AIML - A Career or Confusion?
Abstract: This blog is aimed towards professionals across business verticals who are eager to upgrade their skill set with AIML learning. Professionals who would like to secure, upgrade and re-fit in post chatGPT, Grok world. Without understanding the context of AI specific to one's personal use case, one might end up paying a hefty amount to some institute and end up wasting time and feel bewildered at the end of it. What should i do now? What's next? Questions like these. This blog aims to clarify the world AIML beyond marketing objective. What it means to a professional at different levels and disciplines.
Terminology: The term AIML is an merger of two concepts - Artificial Intelligence and Machine Learning. Artificial Intelligence is an outcome of Machine Learning. Just like a subject matter expert is a product of knowledge he/she has consumed. Add to it, his/her experience of applying them in real life. In the same way, it is with digital computing devices. We call it Machine Learning. Artificial Intelligence mainly refers to the generative capability of this discipline. Generate a response in text, image, audio or video format which is fresh. The input data may be newly generated raw data or today in 2024, it may have already generated by another AI. This concept is not actually new. Earlier also, historical data was used for future prediction. The tradition goes back to even earlier ages. Astrology is defined on the basis of same logic. I remember in my career in the early 2000s, FMCG giants like HLL (Hindustan Lever Ltd now Unilever) used historical data for material requisition, warehouse planning and a lot of things. These were Data Warehousing projects which was run for sometimes 15 days. Those were days when Intel was the most valuable company and everything was dependent on CPU power. But there was a twist.
Theses traditional prediction models or if you call Artificial Intelligence in today's term, were completely dependent on structured data. Means data that has very definite structure and interrelationships and even constraints. It was stored in RDBMS like Oracle, SQL server etc. The query or logic was mostly written in SQL - a language to deal with structured data. But since then, a lot has happened. The source of data today is mainly the other one - Unstructured Data. It refers to expressions in text, images, videos, audios in any or all of these. Here comes the modern age Machine Learning and Artificial Intelligence. Machine Learning has implemented theoretical base of statistics beyond 3 or 4 dimensions. Earlier Oracle Data Warehousing was talking about 16 dimensions which was considered as theoretical limit. No practical and commercial solution at least in my knowledge implemented and benefited from all 16 dimensions. Here is the magic of ML courtesy GPU and its parallel processing. Today we, the python programmer group behave like spoilt child of millionaire! We care very less for memory, processing speed, context switch overhead! Most of the people not even aware about difference between compiler and interpreter. The GPU has opened the flood gate. So essentially the scope has widen to experiment with any number of dimensions. We may never be able to express or even know the function. Or imagine the graph of say 590 million dimensions, but that does not stops us to build a machine which can remember the equation and use it to interpolate or extrapolate for new data. That brings the the job of a statistician to be replaced for better may be, by a AIML professional.
But wait! We are now into post ChatGPT era. Means all of the tasks that can be thought of with unstructured data like text and multimedia inputs are already done by someone or multiple people/entities. They are now available for free mostly. All that you need to do is to just download the model and start using it. Even for that, if you don't have enough infrastructure, cloud is available. A considerable portion of real life work can be managed by using free versions cloud hosted solutions.
So, the job remains is to supply these AIs the correct prompt. The birthplace of another term or wing of AI - prompt engineering. Essentially it is the same as putting the right text in search engine query box so as to get best result. But much more complex and complicated than that. To the extent that in 2nd year of Post Graduate Degree program in some cases, the focus is to teach the secret of it!
So, AIML as discipline affects people in many ways. People with technical background after learning all the fudas of stats and algorithms, may just be paid heavily for effective prompt engineering! Which apparently at a high level seems to be disconnected from custom computing, application of statistics or any calculation for that matter. Of course the deeper understanding helps in many ways but that is not the essential pre requisite for one to become a highly paid prompt engineer.
Today, when we hardly require any new or custom algorithm to catch the low hanging fruits of AI, many business houses are inclined to these. In Indian context to be particular, i attended an webinar organized by a Indian IT giant in early 2024. It was attended by who's who of all Indian legends of Indian LinkedIn profiles. All were focussed on this. Someone even calls it "Jugad AI"! The most powerful person on that platform, the chief of the sponsor entity uttered the phrase "Catching Low Hanging Fruits". It is a standard answer against any original R&D attempt be it making a movie or experimenting in any domain especially in IT. No need to reinvent the wheel.
