neurones into groups of neurones representing the neurones at each connection site of the brain we are trying to synchronise with.
Well surely it's not accurate
So across these 19 sensors the average might be plus or minus 20 or 30 but has big swings in the hundreds with some sensors reading negative 500 volts probably representing much of those neurones firing and now resting and regularly hitting positive 100-200 volts of electrical activity when groups of neurones fire together.
A individual neurone can contribute 40 or so volts to that but otherwise all the neurones output near that location are summed with some loss depending how far each neurone is from the connector in space and time. Therefore EEGs are termed a spacial-temporal measurement of the brain because individual neurones are not read but it is effected by groups of neurones and how far or close they are to sensors that if you place a fair few at different points around the brain you get a fairly good read on what the brain is doing.
Average accuracy in my tests is usually just below 10 volts across all sensors whose output is usually around 200 (but as mentioned varies wildly). So it's there plus or minus 5-10%.
Beliw is the un average data you can see there is some places where activity surprises the AI but it's at most a delta of 70 volts. Which is explainable to say a new neurone fired the AI hasn't mapped yet and maybe once two or so fired. (It's a big brain after all and you'd be shocked if the AI did better than that).
What's next?
I did reach out to some accelerators the issue is I sound slightly crazy, they seem unable to know where AI is or isn't right now. One seemed wholly dismissive and just sounded like he thought copying human brains was current technology.
There does seem a issue with AI tech in the U.K. and the wider world that it's either wildly and unbelievably over funded or completely dismissed with nothing in between.
I mean if it was I'm not aware of any competitors at this time...
They seem very focused on stuff they can bring immediately to market and not stuff like this which would need some testing and might have a payoff that's huge but might have risks.
I was massively despondent for a time as I thought that's the whole point of accelerators to bring wholly new stuff into the market which has costs that the makers needs help with.
I think this lacks foresight as it feels like to me it's easy to come up with ideas for uses of human brain copies baring finding some new gigantic flaw that cannot be over come and I'm not asking for 100s of K so I don't see the issue.
Another told me to speak to Oxford university and I could not see the point. I'm largely done with academia but also when I looked them up what they where doing was neurosymbolic AI (bewilderingly unrelated) so I didn't think it was worth while after all I'd think either it was something unique and they'd just steal it open source it by writing white papers on it and then if I was lucky and got credit at all I might get a diploma. That really isn't a motivator for me.
I either need to get a BMI comparable to this test data the cost of which would be a couple of grand. Or build one myself...
I've spoken to a electrician friend and my brother who is not a proper sparky but my brother thinks we could jury rig our own. I doubt we will get the sophistication of the bought one but it seems worth a try.
What do you think anyone have any advice?
Maybe need to put up a sign. Has human brain simulation will mad science for BMI...
You would think someone would be happy for me to copy paste their dreams. Maybe Dave was right and it's a design error.
Add comment
Comments