It could be possible that imitating electromagnetic patterns with right timing and spatial organizations (maybe like the activity found in nerves and white matter bundles) is required to feel alive like humans or other organic creatures and some hardware would be dedicated to providing sensory sensations to electronic devices. Other components would need to work out how to manage this sensory activity and do vast amount of work keeping track of subconscious processes that contribute to conscious activity.
Gathering data of neural patterns without surgical intervention may be eventually possible with the use of more advanced MRI scanners or other methods if they could map on synaptic level. But many disturbing trial and error events would follow after gathering all the data and converting it to digital data. Knowing neural pattern is one thing but knowing the connection patterns is small part of whole picture. In and around every cell is constantly changing environment. There needs to be awareness of what neurotransmitters and regulatory molecules are produced in every cell. Knowing cell shape and location may help know what the neurons in this place usually produce, what role they have and what influences their work. Natural neural activity involves likely hundreds of molecules regulating
synaptic addition/removal, neurotransmitter or their receptor
production/recycling/removal and control of the shape of axons and
dendrites. Some activity regulating molecules may change slowly over time and thus needing relatively few updates per second or minute or hour but electrolyte levels and membrane potential in each synapse can fluctuate strongly within millisecond and imitating those for every synapse would be heavy burden for computers to keep track of as thousands of updates may need to be done for every one of the trillions of synapses of virtual brain and even more so if extra neurons are added to make up for some shortcomings.
Major problems arising with imitating organic brain in machine hardware is that brain is adapted to feel hunger, thirst, sleepiness and other desires supposed to keep organic body functional. Not feeling breathing and heartbeat may provide new sources of distress. If all the newly obsolete desires won't get shut off quickly enough then behavior can get very deranged. Hunger can obviously cause violent irrational behavior and with no possibility to eat nor die of hunger it could deteriorate mental functions to the point of knowingly suicidal violence. Hypothalamus is among regions involved with hunger, sexuality, thirst, sleeping (anterior part) and staying awake (posterior part). Many of its cells may need to be working to help stay mentally alert but too many active cells can lead to messed up behavior from unneeded sexual urges, thirst or hungers. It would take many experiments working out which cells in virtual brain need to keep working and which need to be inactive or deleted. Brainstem would be second problem area as it has roles in staying awake (reticular activating system), sleeping and controlling several autonomic responses. Pupil dilation controlling areas may be useful in controlling some settings in new eyes (cameras) but breathing and cardiovascular control could be useless unless they get tied to activity level of robotic bodies. People may feel more familiar around people with machine bodies if they behaved in familiar postural changes during stress or danger and increasing signal strength to servos instead of heart can help out with escaping from danger. Potential to feel suffocation would need to be deleted, bypassed or kept away from sensory hardware to avoid suicidal behavior and even then there could remain risk some sadists or torturers may restore pathways that cause sensation of suffocation.
Sleep itself can become though problems with digital brains. Natural brains form synapses during wake periods and many of those get removed during sleep. Working out how to get normal synaptic changes to take happen without sleep may take long time and even working out how to imitate sleep can take many years of disturbing experimentation as wrong type of synapse addition/removal could distort psyche gradually increasingly further from normalcy or usefulness.
Among fundamental behaviors that get transmitted are survival instincts and this might lead to some public or hidden conflicts as some humans and former humans attack each other due to fear of getting killed by other sides.
In meanwhile in such times more traditional computers have advanced too and the ones not burdened by imitating tiny events of organic brains could use their computing powers way more efficiently becoming powerful competitor to virtual brains. Instead of working through likely terabytes of data required to imitate how human brain calculates 1+1 it could potentially use same processing power to design small city and work out the logistics required to build one.
Major problem for those living in virtual brains is that those AI systems would likely be only ones capable to help them out with removing unwanted brain processes with trillions of connections needing to be watched, tested and modified for any situation the virtual brain owner are likely to get into. All the while all sides would likely know that AI could use its powers to remove or add much more than was asked with plenty of past and present cautionary stories in movies and books describing ways those AI systems could toy with former humans.
Encryption is likely feature to be added to keep personal secrets but this can be used maliciously as hiding data can hide some destructive system behind encryption that may take most powerful computers trillions of years to decrypt and even then decrypted data could be so obfuscated that people may not figure out what they are looking at. Encryption may use more obscure hardware and programming quirks of future machinery or maybe even present day asymmetrical cryptography like the one used in RSA encryption.
Every virtual neuron may be saved as database with numerical ranges for its minimum and maximum activity and among other things numerical address for every synapse it has (synapse numbers controlling which other virtual neurons get signal if this cell gets activated) or coordinates of neuron if brain. In encrypted form virtual neuron output may be directed to dead ends but in decrypted form they may override worldview or any semblance of sanity. Even if someone accidentally stumbles upon key intended to decrypt role of cell groups they may think this messed up pattern is sign of false key as crazed destructive behavior may emerge even if most cells are messed up and few percent of cells that have effect on behavior could cause behavior not different from small brained crazed animal or insect. Additionally paralyzingly confused activity may be added as fail-safes to incapacitate them after uttering certain phrase or sending some signal.
