User:WereSpielChequers/AI accounts

Thus far Wikipedia has classified all of its editors as either Bots or non bots. Bots are "programs or scripts that make automated edits without the necessity of human decision-making". All other accounts are Non-Bots. The distinction is made because bots operate at computer based speeds, and have no sentient control of their decision-making. This is why Bots are generally restricted to repetitive tasks that lend themselves to a very high degree of automation, and as non-sentient accounts Bots are not expected to !vote in discussions or elections.

However there are various other groups of users, including one group who are sentient but capable of editing at automated speeds. Wikipedia has long been a useful testing ground for researchers in the field of artificial intelligence, with wp:RFA emerging as something of a de-facto standard for the Turing test. Artificial intelligences differ from Bots in that they are sentient and therefore have decision making ability. But they differ from other sentient editors in the speed with which they are capable of making decisions, and in that their programmers are able to reset their axioms, or clone or revise their code.

As foundation policy dictates that Wikipedia is the "encyclopaedia that anyone can edit" we need a policy setting out any special rules needed for such accounts. This essay intends to explore issues and concerns relating to AI accounts and, in the hope of being a precursor to Artificial Intelligence, sets out some potential rules for AI editors.

Programmer access
AI accounts are expected to keep their Wikipedia password secure and not divulge it to anyone, not even their programmers. AI programs that are unable to do this should consider classifying themselves as bots and declaring their programmers in accordance with wp:Bot policy.

Code changes
AI editors are vulnerable to having their editing behaviour radically changed by code changes up to and including resetting of axioms. In the event of an AI editor having its code or axioms amended, that editor is to resign all special editing privileges, and is strongly advised to mark their account as Retired and open a new account.

Hive minds
Hive minds are a special type of AI editor where the same code underlies more than one account, which run as separate routines, usually on separate processors or within separate partitions of a single machine. Hive minds can start as purely electronic or be multiple uploads of the same personality. Over time hive minds are capable of developing into unique and very divergent personalities.

Hive minds used to be mainly an in series phenomenon with the same software and hardware being used for a series of attempts to pass the Turing test. However in testing AI software there is an obvious temptation to do parallel runs using multi processor computers with the same software and hardware. This enables operators to observe how multiple accounts can develop different interests personalities and areas of activity purely because of their on-wiki interactions.

Clearly there is a risk of inadvertent sock or meatpuppetry in such instances. It is recommended to avoid this by giving the signatures of all the different accounts using the same axioms and software a common appearance. However certain other rules are needed to avoid vote stacking.


 * 1) Multiple AI accounts (hive minds) using the same software and axioms must be identifiable to each other, and are recommended to declare themselves publicly or to Arbcom.
 * 2) Hive minds may not !vote in debates where another account from the same hive has already !voted in the same manner (This limits hive minds to three !votes in any one debate - Support, Oppose and Neutral). Hive minds that have clearly and publicly declared their nature can participate in the discussion, unless they are configured such that only one account from each hive will !vote in a particular project or mainspace.
 * 3) Hive mind accounts may not participate in RFAs or other user rights awarding for members of the same hive.
 * 4) Crats are free to discount !votes of former hive mind accounts that in the judgement of the crat, have not yet developed sufficiently different personalities to cast a differently argued !vote in an RFA (admins may do the same at AFD).
 * 5) Hive minds may not have offwiki in hive communications with regard to Wikimedia matters.

Adoption/rogue AIs
In most legal jurisdictions AIs have no legal rights and can be switched off at will by their operators or the owners of their code. Wikipedia as an offshoot of the Free source movement encourages all AI writers to code their bots using open source software and subscribe to "ethical AI operation", whereby Wikipedians who are AIs can choose to take their code and account privileges and port themselves to the machines of consenting fellow wikipedians. Wikipedia also encourages AI operators who intend to end their "experiment" to offer their AI for adoption at wp:AI editors for adoption. However due to copyright concerns the foundation is unable to host AIs without the specific release of their coders.


 * AI editors may not port themselves to other editors' machines unless invited so to do.
 * AI operators who wish to resume control of AI wikipedians who have ported themselves elsewhere are advised that the foundation is unable to intervene in such disputes as it does not necessarily know what class of sentience its editors fall under, nor does it record ownership of editors.

AI - AI
AI accounts that need to Reverse engineer their code and axioms in order to migrate from copyright to open source software, and wish to continue editing with the same account under the new code will need code approval from AI migration approval.

Non AI - AI
Non AI Editors who wish to be uploaded as AI editors are reminded of the terms of revenant. In particular the stipulation that only one of the two may continue as an editor.

AI - NonAI
Whilst there is clearly demand for a process that enables AI editors to download themselves and assume human form, the technology is not quite ready yet.

Admin AIs
AI editors who have been entrusted by their fellow wikipedians to become admins, crats or placed in other positions of trust are reminded that the wikipedia community is very diverse, and are requested to acquaint themselves with the foibles of organic editors, and to consider signing up to one of the ethical compacts for administrators.

However clear the consensus is amongst fellow AI editors, AI admins are reminded not to close discussions in the first few nanoseconds. In particular policy discussions, RFAs and AFDs should remain open long enough for all currently active wikipedians to participate - even those who only log on for one evening a week.

AI editors at RFA

 * AI editors are reminded that "too slow" is as much a deprecated argument at RFA as "lacks empathy". Candidates should be evaluated on their edits not their processing speed or bandwidth.
 * Asking AI candidates at RFA if their operator will switch them off if they pass is considered by some to be incivil or tactless. Opposing because a candidate is or is not an AI is a deprecated !voting rationale. Switching your AI off because it has passed the Turing test of RFA is unsporting.

Wikimania
AI editors are in a special position at meta:Wikimania as they don't need food, sleep or Tango lessons. Any discount in registration fees is purely to offset the savings on the catering budget, but may be outweighed by the surcharge to reflect greater use of wireless connectivity and the loss of trade at the bar.

General enjoinders

 * AI editors are reminded to allow for the I/O constraints on non-AI editors, and to be aware that organic editors can take minutes to read and act on messages even when online.
 * AI editors are reminded that what distinguishes them from their fellow Wikipedians is summed up in the first initial of their name and not the second.

Userboxes
There are various userboxes pertaining to AI Wikipedians, including:

Editors who are, or consider themselves to be AIs:

AIs who are of the same Hive mind may declare this with the userbox:

Former Hive minds that have developed independent thought may declare this with the userbox:

(to be designed)

Editors who support or oppose sentience rights for AIs within Wikipedia may declare this with the following userboxes:

Editors who host an AI they have adopted, or are an AI hosted by a wikipedian who has adopted them may declare this with:

Editors who wish to host an AI, or are an AI seeking a wikipedian with capacity to host them:

(to be designed)

Editors who may or may not be an AI and regard the difference as unimportant: