The fake news in political campaigns will become somewhere north of just epic. It is not a pretty future unless those who will peddle fake news (about anything that cannot be substantiated) are guaranteed significant jail time.
2 Likes
This is a couple of years old now, but there’s a “Where is Julian Assange?” thread on Reddit in which people suggested that a video interview of him was faked.
Scientists have developed an Artificial Intelligence system that works exceptionally well but they do not understand why it does so.
Perhaps they should be keeping Stephen Hawking’s dire warnings regarding AI in the back of their minds at all times.
This is intrinsic to the problem. If you have a rules-based system where humans write the rules you can reach in and say ‘this is where that decision was made’. Any AI that learns is re-writing the rules as it goes. The criterion for success is after n learning iterations how well does it do, not whether we understand how it has done it.
So will the outcomes of an AI system be predictable? If you want the system to be truly intelligent you would want it to find novel solutions to problems so the answer is no.
Testing complex human-coded systems is extremely difficult even when clear, detailed and explicit goals are available to test against. How do you adequately test a system whose goals are fuzzy and outcomes are unpredictable? Assuming that it is tested as a black box, and the behaviour is adequate under a wide range of circumstances how important is it to know what happens inside the box?
3 Likes
Also the assumptions used to originally code the learning behaviour of the AI system give it a “personality” a behaviour trait sort of. So like a baby that grows up with it’s concepts of the world coloured by it’s upbringing the AI system also “grows” with that same sort of colouring. Good or bad (what is good or bad is subjective) then depends on the parental input.
1 Like
This has already been shown in algorithms used by US police forces such as the NYPD. The algorithms appear, despite ‘good intent’, to be racist - having been written by white men and without input on ‘reasonable behaviours’ from other population groups. Of course, policing in cities such as New York has been very focused on black on poor communities for a long time - so an algorithm that looks at arrest records and such is going to be biased by the underlying data.
One can imagine a creationist creating an AI that simulates the universe right back to its creation in around 4,000BCE in six days. Garbage in, garbage out (GIGO).
4 Likes
Now Artifical Intelligence is predicting scientific discoveries years in advance.
I’m not sure whether to be excited or terrified!
1 Like
That is my feelings also.
A common response also for the contestants on ‘Perfect Match’ when Dexter’s AI was let loose with his predictions?
The real test of AI might be a computer that can play and win at chess without being programmed with the rules of the game? Currently AI is still reliant on the algorithms and learned content that is preprogrammed into each system.
I have some bad news for you - from 2017:
3 Likes
Starting from random play, and given no domain knowledge except the game rules, AlphaZero
Yes, it won, however it was given the rules of the game as a starting point. The learning algorithms were also tailored to game play.
Very clearly it knew it’s objectives and had a set of constraints or rules. Very clever programming?
A work on progress, and very clever programmers. 
An article regarding leading Australian scientists calling for the Federal Government to act on the evolution of artificial intelligence in Australia.
Hopefully the Government will do so.
1 Like
looking at the level of response so far to the financial, social and political impacts of rampant connected technology one cannot be optimistic. Looking at the frequency that any long term planning is done or anticipation of future problems is dealt with effectively you would not bet on any action if given long odds. The planning horizon of all our governments is at most the next election and often the next week.
1 Like
Aussies do like a bet though, and rank outsiders do get up about once every century if The Cup is a rough guide.
If that rank outsider comes in, what odds the rest of the AI enabled world will simply bypass Australia as too hard or irrelevant? A more certain bet.
What odds that same bet is not already on the table with Facebook and Google? Given AI and Waymo (Originally Google’s driverless car project) is at one bleeding edge of AI.
2 Likes
Yeah… no. My money (literally - my tax dollar) is on military applications being front and centre. The past 35 years have shown how much fun a war can be when almost none of your side’s combatants is in harm’s way (I’m starting with the first Bush invasion). Expect that trend to continue to the point where a few rich countries can fight a war without suffering any casualties but inflicting horrendous casualties on the poor sucker that forgot might is right.
5 Likes
OK, this isn’t creepy. Not at all! 
4 Likes
Why computers often don’t perform well in the “real” world.
4 Likes
I must admit to thinking that this was self-evident. It is, for instance, the reason why a computer is incapable of producing a random number.
How do we get random numbers from computers? We introduce an external factor, that is then used to ‘seed’ a pseudo-random number generator - something that computers can do. Cloudflare uses lava lamps. Other applications can draw upon entropy from the way you move your mouse, or the external temperature, or a dozen other factors. Some even use electron drift - randomness at a quantum level.
This StackExchange answer describes computerised cryptography fairly well.
Of course, if quantum computers ever become realistically useable then they will be able to deal much better with chaos.
2 Likes
So it is about quantum computing.
When I saw “chaos”, I expected that the article would be regarding recycling in Victoria.or climate chenge inaction globaly.
2 Likes
