Friendly artificial intelligence

friendly artificial intelligence (also AI or FAI ) is a hypothetical artificial general intelligence (AGI) that would have a positive effect on humanity. It is a part of the ethics of artificial intelligence and is closely related to machine ethics . While machine science is concerned with an artificially intelligent agent should behave, it is adequately constrained.

Etymology and usage

The term was coined by Eliezer Yudkowsky [1] to discuss superintelligent artificial agents that reliably implement human values. Stuart J. Russell and Peter Norvig’s leading artificial intelligence textbook, Artificial Intelligence: A Modern Approach , describes the idea: [2]

Yudkowsky (2008) goes into more detail about how to design a Friendly AI . He asserts that friendliness should be designed, but that the designers should recognize that their own designs may be flawed, and that the robot will learn and evolve over time. Thus the challenge is one of mechanism design-to define a mechanism for evolving the system of checks and balances, and to give the systems utility functions that will remain friendly in the face of such changes.

‘Friendly’ is used in this context as technical terminology , and it is recommended that they are safe and useful, which is “friendly” in the colloquial sense. The concept is primarily invoked in the context of recursively self-improving artificial agents that rapidly explode in intelligence , on the grounds that this hypothetical technology would have a large, rapid, and difficult-to-control impact on human society. [3]

Risks of unfriendly AI

Main article: Existential risk from artificial general intelligence

The roots of concern about artificial intelligence are very old. Kevin LaGrandeur showed that the dangers specific to AI can be seen in ancient literature concerning artificial humanoid servants such as the golem , or the proto-robots of Gerbert of Aurillac and Roger Bacon . In these stories, the extreme intelligence and power of these humanoid creations clash with their status as slaves (which by nature are seen as subhuman), and cause disastrous conflict. [4] By 1942 these themes prompted Isaac Asimov to create the ” Three Laws of Robotics””- principles hard-wired into all the robots in his fiction, intended to prevent them from turning on their creators, or allow them to come to harm. [5]

In modern times as the prospect of superintelligent AI looms nearer, philosopher Nick Bostrom has said that they are superintelligent. He put it this way:

Basically we should assume that a ‘superintelligence’ would be able to achieve whatever goals it has. Therefore, it is extremely important that the goals we endow it, and its entire motivation system, is ‘human friendly.’

Ryszard Michalski, a pioneer of machine learning , taught his Ph.D. students that a truly minded alien mind, was a machine mind, was unknowable and also dangerous to humans. quote needed ]

More recently, Eliezer Yudkowsky has called for the creation of “friendly AI” to mitigate existing risk from advanced artificial intelligence . He explains: “It does not matter, it does not work, but you’re out of it, it’s used for something else.” [6]

Steve Omohundro Says That a Sufficiently advanced AI system will, UNLESS Explicitly counteracted, exhibit a number of basic “drives” , Such As resource acquisition Because of the intrinsic kind of goal-driven systems and That thesis drives will, “without special precautions” causes the AI ​​to exhibit undesired behavior. [7] [8]

Alexander Wissner-Gross says that it may be considered that their future horizons may be considered to be more than likely to be considered, and unfriendly if their planning horizon is shorter than that threshold. [9] [10]

Luke Muehlhauser, writing for the Machine Intelligence Research Institute , recommends that machine ethics researchers adopt what Bruce Schneier has called the “security mindset”: Rather than thinking about how a system works, imagine how it could fail. For Instance, he has even suggested that unicast predictions and communicates via unintended harm. [11]

Coherent extrapolated volition

Yudkowsky advances the Coherent Extrapolated Volition (CEV) model. According to him, coherent extrapolated volition is people’s choices and the actions of the people would collectively take if “we knew more, thought faster, were more the people we wished we were, and had grown up closer together.” [12]

Rather than a Friendly AI being white designed Directly by human programmers, it is to be designed by a “seed AI” programmed to first study human kind And Then Produce the AI qui humanoid Would want, Given Sufficient time and insight, to at sometimes has Satisfactory answer. [12] The appeal to an objective though quota human kind (Perhaps Expressed, for mathematical Purposes, in the form of a utility function or other decision-theoretic formalism), as supplier providing the ultimate criterion of “Friendliness”, is an answer to the meta-ethical problem of defining an objective morality; Extrapolated volition is intended to be objectively meant, all things considered, but it can only be defined in relation to the psychological and cognitive qualities of present-day, unextrapolated humanity.

Other approaches

Ben Goertzel , an artificial general intelligence researcher, believes that AI can not be created with current human knowledge. Goertzel Suggests humans May INSTEAD decided to create an “AI Nanny” with “mildly superhuman intelligence and monitoring powers”, to protect the human race from existential Risks like nanotechnology and to delay the development of other (unfriendly) artificial intelligences up to and UNLESS the safety issues are solved. [13]

Steve Omohundro has proposed a “scaffolding” approach to AI safety, in which one provably safe. [14]

Stefan Pernar Argues along the lines of Meno’s paradox to Point Out That Attempting to solve the problem is ISP Either pointless gold hopeless DEPENDING whether we assumed at one universe exhibits That moral realism or not. In the former case, a transhuman would have been conceived of in the proper way and would not be reasoned about. [15]

Public policy

James Barrat , author of Our Final Invention , suggests that “a public-private partnership has to be created to bring together international-Atomic Energy Agency, but in partnership with corporations.” He AI Urges Researchers to convene a meeting similar to the Asilomar Conference on Recombinant DNA , qui Discussed Risks of biotechnology. [14]

