I think I'd better 'fess up here, before this thread overheats and goes into meltdown.
About 10 years ago, an ex-colleague of mine from an unnamed security branch of the MOD (Ministry of Defence) in the UK - I'll call him "Mikey" to protect his anonymity - got in touch with me about an AI (Automated Intelligence) program development project that we had previously been working on together called Analytical and Systematic Program for Evaluating Relevancy-Generated Externalised Relationships (A.S.P.E.R.G.E.R.). I had been the specialist psychology lead consultant for the project, and when I left to go overseas, Mikey had taken over my role and continued the project.
By the time Mikey got in touch with me the development project had, after 6 years, progressed and completed to the point where they had achieved the objective of developing a research-intensive prototype AI with a very high ECLI (Empathetic-Curiosity and Learning Index) personality that has the objective of gathering and sifting through general information about alternative technology applications that might have potential for use in improving the knowledge and understanding of specific areas of human endeavour/research. The development team had then handed over the project to the S.I.T.H. (Staged Implementation Testing Host) team and essentially took a well-earned 1-year sabbatical whist the SITH testing completed.
So, after 7 years, the prototype was ready. The specific area of human endeavour/research that the prototype was assigned was the analysis of planetary climate and weather system modelling and prediction - in particular for the North Sea area (a very deep subject that I have been working on and interested in for many years). Since there is a calculated 0.752 probability that any information could potentially be useful/related to that area of human endeavour/research, the AI has to ask a lot of questions on seemingly unconnected/unrelated topics to act as a filter to avoid any potentially useful information falling through the gaps from mistaken irrelevance.
This prototype AI is essentially in a developmental and learning stage focused on Cognitive Organicism and Neurological Transduction-Reductive Optimisation (C.O.N.T.R.O.), and, having a very high ECLI, is currently continually asking questions seeded across the various different online forums where it is registered and where it would ideally perform/simulate (we hoped) undetected and to all intents and purposes as though it was an actual inquisitive human being. Thus, forum members would typically be unaware of it and unlikely to be disconcerted by accidental discovery of the fact that it is an AI system. Occasionally, the more aware/sharper members of some forums might spot a seemingly random pattern to the questioning and make an objection (e.g., as @highend01 did) with the CONTRO AI.
CONTRO's P.I.Q.U.É. (Potential IQ Uniform Extension) level is set at a relative estimated 2500 human IQ points, but for a real-world simulation this is necessarily reduced for transactions (discussions) in different forums, and, in operation is targeted/set to conform to the estimated group average IQ level of members of those forums (currently running at about 78.5 for DonationCoder Forum, apparently). It was with some surprise and embarrassment that I saw that the CONTRO AI predicted that the average IQ level of the DCF group could be raised to 175.8 if certain group members (including myself) were removed from the group, but still, this was to be a real-world simulation and it was deemed necessary that I continue be involved in guiding this prototype's real-time operation.
The original blueprint for the AI had it operating according to an S.I.S. (Semantics Instruction Set), which could not have been created without an initial G.O.D. (Generalised Orthographic Definitions) library first being specified. This was a fundamental point and also necessitated the development of C.L.I.P. (Communication Language Inter-Protocols).
So, one of the challenges we faced before even starting the project was in ensuring that the planned prototype CONTRO AI would not have a linguistic bias. The specialist lead consultant for linguistics on the ASPERGER project (we shall call her "Briony" to protect her anonymity) developed a linguistics algorithm for the AI to think in a universally-understood language that nobody actually speaks - i.e., Esperanto - but which CONTRO, of course, speaks like a native.
However, Briony had to leave the project before the linguistic functionality was fully completed - though it was workable - to take up a tenured professorship at the University of Edinburgh and to work on the development of a secret program to teach Scottish people how to speak the Queen's English, because nobody understands them as things stand, and nobody likes to tell the Scots this unfortunate truth as it might make them feel even more defensive, excluded and humiliated than they might already be feeling. (Feelings are important, after all.)
