List of arguments that AI poses an existential risk
This page is under active work and may currently be incoherent or inaccurate.
This is a list of arguments that future progress in artificial intelligence may bring about the extinction of humankind or drastically limit human influence over the long-run future.
Arguments
This list does not contain evidence from opinion.
Risk from competent malign agents
(Main article: Argument for AI X-risk from competent malign agents)
Summary:
Some advanced AI systems will very likely be built to pursue 'goals'
The aggregate goals of these systems may tend to be bad.
Such systems will likely have the power to achieve their goals even against the will of humans
Thus, there is some chance that the future will proceed in opposition to long-run human welfare, because these advanced AI systems will succeed in their (bad) goals
Key counter-arguments:
Second species argument
(Main article: Second species argument for AI x-risk)
Summary:
Humans dominance over other animal species in controlling the world (including sending them extinct) is primarily due to our superior cognitive abilities.
Therefore if another 'species' appears with cognitive abilities superior to humans, humans will lose control over the future, and the future will lose most of its value.
AI will replace humans as 'species' with the most superior cognitive abilities.
Key counter-arguments:
Intelligence in animals doesn't appear to generally relate to dominance (insects are found everywhere but are not very intelligent.
The notion of 'cognitive abilities' is very broad. It is not clear which need to be dominant for a species to be dominant. A calculator is cognitively superior to all humans in terms of doing sums but they aren't dominant.
Endorsed by:
Loss of control via inferiority
Summary:
AI systems will ultimately be much more competent than humans
Thus most decisions will probably be allocated to AI systems, because they will make more competent decisions
Also, AI systems will often be able to choose to make decisions not intentionally allocated to them, because their superior decision-making will allow them to manipulate situations markedly better than humans can.
If AI systems make most decisions, humans will lose control the future
If humans don't control the future, there is a high chance the future will be bad according to human values
Versions:
Version of this argument may work with different forms of AI superiority:
quality of thought
speed
number
copyability
co-ordination
transparency
non-susceptability to permanent death
other AI advantages
Key counter-arguments:
Loss of control via speed
Summary:
Advancing AI may tend to produce very rapid changes
Faster change reduces the ability of groups of humans to maintain safety e.g. reviewing and understanding it, responding to problems as they appear, adjusting course, preparing, negotiating.
The pace of events could become so fast as to allow for negligible human safety efforts
Human efforts to maintain safety may be important for avoiding arbitrarily bad situations
Versions:
This argument may work with several versions of speed:
AI systems will likely act much faster than the human activity they will replace: this is a form of 'Argument from loss of control via superiority'
New AI systems will be developed much faster than similarly impactful technologies previously have
AI systems will produce new non-AI technologies (e.g. weapons) much faster than similarly impactful non-AI technologies previously
Technological changes will lead to changes in society much faster than previously
Key counter-arguments:
The burden of proof could be high for an implausible event such as the destruction of humanity (as opposed to smaller scales of catastrophe)
This argument also seems to support concern about a wide range of technologies, It is unclear if this is predictive of which are worth worrying about
Vulnerable world triggered by AI
Summary:
Technologies may be possible that grant extreme destructive capabilities to small groups without granting other people defensive capabilities
Some people and collectives would like to destroy humanity, or would risk that for other aims
Advanced AI may be such a technology, or may produce such technologies
Key counter-arguments:
AI empowers lone actors on catastrophic projects
Summary:
Until now, large projects have required the labor of many people
This disadvantages projects most people would not work on, and more strongly disadvantages projects others would work to end if they knew of them, e.g. highly destructive or already illegal projects.
Projects to end humanity are in both of these categories, thus have been disadvantaged until now.
AI will allow large projects to proceed with minimal human labor, and thus with the cooperation of very few people
Thus AI will make projects to destroy the world easier, raising the chance of one succeeding.
Key counterarguments:
Normal people's utopias are catastrophic to one another
Summary:
People who broadly agree on good outcomes within the current world may, given much more power, want outcomes that one another would consider catastrophic. e.g. A utilitarian and a Christian might both work to reduce poverty now, but with much more control, the utilitarian might replace humans with efficient pleasure-producing systems without knowledge of the real world, and the Christian may dedicate most resources to glorifying God, and both may consider the other future a radical loss.
AI may empower some humans or human groups to bring about futures closer to what they desire
From 1, that may be catastrophic according to the values of most other humans
Powerful technologies raise the chance of catastrophic accidents
(Main article: Argument for AI x-risk from potential for accidents and misuse)
Summary:
Advanced AI could yield powerful destructive capabilities such as new weapons, new computer viruses, new routes to interfering with other actors.
Prevalence of very powerful technologies raises the risk of cataclysmic accidents
AI may produce or accelerate destructive multi-agent dynamics
(Main article: Argument for AI x-risk from destructive competition)
Summary:
Competition can produce outcomes undesirable to all parties, through selection pressure for the success of any behavior that survives well.
AI may increase the intensity of relevant competitions.
Large impacts suggest large risks
Powerful systems we don't understand suggest large risks
Summary:
So far, humans have developed technology largely through understanding relevant mechanisms
AI systems developed in 2024 are created via repeatedly modifying random systems in the direction of desired behaviors, rather than manually built, so the mechanisms the systems themselves ultimately use are not understood by human developers
Systems whose mechanisms are not understood are more likely to produce undesired consequences than well-understood systems.
Key counterarguments:
See also
Notes