Usually we find two tribes of thought leaders circulating in the market; one category has experts with great proficiency in the sphere of their choice but great believers of the self-defeating concept of power distance; this tribe typically has a machinery of pseudo marketing and social media agencies working overtime on their behalf to grow their fanbase and followers. The other category has leaders who are highly accessible and incredibly humble, but cursed with a superficial hold on the subject matter, which is what they try and compensate through polite and proactive communication with the community at large.
Hearteningly, there’s a rare category of mavericks who do not crave market attention, who are extremely democratic in their communication with the world at large, and who have achieved great length, width and depth in their disciplines and vocations. No wonder, this tribe is a rare and endangered species but hearteningly, the number is more than what we would instinctively like to believe. We have the pleasure and privilege to meet one of them in a Q & A session: Bhaskar Roy, renowned expert in AI, data science, and sustainability and revered Startup Mentor and Advisor.
Talking to Roy, I am instinctively reminded of legendary filmmaker Satyajit Ray’s summary of his first interaction with French Avant Garde director Jean Renoir when Ray was yet to become the Ray we all know and rave about: “Renoir was not only approachable, but so embarrassingly polite and modest that I felt if I were not too careful I would find myself discoursing upon the Future of the Cinema for his benefit.” Just replace cinema with Data Science and the same sentiment applies in toto!
Excepts of my conversation with him, presented as a Q&A interaction for ease of readability and comprehension.
Broadly speaking, in your reckoning, where are data-driven insights taking us, and what are they fetching us across different spheres.
That’s a great question to start with, Sudhir! At its most basic level, data-driven insights are taking us into a new era of decision-making, where businesses and organizations can make more informed choices based on real-world data; no wonder, the world beyond is unraveling at a very fast pace with new tools and technologies surfacing on the shores with astounding regularity. This pace of development is something that’s not been seen for quite some time, most certainly an inflexion point for envisioning how and where the world of AI can lead us. By having access to data and insights, businesses can make better decisions about everything from marketing campaigns to product development.
Data-driven decisioning brings us several concrete benefits across different aspects of businesses. Few examples worth a mention include:
Efficiencies: Data-driven insights help businesses identify areas where they can improve efficiency, such as by streamlining processes or identifying areas for cost savings.
Customer satisfaction: By understanding customer behavior and preferences, businesses can tailor their products and services to meet exact needs and even surpass them, which in turn boosts customer satisfaction and loyalty.
New product development: actionable data can be used to identify new product opportunities, as also to test and refine new products before they are launched.
Innovation: Data can be used to identify new ways of doing things and to develop new products and services.
Reduce risk: By understanding potential risks, businesses can take steps to mitigate them.
Better resource utilization: By understanding how resources are being used, businesses can make more efficient use of them.
Stay ahead of the competition: By using data to identify trends and opportunities, businesses can stay ahead of their competitors.
These benefits are seen across industries like Healthcare (improving patient care), Finance (better investment decisions), Retail (personalized shopping experience), etc., even though the maturity of each industry, and organization within it, differs!
As AI moves from general intelligence to super intelligence, will human intervention eventually become a disposable commodity?
Let’s take a step back to frame the question better. The maturity/sophistication of AI systems is typically classified into 3 broad parts –
Artificial Narrow Intelligence – learning algorithm is designed to perform narrowly defined tasks. Specialization is in one area, solving one problem. Most typical uses cases today fall here, for e.g. image and facial recognition systems, voice assistants, machine’s playing different games like Chess, Go, etc.
Artificial General Intelligence – here, machine can be made to think and function as a human mind, especially when faced with an unknown task. AGI is typically assessed on parameters like Common Sense, Context awareness, Transfer Learning ability, and causality, but we are still some time away from it.
Artificial Super Intelligence – this is where machine consciousness comes into play, and the machine becomes more like an intellect that is much smarter than the best human brains!
Having defined the different phases of AI sophistication, let’s go back to the intent of your original question – where is this all leading us, and what role do humans have in it!
As we look at the current state, we have done reasonably well on ANI, with some more years to go before we get to AGI (although the whole GPT revolution is helping give it a significant push). In the coming times I expect a lot of the mundane to get automated, to start freeing up more and more of the human brain to focus on creative tasks.
With that said, even for the automated tasks, human-in-the-loop is going to be the right way to go about things. This is because a key part missing for machines is the context and feelings in which decisions are being made (getting towards sentience, if I may call it that!), and ability to take learnings from one situation to another. This is going to continue for some time to come, and I fully believe that the current wave of AI is only going to take human creativity and ingenuity further.
At what level of maturity is the industry in countering typical ML challenges In your experience, how is the quality of entry level data science professionals in India?
Interesting question and pertinent too! In my limited understanding, the art of data science (Machine Learning being one of the arrows in the data scientists quiver) has lost its way a little bit.