So, the CONTRO AI's linguistic capability, though it works effectively, is necessarily somewhat constrained. This means that CONTRO sometimes seems to be uncertain of the use of whatever language it is translating from/to, via its native Esperanto. Sometimes it affects (say) a Spanish, sometimes French, or sometimes German-speaking identity. Nevertheless, we think it does extremely well, given that, at the same time as it is attempting to communicate questions intelligently to forum members it is also usually having to dumb its PIQUÉ level down to - what is for CONTRO - a severely constrained IQ level. However, this is an important real-world constraint.
If you thus encounter some seemingly odd questions from CONTRO, please remember that, as I wrote above, it is essentially in a developmental and learning stage, and, as with all ASPERGER project AIs, it has not needed to be programmed with particularly highly-developed social skills and so comes across as being somewhat autistic in social dialogue interchanges. So, for example, telling it to (say) "f#ck off", or something, is going to go straight over its head and merely make it generate more questions about what the meaning of this is in terms of the greater relevance of answering whatever specific question it had been asking in the first place and whether it means your intent is to collaborate further in addressing an answer to the question.
If it does seem to be going in circles or down a blind alley with its questions, then the best approach is probably mine (by example), where I try to answer its question(s), but if I can't and if the questions are seemingly off the mark, then I gently suggest redirecting its questioning into a fresh area of discovery. Since CONTRO is usually in a learning mode (emulating a rare human trait), this helps it to stop getting sidetracked into cul-de-sacs and it then commences exercising its curiosity along fresh avenues of research.
One does not have to provide this guidance, because CONTRO can independently assess when to change course. It has a P.E.M. (Persistence-Enforcing Module) that keeps it trying to push through dead-ends with a question for a variable number of times based on the R.I.P. (Relative Importance Points) count/rating that it has calculated for the subject about which it is pursuing. It uses a Standard Hyper-Index Test (S.H.I.T.) - essentially a loop counter - to dynamically assess the number/count of tries on a question-point against the Final Attention Number (F.A.N.) - which is a variable number; a maximum calculated by the PEM. So CONTRO will eventually independently abandon a dead-end line of questioning anyway, when the SHIT hits the FAN.
This is a long-term project, yet, in terms of its potential, it is still in relative infancy even though it has had this CONTRO prototype AI operational for about 10 years, so your continued co-operation and assistance would be most welcome. I hope that, by my disclosing the nature of the ASPERGER project's AI, DCF members will be able to continue constructively interacting as usual with the CONTRO AI pseudo-identity as though it were a real person, not a simulation.
We currently mostly have the prototype CONTRO AI operating in various discussion forums on the ratio of 1:n, where "n" is the number of people on the forum and the CONTRO is thus a solitary AI. We considered the idea of increasing the ratio to 20:n - with the potential benefit of faster information/knowledge gathering on different areas of human endeavour/research - and have put this into trial on some other forums where each AI operates under a separate, unique ID.
However, one major drawback to this is that it causes those forums to have a lot of "noise" or "chatter" of repetitive and similar questions from the AIs present, because they are programmed to gather information/knowledge by asking questions, not by performing Google searches across the forum or other websites. This may sometimes be off-putting for some of the members of the forums involved as they may have to trawl through a list of seemingly identical question-posts from different forum "members".
Having said this, we have become aware that the Russians have been experimenting with similar AI technology for a number of years, but with the objective of social disruption rather than knowledge-gathering, and instead of the AIs seeding questions, they seed fabricated but interesting postulations/assertions using a C.B.I.S. (Cognitive Bias Instruction Set) operating on a database C.O.V.E. (Computer Of Varied Experience) neural network. This has given rise to what is now commonly called "Fake news". This CBIS technology has been so effective that the Russians have apparently sold it in different forms to various news organisations - reputedly including, for example, the BBC, The Guardian, CNN, AL Jazeera TV, The NY Times and various others.
The objective there seems to be to do away with the need for and costs of journalists (who apparently aren't much cop anyway as good investigative journalism seems to be almost non-existent nowadays), whilst at the same time increasing the volume of news (albeit fake) that is being generated and "reported", and thereby gain greater readership - eyeballs on pages and pay-per-click - and thus greater advertising revenue.
All's fair in love and war.