Science fiction : The Drift

 

Science fiction : The Drift

I was dusting the book shelves and trying to retrieve whatever old and unused items I could secure from the discarded corners and shelves of my room, when suddenly, my fingers felt something like a keyboard. I pulled it out along with the mouse forced into one of the corners in the carton box where this keyboard was placed. 

It brought back many memories, memories of the bygone days, when keys on the keyboards needed some real tapping in order to make the letters appear on the screen. And do not ask me about the mouse. Mouse was almost like someone you really love and you want to convince her that you really do. And the wooing process takes unusually longer than expected, depending on the perseverance of the young boy, and the willingness of the girl to realise his amorous approaches. 

Ah, romance at its peak!

When it came to controlling the movements of the mouse to manage the cursor/pointer on the screen. The user had to invest a lot of time and attention, and finally when the mouse felt sufficiently admired, it somehow let itself be used with ease and without annoying the user anymore.

A reality of the days when personal computers were becoming popular and gaining attention from everybody: scientists as well as an ordinary witted person like myself. And when I held the mouse in my hand, it felt as if I was holding the past in my hand and interacting with it. For a while everything appeared to be a reminder of the past that I had experienced in the early 90s; where I had romanced the mouse for at least 1 hour, to manage the movements of the cursor on the screen. 

Highlighting a letter in the Ms. Word used to be the toughest part, but once you did it. You realised now the mouse does not think that the user at the other end is a mean cat. For me it took 1 hour of persuasion and training to convince the mouse that the hand that felt its surface, was that of a human trying to control the placement of cursor on the screen and not the paws of the cat that it dreaded.

Yet, sometimes the hand would lead the mouse beyond the boundaries of the table and almost stay suspended in the air for less than a second. Then it took the user a while to drag it back to the desired position, and recommence the romance. It took most of the users a day or more, to actually learn using the mouse like a pro.

That was the mouse then, and if we compare it with the mouse today. The mouse does not need any romancing, it is already in love with the user. So it seems! The agony and despair of the cat is only going to deepen with every milestone achieved in the technology focused on computers and related technologies. 

And in 2023, Artificial Intelligence and Machine Learning are technologies that are very much like the old mouse that need a lot of romancing and wooing, in order to be managed with highest accuracy and zero probability of error.

Yes, ZERO. An insignificant value when it is alone, but a significant force when combined with other numerals. But in AI and ML, ZERO, even when it is alone, holds infinite value. Because in AI and ML, achieving ZERO error is the primary goal of all companies who have invested heavily in AI. However, given the fact that as of now these technologies are like the old mouse, achieving and then claiming ZERO error will be like being able to tell, when the cat will show up, without having belled the cat. And that is impossible to achieve.

Maybe you are already wondering why. If you are, here is why.

There is something called AI DRIFT. And this Drift exists in ML as well. It is nothing, but with time when a software that generates intelligent outputs based on AI and ML models, develops errors and enters into a degradation phase, as far as its data accuracy is concerned. 

It is at this stage that experts say, “AI model has entered into AI Drift!” It is like developing an anomaly that is inherited by the AI model based on its data rules and conditionals, that fail to generate accurate output, when its algorithms fail to cater to the growing database and its arrays of attributional qualities. For example:

A weather prediction system

Let us consider a region that experiences 4 seasons:

1. Spring

2. Summer

3. Autumn 

4. Winter

Few of the data sets

1. The fluctuation in temperatures is always recorded ( It is usually between 2% to 5% )

2. The winds that flow annually are recorded

3. The humidity level is registered

4. Rainfall is registered

5. Sunny days ( with hottest and coldest temperatures ) is registered

And lot of other parameters

Based on these datasets, the weather prediction system predicts weather in the target region.

And one of the conditional arguments passed into the AI model says: If wind x flowing from region Y, is carrying 60% humidity, and the temperature in the target region is 30 degrees or above. Then there may be rain in the target region on the 2nd day, from the day and time this data pattern was recorded.

So far everything is perfect. 

The database over the years gets larger and grows quite colossal. At a certain point, the system fails to realise that when the speed of the wind is higher than value ( let us call this value XWIND ) then it rains sooner or on the same day, but if the speed of wind is lower than the value, XWIND, it does not rain at all. Because by the time, X flowing from region Y, carrying 60% humidity reaches the target region, it loses all its moisture.

Over the years, these incidents increase, and the value of these data sets increases too. But the AI algorithm was not designed to factor these conditionals while arriving at a decision. 

So, as these data sets compound like compound interest in this database. The AI model now begins generating incorrect output, because it will tend to get confused whenever these unaccounted for datasets appear in the database. 

And in an AI database, all data sets, whether accounted for or not accounted for, tend to have an impact on the overall accuracy of the database, not at the early stage, but at the later stage.

