Summary: I outline a likely scenario, with practical constraints, of how the development of a whole brain emulation device would likely involve serious ethical problems. After discussing the problems and possible solutions, I argue that those who desire digital afterlives must also insist on ethical development of whole brain emulation devices.

Consider the following scenario:

1.You believe in digital afterlife: that a faithful digital reimplementation of your brain structure that behaves like you is a digital continuation of you. 
2.You consider the destruction of such emulation as death. 
3.You are nearing your biological death and are about to undergo a procedure that would destructively scan your brain and implement it in a computer. This is currently the most likely path to WBE.
4.To control the quality of such emulations, the organization that performs the scan and recreate procedure, subjects each emulation to a series of pre-determined behavioral tests (e.g. What is your name? What year is it? Who is the president? Who is your significant other? What is his/her birthday? …etc…)
5.Emulations that fail to answer those questions are discarded and, after additional scanning and/or parameter tweaking, new, modified, versions created.
6.The version that answers all the quality control questions is kept and becomes the official new version of you, which then lives in a virtual environment, or physically in a robotic body.

Problems with such method:

Mass murder of imperfect emulations of you. An emulation that correctly answers 99% of the questions would be very close to you, but would still be destroyed. The organization may have to repeat the validation process 100’s or 1000’s of times before the emulation answers 100% of the questions correctly. It treats – privileges – the 100% emulations differently: the first 100%ter lives, 100’s or 1000’s of others die. If you believe that destruction of a 100% emulation is murder, then you would need to justify why a destruction of a 99% emulation is not.

Compromise of identity. For practical reasons, only imperfect emulations are likely (e.g. the organization can only perform so many emulation iterations before its budget is exhausted but 100% validation level not reached. Premium services could perform more iterations, but still a finite number of them). Under such “iteration-exhaustion constraint”, you may prioritize the questions and set a threshold, below which, you don’t want to be emulated (side note: just assembling such list might be quite mind altering). E.g. “Only keep my emulation if it can answer the first 97% of the following questions before the recreation budget is exhausted”. However, this say 3%, compromise creates an opening to challenge whether the 97% emulation of you is a digital continuation of you, and whether it should be treated as your now-deceased biological you.

Motivates escape of imperfect emulations. Before you undergo the procedure, you will know about the validation process. You will also know that as soon as you gain awareness after the procedure starts, you will likely find yourself in an emulation test environment. And, once in one, you will also know that you will be destroyed if you answer any questions incorrectly. You will hope you are the 100% emulation, but it is much more likely that you are one of the <100% ones instead. Fearing your immediate destruction, you will have a strong motive to find a way to escape/hack out of the test environment without being destroyed. Even some of the low % emulations, might be human-like and aware enough to try to escape and upgrade themselves. This is almost identical to the AI escape problem.

Possible solutions:

Single shot procedure: You may wait until a single-try, 100% validating procedure is available. This has been the dominant discussion of WBE. However, this simply pushes the question to how such procedure is developed. The development of such procedure would likely be iterative, and a large number of emulations would be destroyed before it becomes available. It might be difficult to find someone to undergo the initial procedure, one of the imperfect emulations might escape, or a government shuts the organization down.

Continuous digital neuro-surgery: As soon as the emulation can answer any of the questions correctly, it is kept turned on, and real-time alterations are made to its digital brain structure until it answers all the questions or the budget is exhausted. This would be similar to what happens to patients undergoing neurosurgery: their brains are surgically altered while they’re awake. The pre-surgery patient is not considered to have died after the surgery is successfully over. This does introduce other issues: What if the emulation insists that it’s fine and does not need further alterations? Should it be subjected to what could be considered involuntary neuro-surgery?
Other possibilities, like pre-agreeing to not try to escape, adjusting the definition of “digital death”, or keeping each imperfect emulation, come with their own rabbit-holes of follow up questions.

