Artificial Intelligence? Really?
This (https://www.wired.com/story/guide-artificial-intelligence/) article from Wired magazine is worth a read as it explains that “Artificial Intelligence” has been nothing more than a myth from the time the term came into existence and what is actually peddled under that name is no more than machines that have been “taught” to recognise certain patterns in certain, very limited, circumstances.
There is an age-old IT industry truism that states “Garbage in, Garbage out” – the translated meaning of which is that if you feed erroneous sales figures into a computer designed to analyse sales patterns the resultant output will be entirely wrong and worthless. Following this fact it is obvious that any decision or action taken based on the machine’s output is equally wrong
In the same way, “training” an “AI” using (say) only white, male faces will cause it to develop a bias against anyone who is female, coloured or disfigured. Feed one only Bible-belt America language before setting it loose amidst today’s social media driven frenetic world and it rapidly turns into the equivalent of a far right racist as it “learns” from exposure to a broader set of inputs that other language forms exist and adds weight to “its brain” based on the algorithms it consists of that tells it to prefer language forms it comes across that are new or just appear more frequently – the equivalent of being shouted loudest or posted more prolifically on social media or being repeated or stated by people it has been “taught” carry more significance – like, say, the American President.
Microsoft had to turn their “highly advanced” customer service AI attempt off after just three days after it began hurling racist and discriminatory insults at folks who phoned it asking for their washing machines to be repaired. Microsoft’s error? It had “trained” the machine with the largest human language set available in digital form – social media posts – and the machine had happily adopted the mindset of the average social media user. The result – not pretty
There is no intelligence in these machines. They cannot see someone walking away from them in a street and “think” to themselves ‘Hey, that looks like Gary – I should go and say hello’ because they lack both the human mind’s ability to collect many orders of magnitude more “training data” from just a glimpse of the back of Gary as he leaves your house after a dinner party. Or facially recognise Sue from an 80% rear angle because you once saw her turn her head away from you to talk to someone else across a dinner table, revealing the full gamut of her profile right to the back of her head.
“Intelligence” (in just this most limited field) is the mind’s ability to take that glimpse you once had of someone called Sue and use imagination to “join the dots” and play probabilities that tell us that a view seen at 82% rear or 25% side of the opposite side of a head glanced at random in a busy street is the person called Sue – and instantly fills in not just everything we know about Sue (marital status, kids, parents, relationship to you, job …) but emotional values – do you like her, agree with her politics, want to say ‘Hi’ or walk in the opposite direction hoping she hasn’t recognised you!
The simple fact is that current computing technology can do some things incredibly fast (as perceived by the human brain) but when tasked with jobs that require the simplest imagination or emotion fall at the first step.
There’s a Nobel prize waiting for the first person who can explain what “imagination” or “emotion” even is – let alone develop an algorithmic expression that enacts it
Artificial intelligence does not exist. And, if it ever does, is many, many lifetimes in the future. It requires technology as many unknown orders of magnitude removed from today’s technology as our pocket computer marvels and massive computing clusters are from Alan Turing’s first postulation of the Imagination Game
Why is this important? What dangers does the current belief in AI cause us all?
Just this week, the London Metropolitan Police has responded to a Freedom of Information request with the startling news (to anybody who doesn’t understand any of the above) that in the entire time (years!) they have been busy running facial recognition technology against the millions of citizens daily going about their lives across the UK’s capital city just TWO “criminals” have been flagged by the system – one accused of non-payment of a parking fine (falsely as it turned out – he had the bank records to prove the fine had been paid) and another who was sought as a non-cooperating witness in a case the Met had dropped for entirely unconnected reasons more than a decade earlier. Of terrorists and members or organised criminal gangs or extremist protesters the system identified not one.
And … the Met had to write down the slightly embarrassing admission that 98% of faces “recognised” by the system turned out to be completely wrong. The wrong person entirely.
A fact that hadn’t deterred the ever vigilant Met from going out and harassing several thousand entirely innocent people who knew absolutely nothing about the crimes the Met was accusing them of being involved in.
On the simplistic belief that if a computer says something is true then it must be so.
Clearly those running the Met and other police forces and intelligence services round the world learned nothing from Operation Ore (cf; https://en.wikipedia.org/wiki/Operation_Ore and https://www.mintpressnews.com/operation-ore-how-sloppy-work-by-the-fbi-and-the-press-led-to-suicides/240277/) – or the hundreds of similar scandals where “2+2=99” has been fed into a computer and the resultant nonsense that comes out the other end is swallowed as gospel.
Hence my statement that the only difference between a bored Texas cop and an advanced “AI” machine is that the machine gets to the wrong conclusion a million times faster.
Read the article – I don’t need to explain any more.