So, new jobs are created and old ones are deleted. Definitions are rewritten. A techie or programmer may not essentially mean computational programming. It may just mean the ways to control the AI. Here domain knowledge is more important than the skill set. In many organizations, the team is divided into grids of horizontal experts and vertical experts. Horizontal refers to the skill set that affects or works across business verticals. Like be it Retail or Manufacturing or Energy, you need .Net, Java professionals. Their value is the knowledge of technical stack rather than business verticals. The more computing framework advanced, focus has moved on steadily from horizontal to the vertical. It is the vertical guy who owns the problem at the end of the day. He needs a translator to tell it to the computer and that's where the horizontal fellow steps in. Now what can be a more ideal solution for business where the problem owner is directly talking with the machine. Negotiating, calibrating the output with the machine directly. Saves a lot of time, effort and money. Makes sense.
So, if you are associated with a Vertical, check out for the latest AI offering in that space. There is a good enough chances that you will find already available COTS (Commercial Off The Shelve) products in that space. You may start learning it as a power user right now or you may come back after 1/2 years of full fledged PG degree or certificate and start the journey more confidently. But end of the day, realistically, this is the place you will find yourself from practical point of view.
If you are looking for change in vertical and begin fresh in AIML as a horizontal domain resource, you will most probably start with a Business Analyst role. Here, you will analyze the data, produce a lot of graphs may be interactive ones to enable better informed decision making. Here also you need a very thorough understanding of the vertical which is called domain in AIML vocabulary.
Domain dominance is apparent as more and more training service providers now focussing on additional specialization/moule with one or max two domains.
It is not easy to start a fresh with AIML foregoing the past. As a professional one has achieved something like a certain decent pay package, designation etc. He/She has also aged. Belong to a different age group now. All these are difficult to forego for the professional and even for the employer also. You may be willing to accept a 4 Lakh per Anum package while your current CTC is say 7.5 Lakh but the employer may not be confident to offer you. Today while 3/4 years regular specialized degree course pass out is available, the employer is not bound to compromise at this point.
Overall, it is better to start with a vertical or use an existing AI like ChatGPT, Grok etc to start with. Going for a long journey which focuses more on theoretical aspects or behind the scene logics, may not find an appropriate application area with monetary rewards easily.
Training: Like today, we can feed an AI the base data which is already generated by another AI, there are professional trainers in large numbers who earn a living by teaching AI while they themselves have never found a chance to be a part of a business use case. The platforms under pressure of providing employment opportunities to the students, encourage this practice. It is a potential employment area but with very high volatility, high stress level and less rewarding in monetary terms. Although it has some other softer benefits like being recognized as a teacher, it is not a very stable and long term solution for all.
Entrepreneurship: AI has created an opportunity. Today even one single person with a viable use case and business vision can dare to build up something to compete in the open market space. But nevertheless it is not free. Be prepared for cloud costing which can range from 20K to 1 Lakh per month INR. It is best if you can form a group of like minded people a bit early. May be a group of 3 or 4 people who can contribute and form a kind of co operative or partnership firm. Working on any idea/project needs multiple considerations. If you think NLP, the copyright question is the one most probably another cost center. Data procurement, preparation, check out the legal check boxes are 80% of the work probably. Technical solution design and fine tuning may even be completed within a few hours. Of course you can continuously evolve. Make some more fine tuning, improving performance etc but still technical work does not mean anything if you don't have customers, live feedback.
There are quite a lot of ideas that can be taken up by a group of enthusiastic people from social and cultural point of view. These are the areas where society will greatly benefit from and the creators may get satisfaction and return beyond expectations. Here are some of the ideas:
There are quite a lot of ideas that can be taken up by a group of enthusiastic people from social and cultural point of view. These are the areas where society will greatly benefit from and the creators may get satisfaction and return beyond expectations. Here are some of the ideas:
- An NLP chatbot that can be my spiritual guru. One can take a base model and train additional copyright free texts of ancient wisdoms, one or more guru's writings, speeches to train a model which can guide an individual. It is like Yudhisthir seating besides Bhishma while he was lying in arrow bed and asking what is his Dharma at that point of time. User may ask questions about self, family, relationship many things. Depending on the use case specification political questions may or may not be under scope. Btw, a political guru can also be the primary source of reference. Makes it interesting! Isn't it?
- An NLP chatbot for healthcare: Healthcare is becoming a global issue. Every family is under financial pressure and threat from it. Most of the people try to avoid a Doctor's visit unless the issue becomes critical. Also, sometimes local doctors prescribe costly branded medicines which may not be necessary or even sometimes can create other side effects. So, why not make a generic medicine prescribing AI doc as solution? With all disclaimers and possible problems of such prescription may end of the day work far better than 'expert advice' or common knowledge. This one can be further extended to sub domains like ophthalmology, dermatology etc.
Comments
Post a Comment