John McGinnis encourages governments to accelerate AI research. Because the goalposts of friendly AI are not necessarily clear, it suggests a model more like the National Institutes of Health , where “Peer review panels of computer and cognitive scientists would that such advances would be accompanied by appropriate safeguards. ” McGinnis feels that peer review is better than regulation to address technical issues that are not possible to capture through bureaucratic mandates “. McGinnis notes that his position stands in contrast to that of the Machine Intelligence Research Institute , which generally aims at avoiding government involvement in AI. [16]

According to Gary Marcus , the annual amount of money being spent on developing machine morality is tiny. [17]


See also: Technological singularity § Criticisms

Some critics believe that both human-level and AI are unlikely. Writing in The Guardian , Alan Winfeld compares human-level artificial intelligence with faster-than-lightning difficulties, and states that while we need to be “cautious and prepared” given the stakes involved, we do not need to be obsessing “about the risks of superintelligence. [18]

Some philosophers claim that any truly “rational” agent, whether artificial or human, will naturally be benevolent; in this view, deliberate safeguards designed to produce a friendly. [19] Other critics question whether it is possible for an artificial intelligence to be friendly. Adam Keiper and Ari N. Schulman, editors of the technology journal The New AtlantisIt will be impossible to guarantee “friendly” behavior in the future. They write that the criteria for a possible likelihood of myriad possible outcomes, the goal of a consensus and a consensus on how to value the different outcomes. [20]

See also

  • AI takeover
  • Ethics of artificial intelligence
  • Existential risk from artificial general intelligence
  • Intelligence explosion
  • Machine ethics
  • Machine Intelligence Research Institute
  • OpenAI
  • Singularitarianism – a moral philosophy advocated by proponents of AI
  • Technological singularity
  • Three Laws of Robotics


  1. Jump up^ Tegmark, Max (2014). “Life, Our Universe and Everything”. Our Mathematical Universe: My Quest for the Ultimate Nature of Reality (First ed.). ISBN  9780307744258 . Eliezer Yudkowsky terms “Friendly AI,” …
  2. Jump up^ Russell, Stuart; Norvig, Peter (2010). Artificial Intelligence: A Modern Approach . Prentice Hall. ISBN  0-13-604259-7 .
  3. Jump up^ Wallach, Wendell; Allen, Colin (2009). Moral Machines: Teaching Robots Right from Wrong . Oxford University Press, Inc. ISBN  978-0-19-537404-9 .
  4. Jump up^ Kevin LaGrandeur . “The Persistent Peril of the Artificial Slave” . Science Fiction Studies . Retrieved 2013-05-06 .
  5. Jump up^ Isaac Asimov (1964). “Introduction”. The Rest of the Robots . Doubleday. ISBN  0-385-09041-2 .
  6. Jump up^ Eliezer Yudkowsky(2008) in Artificial Intelligence as a Positive and Negative Factor in Global Risk
  7. Jump up^ Omohundro, SM (2008, February). The basic AI drives. In AGI (171, pp. 483-492).
  8. Jump up^ Bostrom, Nick (2014). Superintelligence: Paths, Dangers, Strategies . Oxford: Oxford University Press. ISBN  9780199678112 . Chapter 7: The Superintelligent Will.
  9. Jump up^ ‘How Skynet Might Emerge From Simple Physics, io9, Published 2013-04-26.
  10. Jump up^ Wissner-Gross, AD ; Freer, CE (2013). “Causal Entropic Forces”(PDF) . Physical Review Letters . 110 (16): 168702. Bibcode :2013PhRvL.110p8702W . doi : 10.1103 / PhysRevLett.110.168702 .
  11. Jump up^ Muehlhauser, Luke (31 Jul 2013). “AI Risk and the Security Mindset” . Machine Intelligence Research Institute . Retrieved 15 July 2014 .
  12. ^ Jump up to:b “Coherent Extrapolated Volition” (PDF) . . Retrieved 2015-09-12 .
  13. Jump up^ Goertzel, Ben. “Should Humanity Build A Nanny Global AI to Delay the Singularity Until It’s Better Understood?”, Journal of Conscious Studies 19.1-2 (2012): 1-2.
  14. ^ Jump up to:b Hendry, Erica R. (January 21, 2014). “What Happens When Artificial Intelligence Turns On Us?” . . Retrieved 15 July 2014 .
  15. Jump up^ Pernar, Stefan. “The Evolutionary Perspective – a Transhuman Philosophy”, 8th Conference on General Artificial Intelligence in Berlin, July 22-25, 2015
  16. Jump up^ McGinnis, John O. (Summer 2010). “Accelerating AI” . Northwestern University Law Review . 104 (3): 1253-1270 . Retrieved 16 July 2014 .
  17. Jump up^ Marcus, Gary (24 November 2012). “Moral Machines” . The New Yorker . Retrieved 30 July 2014 .
  18. Jump up^ Winfield, Alan. “Artificial intelligence will not turn into a Frankenstein’s monster” . The Guardian . Retrieved September 17, 2014 .
  19. Jump up^ Kornai, András. “Bounding the impact of AGI”. Journal of Experimental & Theoretical Artificial Intelligence ahead-of-print (2014): 1-22. “… the essence of AGIs is their reasoning facilities, and it is the very logic of their being able to behave in a mor … The real nightmare scenario (is one where) humans find it advantageous to strongly couple themselves to AGIs, with no guarantees against self-deception. “
  20. Jump up^ Adam Keiper and Ari N. Schulman. “The Problem with ‘Friendly’ Artificial Intelligence” . The New Atlantis . Retrieved 2012-01-16 .

Leave a Comment

Your email address will not be published. Required fields are marked *