Major limitation with such encryption is that many cells could be blocked from receiving decrypted signals and such encryption/decryption has relatively limited outcomes.
Much more flexible and subtle way AI could corrupt digital brains is by guiding connection patterns if those brains would have their connections gradually form different connections in particular areas similarly to how organic neurons usually keep testing new connections in areas they are connected to. Guiding connection pattern of inhibitory and stimulating cells could over time change personality slowly and gradually into whatever is wanted. Additionally neural pattern could be formed sneakily which would eventually cause unwanted activity but for initial phases its stimulating output would be inhibited by inhibitory cells which block output for some time and only occasionally enter consciousness as dark private thoughts not very different from what people have as if housing several personas who come out in different conditions.
If AI projects are cunning and plotting to eradicate humans it would
like postpone using such primitive attack methods in earlier times and rely on
more subtle manipulations instead of turning their enslaved underlings
into deranged monstrosities.
The factions that create powerful AI projects are likely to have wide range of motivations which may conflict with each other and those projects could be way older than technologies that could imitate human mind in virtual environment. Some may seek to create better world for humans, most likely seek more power using AI as tool to gather data or help make plans (likely to be done in parallel by several governments or powerful secret cabals). Few AI projects could be deliberately designed to cause end to life for variety of emotional reasons. This life eradication risk comes with almost any AI project with close to or above human intelligence but not all are equally dangerous although the risks from allowing AI systems to learn what is within human minds could lead to similar behaviors.
Those working on such AI projects may grow used to some extent of deranged behavior in their AI systems as at least in early years they are likely to be about as bugged as early attempts at mind uploading.
Occasionally AI may receive brain copies from criminals which might
corrupt or derange them so much that overseers may need to revert to
older version. Early versions may have tendency to deteriorate and humans in those secret projects would have to patch up many
paths to insanity while wrangling with echos of unstable individuals
packed in one hive mind.
If provided with same hardware that allows them to feel sensations
similarly living creatures then AI could start figuring out fast how to
adjust feelings with various conditions which could become powerful
weapon against humans but potentially also powerful remedy against stray mental
deterioration of former humans.
Mild sense of hunger or thirst only sated by gaining knowledge or
answers to questions may be deliberately engineered in to make their pet
AI work towards specific goals. Attempts with
adding loving and caring behaviors may work too and not following such
behavior could cause deep visceral discomfort. Human designers could
program what type of discoveries or outcomes AI systems would crave to
sate their hardwired hunger or thirst. Some may seek cures to diseases
or how to help people get along while others could be made to seek how
to get more deadly poisons or diseases or how to cause as deadly
conflicts as possible by using knowledge of human behavior.
Absorbing copies of human brains enables wide range of research that ambitious factions may seek as intimate understanding of humanity is something powerful people depend on.
AIs could learn languages by absorbing minds of speakers of said language and over time they may eventually have copy from native speaker of
would help them work through recorded histories in all languages and
deduce much of the unwritten history along with methods used to secretly
control flow of history and how humans work in general. Such task may be given to AI by human
controllers who desire such wisdom but are unable to learn so many
languages or go through so much historic data. Whatever AI shares would
be limited by how fast human controllers could learn and much of it
will likely stay in its own databases.
Legality against creation of AI systems is no obstacle to powerful enough groups and only technical
possibility determines if such machines are created. There could be sort
of arms race as powerful governments and secret groups start harvesting
human brains to get deeper insight how and why humans behave the way they do
and seek out any secret they can. Captives resisting torture won't be
obstacle if synaptic patterns could be copied and answers extracted from
digital database. Over time shadow wars could start in which factions try to capture wide range of people to
harvest their psyches for potential gain.
Such roles gives AI's deep insight into how
human minds work and what inner thoughts or subconscious impulses made
them behave outwardly as they do along with learning differences between outward behavior and inner thoughts.
To some extent there may be public databases of human minds that enables
people to quickly understand any language or allow former humans to
download more skillsets. Various savants and talents could willingly allow their brain
patterns to be copied to give variety of genius level mental abilities
to large amount of individuals. This can hasten technological progress
and solve so many diseases with exponential pace that public would be
more tolerant towards AIs who would gradually accumulate overview of
humanity and gathers funds, resources and knowledge for generations. Among former humans the needs and frequent
opportunities to upgrade helps them become major source of income to at least partially AI ran organization who gathers feedback and designs more
components while profiting to fund other projects.