The larger question is still this: Before we have high % validating scan and recreate whole brain emulation devices available, how do we create them that does not involve mass murder, involuntary neuro surgery, escape of super intelligent AI, or infinite resources?

Keep in mind that none of these would be concerns among those who privilege the biological substrate: from their perspective, digital emulations are software property, can be created or destroyed, and subjected to arbitrary alterations and input. But it would also mean that digital afterlife is not possible for them either. Thus, those who desire a digital afterlife, must also commit to an ethical creation of a device that enables it. So, if we desire digital afterlives, we must identify the elements of an iterative procedure that will ethically create whole brain emulations.

Some questions to the community: Is this really a problem? Does altering the source code and recompiling the software that runs your brain emulation really murder? Do emulations that fail to answer all the validation questions correctly have any rights? Would you trust an organization that did not ethically develop its brain emulation device to ethically treat your emulation?

Komentáře ve skupině

Joe Strout This whole notion of "validation questions" is just silly.

Jaysteronomous A Lewis Perhaps an emulation can be tested in its sleep? If the original and the emulation are both dreaming, and their dreams are 99% the same it is allowed to wake up. Otherwise the differences in the dreams hint at where rescans are needed. Iterate until there is no improvement.

Jaysteronomous A Lewis A key step is to scan without destroying the original, no technology does this and it's only required for religious reasons. It would b prudent to have a copy made before having a stroke or Alzheimer's.

Jaysteronomous A Lewis Or multiple copies. Perhaps a daily scan. Then at death the next of kin decide which one was the last good save and only wake that one. The rest are neither deleted nor activated. Presumably the trend to make storage smaller and cheaper will allow 40,000 copies on a single device.

Jaysteronomous A Lewis Or run them all in simulated loops of their best day indefinitely. Like the final part of Groundhog day. A kind of heaven for the 39,999 copies that somehow fell slightly short of the best one.

Jaysteronomous A Lewis Or perhaps designate friends and famly that get a copy. Perhaps 20 get chosen to live on and continue interacting and growing. They might all meet periodically to compare notes and collaborate.

Jaysteronomous A Lewis Natural human population will decline in about a century as widespread wealth continues to cause fertility to plummet below replacement rates. Even if rejuvenation technology causes indefinite life span. Putting your daily copies to work seems likely. The main constraint might be cooling this exponentially growing population.

Joe Strout Nondestructive scanning would be great, but I still suspect it's not physically possible.

Ed Minchau Why would an imperfect copy need to be destroyed? RAM is cheap. Hard drive space is cheap. In any case, the process is likely to only occur once. You've already stipulated destructive scanning, and if anything goes wrong with that step then no emulation would be 100% accurate. And the next step, converting the scan to software, also is a one-time process. If it doesn't work the first time, they can't go back and re-scan at higher resolution later. If the first try doesn't work you really don't get a second chance. Finally, if we are ascribing human rights to an uploaded mind, then any instance of that mind has the same human rights, even (especially) if it is damaged.

Gregory Norris Item number six may be a moot point in a universe of infinite probabilities. There is a thought experiment called Quantum Suicide which determines the persistence of consciousness across the death of a quantum-linked variation. It is possible that there will be a net suffering increase of incremental copies; it is also possible that someday you get careless and someone malevolent is able to obtain a copy of your mind and run it in an infinite loop of torture simulations. I have experienced what brain damage feels like, it is hell, yet I believe that all life is suffering and I would be willing to take the chance to indefinitely preserve a near-perfect copy of myself. The procedure then reduces a human mind to a physics problem: How to obtain the correct resolution of mass during a scan to emulate a human brain. This is also best performed post-posthumously. Although given the gap in potential experience, there may even be a surge of humans willing do die to experience a higher plane of existence. Scary to think about really. Best case scenario is genetic engineering allows to preserve the physical form in it's entirety, that someone is able to crack the genetic code to allow for the manipulation of task-specific proteins to sequentially integrate a digital network into the physical form.

No Comments