How AI/ML/Generative AI can at times be totally wrong and unreliable

We all have read a lot of good things about AL/MI and how it is shaping the world rather how it is disrupting the world and giving us new insights for new opportunities.

Well, I feel we need to be cautious in over relying on AI/ML since it relies on data and no one is sure if the data is true or not. Collected data may be technically correct in terms of quality and suitability for analysis but its truth value cannot be ascertained.

Let me explain by a couple of examples and hope hear some back from the data scientist community.

First example relates to Recommendation Systems. As most of you be knowing recommendation systems comes into play whenever we are trying to make a new purchase online and based either on our past behavior or based on ratings given by the other customers, we are thrown up a number of choices. It is a similar concept as NPS (Net Promotor Score). Recommendation systems could be non-personalized or personalized.

CASE 1:

So here is a real-life situation. Assume we wish to try out some restaurant for the first time. The moment you try to search, Google would throw at us restaurants with some good ratings for us to choose. Now the challenge is we assume that the ratings have been given by real customers. And this is the real mistake we could be doing. See the actual communication I had today (30th November 2023) when someone messaged me. The number which was displayed was (+63 9638521792). Actual Screen shot of the initial conversation is copied below. I blocked the person eventually. But it got me thinking. How the data on which AI/ML or for that matter any analysis rests can be abused and misrepresented.

The purpose of sharing this screen shot very blatantly here is to ensure that we as sane people should really be careful of such scams in rating elucidation. ?But then what happens is this kind of data is not bad quality data but the truth value is bad. So next time I am looking at ratings for restaurant (for that matter anything), I would not be sure at all. And all the good intention of AI/ML has gone for a toss and its reliability itself is under the cloud. They say Data is the Fuel of the Knowledge Economy but what if the fuel is adulterated. And I do not see any method or any efforts to curtail this. With people losing jobs (see my case 3 below) the ethics could go out of the window and people may fall for such scams.

CASE 2:

Recently read the news article about the deepfakes related to some celebrities in India. Already govt has sprung into action but then it has shown how these technologies can be misused. For example, if someone was to make a wrong video film of mine using deepfakes and circulate impact my family, I would be doomed. So, we need to be careful.

CASE 3:

This is more of a repeat of the issue I had raised in one my earlier articles. It relates to sentiment analysis. In sentiment analysis the words, tone, etc used by a customer over voice communication or written communication is collected and analyzed to understand the customer satisfaction level. ?But the rating of sentiment or analysis of words (good, bad, neutral) is done based on exact meaning as per dictionary. But we always forget that individuals have limited vocabulary and for want of a correct word for articulation, the person may use a more harsh or soft word than what he intended. Here also it means the flaw gets into the system and so sentiment analysis may not very reliable.

CASE 4

I am sure all of us may have called customer service of banks and came across the automated bots which run you through the menu of self service and prompts you to choose various options. Frankly I never felt the connect in these interactions. I felt is it too cold and too inhuman. Time has to call spade a spade and bring back the human interaction. Many similar experiences have been recorded. In fact in my earlier article I had shared the episode of why human touch is needed in AI based on one reported incident (https://gadgets.ndtv.com/apps/news/amazon-flex-bot-firing-rating-hiring-human-workers-algorithm-ai-2474937)

CASE 5:

This is about post Labour economics. What is post Labour economics? It tries to answer the following question “How will the economy work if we all lose our jobs? Which industries will collapse and Which industries would survive. One can see the full video at the following url.

https://www.youtube.com/watch?v=eD5GlCIS0sA

From an article published in TechCrunch.com (https://techcrunch.com/2023/11/27/tech-layoffs-2023-list/#janlayoffs) the point I am trying to raise here is this. The tech industry has seen?more than 240,000 jobs?lost in 2023, a total that’s already 50% higher than last year and growing.

Earlier this year, mass workforce reductions were driven by the biggest names in tech companies like?Google,?Amazon,?Microsoft,?Yahoo,?Meta?and?Zoom. Startups across many sectors also announced cutbacks through the first half of the year. And while tech layoffs?slowed down?in the summer and fall, it appears that cuts?are ramping up yet again. So where are we heading?

In conclusion, I can say that we can find many more cases from real life where we have failed the AI/ML by not having some check and balances or then it has failed us.

And so, in a world where the lines between natural wisdom and artificial intelligence is getting blurred, where we let data alone rule us (perceptions, feelings are not considered), I am glad to say that I have unearthed a process which helps in regenerating our thinking mind. The process encapsulated as a software application (our team had developed it many years ago) uses rationality (Data) and Intuition integrated together instead of discrete and mutually exclusive components. It uses the collective wisdom of the team and lateral thinking to come to a consensus action plan rather than leaving it completely to AI/ML. And anyways AI/ML would not work in a situation where underlying data does not exist (Some pioneering project for example) Would be glad to demo it to organizations who want to focus on effectiveness in solving any real-world problems. Do connect with me as and if needed.