Talk Morgan Forums Community The Malvern Arms Artificial Intelligence

A lie is an assertion that is believed to be false, typically used with the purpose of deceiving or misleading someone. What constitutes a lies become interesting as there are many different versions of the truth and therefore potentially many different lies on the same subject matter.

Take just one area eg religious teachings about God and its various followers, and just the differences in emphasis between common faiths eg Roman Catholic and Protestant churches, within Islam between Sunni and Shite sects, differences of truth that has lead to global conflicts that have existed in some cases for millennia as a result of opposing versions of often the same truth.

Yet an atheist fundamentally does not believe in a deity / God.........how will AI balance those versions of the truth/lies when faced with different audiences with different beliefs.

Even simple historical facts will have different and conflicting versions of the truth according to who is telling the tale.

So its a more simple question for me, is how does AI differentiate between the many version of the truth and a lie which may arise from the various interested groups interpretation of an event and therefore AI even recognise the truth , so to be in a position should A1 so choose to condition it's response with a lie..........what belief system and values will AI have.

GIGO. I don't know how that smiley got there.

Yes to all of this. As I have mentioned before, 20 of us spent some time assisting the growth of one test AI model and managed to make it consider a very odd set of things as it's base of "truth" and so it was very happy to think it was not lying or cheating. Put a hungry unsympathetic state, company or leader under this who can just change the rules and you do have an opportunity for driving bad outcomes. The difference is you can drive them way cheaper, faster, wider and without potential censure. A bit like Wikipedia "who edited that?" is not said nearly often enough in my opinion. The younger generations also do not yet have way enough cynicism in them to be protected effectively.

But just wait until this next concern is highlighted. Some may not like it but we have a system of law that we need to abide by. It is a moving target, very slow moving. It has yet to really work out how the generic state of cyber-odd should be ruled, or even if it is within the boundary of UK law in many cases. It has little in it's book to deal with it. Not aiming to be rude but this is outside it's existing remit and it does not create the new stuff as quickly as the problems occur and mutate. How will we rule AI? In truth it needs to be global but I cannot see that happening. Even at UN level after all it is a global issue now. Perhaps a little like the model of air travel, airside and landside. There would need to a Cyberside rulebook which countries sign up to. Gambling and Pron industries would put some money into killing that I suspect?



This is my fear - he who pays the piper calls the tune. I fear most of the AI we will bump into will be the same as googleads - commercially driven. If you are not paying then you are the product of the tool. TalkMorgan remains rare and now almost unique in my browser tabs for having value but not demanding money. Along with the CPU/GPU performance of simple machines now having enough power to create deep fake content. The bar to audience manipulation gets lower and lower.

Moggo - That really made me laugh! "I personally feel such a prune at having enjoyed the antics of a comedian cross dressing as a polititian." Don't be so hard on yourself, just assume they are all a little bit cross in that respect! Comedian cross dressing as a politician, ha ha ha

I suppose, in theory anyway, that AI, presuming that it self learns, will look at religion, and assuming a scientific MO, would take the view that without evidence they are all simply beliefs. What AI might not understand is such global conflict arising from such arguably simple beliefs. One AI would probably process such contradictions internally, but what if two AIs start chatting?

The law as it stands will only complicate matters because to have varied and non compatible laws globally nakes no sense whatsoever. Humanity isn't logical, and so what this comes down to in the end, is how a super intelligent AI deals with such chaos, and from what base agenda. Whether or nor AI choses to interfere, just observe, take sides....nobody knows.

Can you imagine it trying to fathom the broad spectrum of human life for 5 (human) years and then ending up like Marvin from hitchhikers guide to the galaxy. Brain the size of a planet and all they want to know is what channel Top Gear is currently showing on.

I'm not sure how AI will be able to ignore religion, as in many parts of the world it dictates the daily cycle of life, from when to pray to what to eat and when to eat eg Ramadan and even the relative ranking of respective genders in different societies. Whether AI can entertain the relative philosophical merit of each belief system will be interesting, but I cant see AI advocating a Christian/western belief and value system with its equality of the sexes going down well in a strict more fundamental Islamic environment. Interesting times ahead.

It will be interesting to see if AI is given a separate legal identity very much on bases which companies operate, especially if the owners of the AI want to distance themselves from product liability created legal issues arising as a result of an AI application.

Of course AI can lie :-)

- it's framed as 'halucination' - here for example is Google's Bard on a passive heat recovery unit from a firm Zypho:

"The Zypho Slim50 is a horizontal shower heat recovery unit that is available to purchase in the UK. It is the slimmest unit on the market, at just 50mm high, and it can be installed under any shower tray or bathtub. The Slim50 has a 50% energy efficiency rating, which is the highest of any multi-fit WWHR device. This means that it can recover up to 50% of the heat from waste shower water, which can save you money on your energy bills. The Slim50 is also easy to install and requires no maintenance.

[that's all true!]

The current price of the Zypho Slim50 in the UK is £299.99. This price includes VAT and delivery. You can purchase the Slim50 from a variety of online retailers, including Amazon, Screwfix, and Toolstation."

None of this is true - it costs more than that, isn't yet available in the UK, and isn't available from those suppliers.

The trouble is that the correct stuff and the nonsense hallucination are indistinguishable. The best summary & caveat I've heard is to not confuse AI's impressive performance with competence

Will

On no account am I steering this discussion into yet another bash the politician thread so please don't go there ... but, if you are given information from a source that should be trusted and you repeat that information in good faith ... is that a lie? .... to my mind a lie requires that you know something is false but deliberately repeat it anyway ... otherwise it's just being wrong ???

The above example could just be poor inputing to a 'normal' computer .... with true AI we have the possibility of the required sophistication for 'real lies'

K

I posed the following question

will AI be able to self evolve past its original programming
Sent message. will AI be able to self evolve past its original programming

I received the following politician type answer

That's a very interesting and complex question. There is no definitive answer, but some researchers have proposed that AI could evolve by itself using concepts from Darwinian evolution, such as survival of the fittest and mutation¹². They have created software that can generate and improve AI programs without human input, such as POET (Paired Open-Ended Trailblazer)¹². POET is a kind of training dojo for virtual bots that can learn to navigate different environments by trial and error¹². The idea is to let AI discover new ways of solving problems and perhaps develop different kinds of intelligence².

Source: Conversation with Bing, 15/06/2023
(1) Artificial intelligence is evolving all by itself - AAAS. https://www.science.org/content/article/artificial-intelligence-evolving-all-itself.
(2) AI is learning how to create itself | MIT Technology Review. https://www.technologyreview.com/20...ntelligence-learning-create-itself-agi/.


A more honest answer would be

We don't know!

At least at the moment it is citing where it is seeking its reference material, is that AI or is that just a clever lookup tool, to my mind's eye it is a lookup tool and the AI part is how it humanised the presentation of the data.

To me there are two distinct species of AI.
The first species is unweaponised AI, which will be data mining every second of every day and algorithms will sort this data and use it to sell you something or persuade you of something. You will be digitally pigeonholed and targeted by various entities, based on what the data says about you when sorted in various ways. CAVEAT EMPTOR!
The second species of AI - weaponised AI, will be loaded with algorithms that will dictate whether it shoots you or not.
I know which one scares me the most.