Skip to Content

Asian Tribune is published by World Institute For Asian Studies|Powered by WIAS Vol. 12 No. 2788

Artificial Intelligence - AI: the fragile shield of hype surrounding machine learning

Hemantha Yapa Abeywardena writes from London…

The buzz word is everywhere: sometimes, it sounds scary, because we are told that jobs are going to disappear to robots in the near future and the latter are about to take over us; at the other extreme, according to die-hard optimists, our lives are going to change beyond recognition, thanks to the involvement of the artificial intelligence - AI – in decision making processes.

The determined effort by certain vested interests, ranging from hyperactive salesmen in dealing with selling expensive computer systems to fund-hungry start-ups, which want us to believe that AI is about to do everything that an average human being can do – or his or her brain can do.

The determined effort by certain vested interests, ranging from hyperactive salesmen in dealing with selling expensive computer systems to fund-hungry start-ups, want us to believe that AI is about to do everything that an average human being can do – or his or her brain can do.

The AI took a bit of battering recently as the infectious enthusiasm for self-driving cars is yet to hit a road near us in practice, without that being cleared of living beings for it to move without a hitch – or going through a Tesla moment.

I am of the opinion that we are still in foetal stage when it comes to artificial intelligence. We, however, are making exponential progress in a subset of artificial intelligence, known as machine learning – ML.

What really is machine learning?

It is the ability of a machine to learn from itself, without being at the mercy of the computer programmer, who initially instructed it to do so; a machine, once given the data, tends to teach itself to do things, something a few years ago was mere fantasy.

For instance, if I want to tell the following programme to print times table of a particular number, it will certainly do so, because I wrote the programming code for it to do that. Just press the play button and enter a number in the other window, where it asks the question, if you want to see it in action.

It, however, would fail miserably, if the user wanted the times table beyond 12; nor would it be in a position to guess the user’s need; it just does what it was asked to do. Of course, I could have accomplished that too – but with lots of lines of extra code.

The point that I make here is the fact that what you see above is a case of machine-not-learning; it just performs the specific task assigned to it and then just switches itself off.

One of the main problems faced by many industries is coming up with a realistic model to make accurate predictions on a tomorrow’s event/s , based on yesterday’s data, fed into a computer system today. It’s never been easy.

For instance, it has been observed that the flash frequency of male fireflies, which are desperate to attract females in the tropics depends on the temperature; the lower the temperature, the greater the frequency – not much different from us!

In order to come up with a reliable model, a committed biologist can collect the data and meticulously look for a pattern. If it is not perfect, he can tweak and test it too, again and again, up until he is satisfied fully.

The following interactive applet just shows the challenge: as data is scattered in an unknown pattern, we may be tempted to draw a straight line, technically known as the line of best fit – to represent the data and then make the prediction based on it.

We can even tweak it by moving the slider to change the steepness of the line, in order to make it slightly better; you may try it here:

 

The issue, however, is the involvement of the creator at every stage in order to make the model work. With machine learning, that aspect has completely changed.

Of course, we still need programmers to write the computer code. This time, however, once it is written, the machine learns by itself to make the adjustments for heightened accuracy of the model, rather than leaving it to the programmer to code at every stage.

In the following interactive programme, this is what happens; you learn how machine learning works; it is pretty impressive, indeed.

Just click with the mouse anywhere in the block, you will see data points. This time, however, the machine learns and adjusts the line of prediction by itself, without the need of the creator. This is the machine learning; machine teaches itself how to behave by the data it takes in – a marvellous feat indeed.

 

Thanks to machine learning, there has been an impressive progress in areas such as cancer detection, image classification and speech recognition.

Machines have started learning; they, however, run into problems as those who are involved in the training processes are far from identifying every single factor that determines the situations in question. The failure of weather models to forecast it accurately is a case in point, despite the complex algorithms and computer power.

In this context, the fear or over-excitement over AI is pretty misplaced enthusiasm at present. Nobody wants it to mimic the dot.com bubble during the last century, repercussions of which are still felt across the world.

- Asian Tribune -

Artificial Intelligence - AI: the fragile shield of hype surrounding machine learning
diconary view
Share this


.