A.I. is magic, isn't it?
I am often baffled by the external perception (ie. people who don't work with this stuff everyday) that is portrayed about A.I.. I also have issue with the usage of the term A.I. - as far as I am aware, we / humans, have not yet successfully created an Artificial Intelligence. We have created human children, that have somewhat of a limited intelligence, but we have absolutely and genuinely no idea "how" we created the childs intelligence, that is a mystery. Before, b!tching at me that we are the all knowing masters of the planet Earth and we are far superior, we have scientists and we know how everything works.... we don't. fact. We can sit down with pen & paper in my garage and draw out exactly how the cars in my garage work. We can pick any piece, strip it down, figure it out & rebuild it. Because we have an understanding, we potentially can even improve certain elements, such as carbs or fuel injection manifolds.
However, we (mad scientists apart), cannot do this with a human. Yes, we have moved on leaps and bounds over the past 150years, we have a lot more understanding of how human parts work, external limbs and internal components. What we haven't truly figured out is the four square inches that sits inside your skull. I'll also agree with you that we, as a race of humans, have figured out quite a lot by various means, some ethically, some not so, in order to understand how the human brain works. What we do not understand is how human intelligence works. Therefore, it is kind of technically correct to state that until that happens, we cannot create an Artificial Intelligence.
What we can create though is a sh!t load of code that performs Machine Learning and uses algorithms to make best guess choices and mis-label that as Artificial Intelligence.
Why the random rant about this? Well, I stumbled over this article on the BBC News website:
https://www.bbc.co.uk/news/technology-56414491
Having been someone working "in the Industry" on this stuff for the past, 5-6years, I can 100% confirm and agree that what you don't see or hear about is the exact point that is being shown here. People are being used to make the "Intelligent" decisions, ie. "look at this image, does it contain an animal, if so what one and could you also draw a box around it for me....thanks!". That takes a human brain a second or so to look at something, analyse it and determine and output and perform an action. We are sooooo far off a computer system doing something at the same speed or accuracy as a human being.
I have had many a discussion with some very senior company executives, who totally understand how to run their business to get the maximum amount of money out of it and still keep their employees employed, but when it comes to the mention of the word, A.I., they glaze over and picture all of the movies they've seen that portray an A.I.. Then, during discussions, they will say phrases such as, "...and then the A.I. will figure it out and provide us with the data to make better more informed decisions, yeah?".....
Well, yes....but...in order to "figure it out", you need data....a LOT of data....and you need to have a way to understand that data.... age old saying = "garbage in, garbage out". Also, there is no magic A.I. robot sitting in a box someplace, just figuring stuff out like a 3year old human child. There are lines of code, written by human beings (and therefore inherently buggy), that execute some logical algorithms that make guesses, albeit intelligent guesses that can then provide you with an output that you can then decide to action upon. The machine is still dumb. It cannot be associated with "blame", if you actioned out the output and the actions caused you to lose $10Bn of your companies assets, you asked it a question and based upon it's limited (yes, I shall still state limited) access to data, it's "training" (we'll come back to this in a moment), it then provided you with a result. There was no intelligence involved, until the result was taken by a human being and assessed. Then intelligence was applied. Limited or otherwise.
Training data is a difficult problem. You think you're doing the right thing by providing relevant data that is associated with the problem you want the M.L. algorithms to solve, however, you can then add in a lot of bias, as has been proven in numerous public situations exposed by the worlds press & media. So, not only do you have a problem about needing a lot of data, you also need a balance of that data and you also need preparation of that data - ie. "labelling". Yep, you know it, a lot of large IT corps. pulled some funny fast one's on the general public (thanks public!), by making phone apps that allowed you to do X,Y & Z, and you did it as you got a benefit out of it, but behind the scenes, you are actually doing the manual "labelling" for them, with it costing them much - I won't say it's without a cost, as they had to make the app, test it, deploy it, host it, etc..etc...but it's still way cheaper than getting dedicated "labellers", however, and this is the ref. to the BBC News article, you also need levels of approval for that labelling and you need humans to do that mundane task. Think about the scenario when you remove those humans who are double-checking.... and some malicious "kids" decide that for fun, they are going to start labelling/tagging all cats in images as dogs, if everyone started doing it, the M.L. algorithms wouldn't have the Intelligence to know they are being duped. Initially, it would be detected as an anomaly, as a glitch, as an acceptable percent error....but as it happened more and more, if left "unsupervised" the M.L. algorithms would then start to retrain themselves based upon the data they have been presented with... and then cats would take over the internet....oh, wait a minute! ;-)
Whilst, we are making huge steps forward with the current tooling and mechanisms we have in place and I am mightily impressed with how far everything has come along over the past decade, we really need an injection of "something" to boost us to the next level, where we can have a genuine Artificial Intelligence made available - however, and I will point this at our American friends, once you have true A.I., you will have fear, human fear, to deal with. In the same light as UFOs and Aliens, if the general public were to be informed that such things existed, then they would feel threatened, suddenly we're no longer at the top of the food or Intelligence chain and natural human behaviour will kick in (literally), humans will then seek to either limit, incarcerate or destroy the Artificial Intelligence. I am saddened by this potential prospect, I seriously think we, as the race of human beings, need something to help us move along/up to the next level and we shall have to get over our ancestors instincts to kill something that threatens us, something we cannot control. We do it fine for the 3 year old human child, who then grows up within our society (that we control every aspect of) and we have absolutely no real idea of what "thoughts" are going on within the same 4 square inches inside that child/adolescent/adults skull - if we have learnt to "trust" human children, surely, we can learn to "trust" an A.I. when it arises.
Well, that all got a little bit serious a little bit quickly didn't it!?!
I'll see if I can find a video that I made back in around 2017-ish entitled, "A.I. is not magic", except I'm not sure if it contains too much work/work content to be made public.
update: found the video and have uploaded it and as it's "old" I don't think there is anything confidential discussed, so enjoy!
-------
The question that I shall giggle about for a short time is: Did "I" write the above content, or was it written by an M.L. algorithm that I fed some data and pointed at some online articles?..... you'll never know! And, that is a slightly scary thought to deal with.... you assumed "I" was talking to you, that "I" as a human being was conveying this information for discussion (albeit one-way).... it's going to be an interesting future. I'm getting a deck-chair and some popcorn....
--------------------
UPDATE:
I just noticed this article (the DAY AFTER I posted the above) - title says it all really :-)
Comments
Post a Comment