And it is at this stage, AI develops AI DRIFT. 

But the human mind is very special when it comes to this, it is capable of identifying these deviations and then factoring them in, to generate best and reliable outputs. Therefore, I believe all AI models would need some human intervention at certain stages of the database growth. Once the database has achieved a certain threshold of complexity and human intervention has resolved it. Then the database can be refined by adding new algorithms that work in harmony with the core algorithm. Maybe then, at a highly advanced stage, no matter how large the database grows, it will be able to redefine and reset logical arguments intuitively.

But that brings us to a new question. Who will define this intuitive level of thinking in the AI models? If it is a person who is a villain by nature, then we all are doomed. Imagine this in medicine, finance, city planning, traffic management, water supply management etc. One bug will have the potential to bring the entire world to a standstill.

Therefore, some sort of human control and override will always be required if we are not to become the victims of one of humanity’s greatest technological leaps.

Human mind has a limit after which it feels an intellectual freeze, but it manages to come out of this freeze after a few pauses or well timed rest. Mr. Isaac Newton is the best example of how far the human mind can push itself.

AI can be limitless, but for it to be so, it will first have to learn to become as good as the human mind, and then by embedding millions of human mind like nodes into these colossal databases. AI drift can be easily managed, because one of these human-like brains will be dedicated to understanding and fixing this AI Drift. However, for every industry, these brains will be different, since data sets vary in every sector. 

Database corresponding to traffic cannot have data sets similar to database dedicated to Healthcare. Therefore, they will tend to experience AI drift of  a different kind and at different stages. My guess is, traffic related database will develop it sooner, as incoming data frequency will be huge. But the healthcare database will develop this drift at a later stage, but the complexity of this drift will be one that would need a lot of attention and human intervention until the drift pattern is understood. The process can make a person feel completely jaded, if not slackened.

But with AI’s proliferation in several and almost all sectors, it is indeed getting smarter, and as it will experience many AI Drifts, it will finally evolve into a mouse that is designed to love the user at the very first interaction. But that will take some time and resolving many AI Drifts. 

In this case, the weight of the evidence is incommensurate with the strength of the claim. Because in AI and ML, it is objective reality, than subjective reality that matters and decides the efficiency. Although the hardest thing in life is taking what one does not want. In databases it becomes an inevitable fact where a database has to learn to accommodate every data set and then arrange it in meaningful ways to achieve highly effective output.

Put in simple words, sometimes and in worst case scenarios, AI Drift is like a toad sitting in an electric kettle, that is plugged in, while the toad is thinking water will not boil.

And to salvage the toad and humans from experiencing an AI drift of apocalyptic scale, I wrote my science fiction novel, They Loved in 2075. It is a science fiction romance, of a different kind. Where reality of 2075, has evolved into a mouse that does not want to be managed by humans at all. It has developed awareness and it is clicking everything and everywhere randomly. And in the midst of this self aware world of artificial intelligence, humans are struggling to emote like true humans, because they have developed a pseudo layer of artificial emotions. And gradually humanity is losing its ability to express true human emotions that spring from the depths of one's feelings and heart. So, it is a science fiction romance with a message that will motivate you to fall in love again with all those people you already love. It is with this hope that I wrote this science fiction novel, They Loved in 2075.

Enjoy falling in love, enjoy the feeling of love, for who knows how long it will last. 

And who knows, if humans in 2075, would even understand the meaning of these lines of a lover from our age.

Even the winking of the eyes had ceased, 

For it disturbed the rhyme of our feelings, 

Love in our life had doubtlessly increased, 

Inspite of times passing moments and their wanton stealings, 

So, she did not wink, and neither did I, 

We lay there still, covered in the morning light, 

I was glad that finally I could say, it was I who she loved, just I, 

Not winking, hearts throbbing, in that love embrace tight, 

In 2075, it may be a Doppler effect at emotional level, only that in this case, the frequency of human emotions will reduce, because the source and observer will continuously be moving away from each other and not towards each other. 

And for one major reason: We humans are yet to evolve a subject that would address a very important aspect of technology, that I call, “Sociology of Technology.” 

In 2023, humanity is at several crossroads, almost like Tom Hanks in the concluding scene of his blockbuster movie The Cast Away. And for us to be able to take the right turns, it is the right time to come up with a new subject like: Sociology of Technology. Only then will human species be able to retain human values and actual human identity in 2075, and say proudly and happily, “They Loved in 2075 just like our ancestors did in 1990 and earlier.”

Let us not be dispirited, because nothing can stop the arrival of 2075. But in 2023, we have the power to define how our future generations arrive in 2075. Let it be a grand entry befitting of thinking beings, like us, the humans!