We have a situation where far too many people calling themselves ‘data scientists’ have no clue of what they are doing. They might know the right libraries in python (or their programming language of choice), know how to import and execute them, but have no clue of how to interpret results that the system spits out. Many a times they don’t even want to massage the data first to understand what it’s all about, clearly a situation far from ideal and moving towards being unsustainable.
For a young data scientist starting out, it’s very important to understand the nuances of what they are doing. They should be able to understand the problem/ issue they are trying to address, and the context in which their final output will get used by end stakeholders (essentially elements of problem solving). It is only when you comprehend this, that the data comes in. It’s important to massage the data to understand what its all about, before starting to do anything else with it.
This is what then enables one to answer all the questions you had around understanding reliability of input data, sanctity of the test and control data considered, and over-fitting.
Will it be a good idea of institutionalizing a robust mechanism across all domains where probabilistic AI results are systemically validated in the light of intuition and discretion, to counter the perils of ‘over-obliging’ algorithms.
A short answer to the question is – Yes, it’s certainly a good idea as it would help to ensure that AI systems are not making decisions that are harmful or unfair, thereby reducing and/ or eliminating inherent bias.
This is typically done by establishing a Responsible/ Ethical AI framework, driven by multi-skilled team of experts in AI, Ethics and Law, anchored on basic principles of Transparency (no black box solutions), Fairness (equitable and inclusive treatment), Explainability (tracing back of decision factors) and Human-centricity (understand and solving for biases).
Of course, there are also some challenges to institutionalizing a robust mechanism for validating probabilistic AI results. One challenge is that it can be difficult to define what constitutes the different tenets of Responsible/ Ethical AI. Additionally, it can be time-consuming and expensive to validate AI results in this way. However, I believe that the benefits of institutionalizing a robust mechanism for validating probabilistic AI results outweigh the challenges.
Is it right to say that predictive and prescriptive analytics are likely to draw maximum mileage from ML models in terms of making intelligence accessible and actionable compared to the descriptive and diagnostic analytics?
Yes, I think it would be right to say that about predictive and prescriptive analytics, but a lot depends on maturity of the organization and industry where the solutions are being deployed.
At a high level, ML models can be used to identify patterns in data that would be difficult or impossible for humans to find, and they can be used to generate predictions and recommendations that can help businesses make better decisions; for instance, ML models can be used to predict customer churn, identify fraud, or optimize supply chains.
Some specific examples of how ML models are being used in predictive and prescriptive analytics:
Customer churn prediction: ML models can be used to predict which customers are most likely to churn, so that businesses can take steps to prevent them from doing so. For example, Netflix uses ML models to predict which customers are likely to cancel their subscriptions, and then sends them targeted offers to try to keep them on board.
Fraud detection: ML models can be used to detect fraudulent activity, such as credit card fraud or insurance fraud. For example, banks use ML models to identify suspicious transactions, and then investigate them further to see if they are fraudulent.
Supply chain optimization: ML models can be used to optimize supply chains, so that businesses can reduce costs and improve efficiency. For example, Amazon uses ML models to predict demand for products, so that they can order the right amount of inventory and avoid stockouts.
One thing to note here is the level of maturity of the organization where these solutions are being deployed. For maximum efficacy, and impact, such organizations already have majority of their data issues (like data availability, quality, etc.) sorted, they have already done some level of Descriptive and Diagnostic Analytics (essentially understanding what has already transpired, and why it went the way it did!) for some period with decent adoption, and are now trying to get ahead of the game.
Currently, every other AI company seems to be recruiting prompt engineers. Notwithstanding the criticality of right prompts in helping AI models generate accurate content, is there an element of fad in the current deluge of recruitment?
An excellent question, Sudhir. There indeed is a lot of interest in Prompt engineering as a field. A lot of it is driven from the fact that mass market Generative AI technologies are still nascent (for e.g. ChatGPT), and still need to evolve to a stage where features that let you co-pilot with them become more widespread. They have begun, but will need more time to mature.
This basically leads to a situation where, in the current context, to drive optimal value from these tools, a whole deal of manual effort needs to be put in to get the desired output from these systems, and that’s what Prompt Engineers help us do.
Having said that, I feel it will be a short-lived phenomenon. On the other hand, the time-tested value of people with the right combination of business understanding and ability to solve problems will continue to hold good for many years to come.
Which sectors are likely to lead the AI evolution and progression in your reckoning, the first among equals so to say?
It’s a little tricky to answer this question. AI has been a buzz word for many years now, but it has impacted different industries in myriad different ways. If we look at the traditional B2C industries like Banking, Telecom or Retail, AI has typically been used across the board to drive customer marketing efforts, optimization of logistics, supply chain, customer targeting, etc. This, in my view, has happened because data is available in humungous volumes and velocity in these industries.
Next in my mind has been the development over the last 5-7 years where significant amount of data has become available in the manufacturing related industries, emanating from sensors embedded in all kinds of machines (from the biggest of the biggest to the tiniest ones we carry in our hands/ pockets. This first led to a unique problem where removing noise from the data was crucial before you could let the magic of AI-ML unfold and drive business value. This value is going to continue accruing for some time to come and will manifest itself it much smarter machines.
