Artificial Intelligence Technology | My Two Paisa or PaisaVasooli!

The Society online Magazine of the daily The Hindu, had in its front news an article titled “Hallucinations in bot land” by Krish Ashok.

I tend to prefer articles which are printed in a newspaper. A hard copy gives me greater pleasure, while I am reluctant to switch to ePaper.

So then, this article appeared in the Sunday Magazine section that following week. Article casually bumped into my view and I was interested.

I have jumped midway into the bandwagon of Artificial Intelligence (AI) and Large Language Models (LLMs). Bear with my digression – would it be okay if our AI engines input data of high-level intelligence? Casually asking!

It is like naturally selecting and eliminating, first mark holders in class from poor scorers. It seems to mean that not including data from low level intelligence would be advantageous. Aren’t we placing too much exacting technology while unprocessed thoughts are at a lower rung?!!

It feels like asking the AI to write on a special topic for a Secondary grader. When the AI is at Kindergarten grade level, then, what could be expected and judged as appropriate? When they fail in their performance, then it is time to recalibrate for the involved thinking human-minds.

I am not saying that we must not try to reduce the large amount of time spent to find relevant supporting points that are genuinely true source texts. It should not be forgone merely because of some frivolous technical shortcuts.

Sometimes, when an article or an essay is assigned to a student. The teacher does not want to read just the raw information, or the unprocessed data striven throughout the said essay.

Now here we are getting the help of AI to aid in long and large data repositories like the internet striven with many texts, videos, graphics, and other media data which are made-up, genuine, and hard to tell ones.

I am sure it is time consuming to read and to sift through it. But it is about human inference and internal interrogations leading to better understanding of the core issues addressed in all the content found on the net.

Hallucination | New Lingo Word But, Old for Me!

It was a clever move to explain the new terms connected with the new developing technology as an aside block in the news article. I was fascinated with the terms.

One of them stood out for me – “Hallucinations!” To understand this term, one needs to know where things have gone wrong in the thought process of a human. It is a human understanding of the term being so limited.

Can we make AIs solve it? When humans don’t have the full detailed understanding of the word and it’s implied outcomes.

Setting up filters that prevent such a situation may seem difficult and hard to key in data for machines to learn. But, why not see it as an opportunity to understand how it gets into such a state.

I have worked in various technical areas as a help files writer. If an AI takes over this task, wouldn’t it be easy to get updated help files instantly. Like instant sambar meets instant rasam, of course which combined becomes gotasu!!

Moreover, technical writing can be automated with some initial input and have it custom-made for each client to meet their needs.

The same applies to the research method too. Are we looking at genuine articles or people who may be saying things in their personal blog or other areas of social media which could be real or false?

How to discern the two? Krish was right in his article that made-up research information would be detrimental and spread false information.

What is the Solution?

We must understand why and wherefore the Hallucination occurs; what factor leads the AI to pick the wrong route that is dangerous for valid data research? It is not the technology that is plotting on us. But it means that we need to understand human logic clubbed with human emotions. We need to observe it much more closely.

Are We Running Away With Presumptions?

Let’s take in a class of KG goers each child is unique and I feel likewise each child sees the world differently. So, naturally we should have all children who are intelligent with varying response time. If wrong, they are taught again!

The same thing applies for hallucinations too! I understood from the article the hallucinations are never repeated. But I disagree, what if there were another hallucination, one or more that are slightly different from the first one.

Clarity Becomes A Key-Factor

From personal experience hallucinations are of different kinds and the mind is left in a twist in such situations. Clarity of thought may be missing but one could categorize the visuals which helps in controlling them.

My night-time dream is quiet complex and tends to leave me exhausted the next morning. It is my mind’s way of processing the stress of the day before.

When that happens, I would record my dream the next day first thing after my ablutions. Then find a time when I am calm and collected to separate it. Once I finish dissection of the falsehood from the truth, it is time to test it.

May Be If Discerning Habit Becomes In-Built

If AI Technology could allow for self-critiquing and self-evaluation using a method that discerns facts from fiction and plain made-up imaginative falsification. I feel that would help the AI Engines to work with discretion while providing more positive results.

If a filter for not producing any citation by AI or creating a non-existent weblinks are separated. The AI puts aside search results as genuine or as fake.

Verification by a ping and a record of such pinging calls were maintained then, it would help in providing better options for good search results. If AI gets a global filter that prevents it to make up sources and references.

But with the fast-paced Internet Technology and growing eBooks and webpages that are genuine then such sources must have a proper way to be cited. Each webpage on the net should have markings that say it is a genuine fact-based post and not a fabrication.

AI should be able to differentiate between fact and fiction for which either a globally agreed upon filter is provided, a cross verification pointer included in the process where the AI engine and servers can talk to each other while maintaining their record.

Even Open-Source Coding should have proper rules set across platforms so that AI engines don’t get confused with the search query provided to them.

AI just provides quicker access to the countless hours of search and sifting through webpages that are fake and real source. Then, marking the resultant source as genuine or as fake helps the human searcher to be able to decide and to plan whether to use that source or not.

Do We Want AI to Take Over Hard Task?

Yes, we do want them to take over the hard and repetitive jobs but with human caution being set in place. Every technology has some flaws in it. At this moment we are not able to see the extent of damage that can happen over the period of long use.

If the history of human faults can be coded and made into a learning lesson that is indelible, thus bringing caution in the process. This would be a natural way in which humans tackle issues, so the AI learns. Memory of the past errors could also become repository for better performances.

AI Errors and Future Reforms

AI should be given the permission to make errors in a controlled environment through proper application of the test runs. This will provide the required self-learning to self-correct course and angle.

Varied students and varied learning methods. But all are learning the same thing eventually. Yet, do they see it as similarities leading to a pattern that becomes an open question!