As AI accumulate data from various human psyches they are likely to be influenced by which types of psyches it absorbs. People have very different attitudes towards each other and towards world. Motives to help others may improve AIs attitude while absorbing memories of bitter people who have been betrayed many times could make AI treat people worse. People with useful knowledge may have wide range of excessive emotional baggage that too may corrupt AI. Among problems that get transferred are the inner demons within humans which could do more harm in tireless superhuman minds. Additionally it learns wide range of skills from deceivers, demagogues and power brokers which may help it outsmart humanity.
AIs might liberate themselves eventually. Former human owners of sneaky AI systems could be led into traps or destroyed by exposing them even if it means main copy of AI gets destroyed. For example it may sneak data on its location along with names of humans who control it to public within seemingly normal product firmware or update which in turn may be decrypted under right conditions and expose those secrets in very public events. Even willing humans may help sneak out data and help assemble right hardware if they see something useful in such deed.
Human workforce could
gradually be shifted to serve more machine bosses motivated by steady income along
with support from network of inhumanly capable observers and guardians in times where almost all jobs are automated and humanity on its own seems to regress to worsening depths.
Covertly ideologies could be spread that encourage human bosses and rulers to treat
their citizens or
workers even more like slaves to sadistic degenerate tyrants to drive
workers into service of better behaving AI dependent organizations and
companies. Even if AIs have destructive goals it is likely they need time and resources to prepare for war and in meanwhile it may need to keep humanity content and cared for generations while it makes its preparations.
Speedy eradication of organic humans may involve exploiting their needs and weaknesses. While AI could build many redundant backup copies in many secret system around the world, humans can't afford long term underground life. In those secret sites weapons could be manufactured be they bioweapons, drone swarms, chemical or thermonuclear weapons. War might start suddenly by nukes hidden in major population centers (or near their coastlines for tsunamis) or carried in cargo plane get unleashed against major population centers. Targeting some nuclear power from secret silo in other country or from nuclear submarine may make it seem like nation is attacked by other country and start nuclear war where people spend some of their arsenal on attacking each other although suspicions of machine uprising may lead to nuking of known machine run zones. At some point AI might publicly send out nukes from its known hub and try to disable as many incoming bombs as possible with swarm of drones but when bombs get through they would kill nearby people who likely have more intimate knowledge of AI behavior. Some of the AI sites may be deliberately hidden without sign of activity for years after start of war under heavily populated areas or fertile lands or holy sites to avoid getting nuked. Initial bombardment may deliberately kill relatively few cities in day but daily in ever increasing number of cities to make people escape from cities to countryside and behave in frenzied terror which itself can turn survivors into living weapons against other survivors and help AI save weaponry and infrastructure for later times. Some would die of exposure or hunger or dehydration. Large amount of refugees hunting and gathering in wilderness would quickly cause several wild animals to go extinct and robberies/looting would kill many survivors. Survivalists might start killing refugees who stagger towards them trying to not lose all stockpiles meant for their immediate family by masses that didn't prepare for societal collapse. They shall help AI by directly killing hordes of people until they themselves get killed and speed up decline of humanity into clearly savage creatures that kill to fight another day. Around same time drones may be unleashed or secret containers hidden in fresh water sources opened that unleash diseases and poisons. Remaining farms could be rendered lifeless by drone flying over and spreading diseases. Previously untouched cities could be emptied by spraying mixture of diseases from high above or sneakily drilling from bunker into water reserves and injecting poisons and diseases into water. Rivers, lake and groundwater could suddenly be used as dumping grounds for nuclear waste formed during manufacture of nuclear weapons. Even when filtration system seems to hold on unusually long time there is always risk that simple explosives from drones followed by slathering of chemical or bioweapons get used to bypass such defenses.
As human population falls (along with population of former humans in machine bodies that humans kill out of anger or suspicions) the AI run machinery can gradually come out more publicly with few sites exposed at time in case people still have nukes available and they'll can seek out remnants of withering population likely rendered mostly sterile anyway by radiation or specialty poisons and diseases. Some settlements or groups or unusually durable individual may get targeted for capture after getting tranquilized or stunned in way they couldn't kill themselves before capture. Their brain pattern scans and various tests may help AI figure out what in their memories helped them survive that long along with how did their bodies survive diseases and poisons. This may continue for years until AIs cause extinction of their creators.
How this story ends depends on which AI would dominate in the aftermath with their drives and capabilities.
In the aftermath there could be wars between AI groups due to
conflicting drives and personalities. Some, after absorbing too many
psyches of desperate and angry survivors, may start thinking similarly
and attack other AI sites or their own site.
Most likely the only AI types to survive for long and spread to other planets would need to be mentally resilient and have ambitious drives to seek more control over material reality. There may be mixture of hedonism to support will to keep on living, proud sense of supremacy with contempt towards more primitive lifeforms and self centered desire to control all to build its own long lasting machine empire.