Next, I expect industries like Pharma and Healthcare to also start seeing significant value realization from AI. These have typically been industries where consolidation across different types of data, namely structured (e.g. patient details), semi-structured/ unstructured (e.g. doctors’ notes combined with diagnostic reports OR different types of image data) has been a key issue. Some of these are challenges are getting resolved and the infusion of AI are making processes more efficient and even amazing for everyone involved!
Do you see some scope for reviving Mathematics as a stream of choice towards creating a pool of potential AI resources in the time to come?
Great question again, and something that’s very close to my heart!
I have a strong bias (!) towards Mathematics, and related fields in the furtherance of a pool of people that are trained on AI. These are the core building blocks of what enables people to have the skills to create not just new kinds of AI, but also to enhance human intelligence and ingenuity. The right understanding of mathematics and related fields provides people the faculties to not only create new AI systems, but also understand what’s happening within AI systems (whether created by them or others) and where changes might be needed to move towards desired outcomes.
Knowledge of AI alone won’t suffice. Along with it needs to be inculcated the ability to problem solve. This is extremely crucial as even the best of AI systems exist to solve for specific problems, within the specific context of the industry, vertical, business or government, where it’s being implemented. Take away understanding of the problem, and the context in which it operates, and you are left with a cutting edge system that does nothing!
Is it right to say that Python is the reigning language of choice for data science? How are R, Scala, Matlab, Hive, Pig and the like faring? How would the language front evolve going forward?
Sudhir, while I’m no expert on this, I can say for sure that there is no one horse for all the races. Python has served the purposes well so far, given its flexibility and ability to adapt to changing tech environments as it lends itself well to integration efforts.
Having said that, tools like R, Matlab and their ilk also have very specific application areas in which they work well. For instance, R, before the advent of Python, was the programming language of choice in academia for those looking to have flexibility in data manipulation and analysis, but it has decreased in popularity over time.
Other languages like Scala, Hive, and Pig have evolved to address gaps in existing systems and languages for specific purposes. I have a feeling that these will also continue to co-exist with other languages, albeit for very specific purposes.
As an aside, what are your views on the open source software movement? Do you see enough mavericks around you who are keen to look at the holistic picture, who are not overtly obsessed with serving the proprietary tribes of coveted brands - both in product and service spaces.
Again a tough one to foretell, especially in the backdrop of the OpenAI movement being bought into by large corporates, who are now trying to monetize their investments into it! I have a feeling that there are enough and more ‘mavericks’ who want to do this, but end of the day it all needs either philanthropists to sustain, or for the value to be monetized. Each has their own interest groups looking to make a buck!
Could you briefly share the ‘what’, ‘when’, and ‘how’ of the patents you hold – viz. Booster centric resource allocation and Resource allocation based on retail incident information?
Sure Sudhir! Let me paint a scenario for you – think about visiting a store. Browsing through the aisles, we have all seen folks shoplifting in different forms – people outright stealing from the store, or those that switch labels from low priced items onto those that are higher priced, and many others. Consequent losses for retailers come to anywhere in the range of 1-2% of revenue!
What we don’t realize is that a large proportion of this activity actually is pretty organized in large global economies and is formally termed as Organized Retail Crime. These group go stealing from store to store, and have pre-identified target stores, timings, etc.
Now, let’s introduce an element of Karamchand Jasoos in the mix, with the carrot in hand, or Mr. Sherlock Holmes for the natively uninitiated. They start by looking at commonalities, and seeing things play out on maps, and patterns start to emerge. Now, once you have caught onto the trends, you can actually be one step ahead of these shoplifting groups and can predict to a large certainty where and when they are expected to be seen next, enabling their eventual exposé and arrest.
This at a high level is what I and my team had done for a large Retail Chain!
Could you share few notes on your personal voyage : education, career, and passion pursuits.
My grounding in left-brain dominated thinking started off with a Bachelor’s in Statistics, before being balanced out with a MA in Economics, and providing me with a perfectly created, but coincidentally arrived at, recipe of getting started in the nascent field of Analytics/ Data Science (it wasn’t even called DS then!).
I’ve worked with a few organizations through my career, both on the service provider side, and with in-house teams at large global corporations, with lots of learnings along the way from great bosses and mentors, that I’m eternally thankful for!
I truly want to have an impact on the world around me, in small but significant ways. Besides my day job, I’m associated with multiple educational institutions in different capacities as advisor to students, mentoring student start-up ideas, technical assessments and the like. These give me a great deal of satisfaction and reinforce the feeling that the true extent of human ingenuity continues to develop and will unfold in ways we can’t even begin to fathom!
Any parting thoughts?
I’ve covered most aspects. Only thing I will re-emphasize – human creativity and ingenuity is foremost. It’s us, Humans, who have created AI and have the power to continue moulding it to serve our purposes. We must stay in the loop and have to stay ahead of the game to lead the world forward with our unique creativity and ingenuity!