Introduction

Today, I shall also be speaking something which is not very, very technical.

I have been speaking at this venue for last almost six months now, and generally most of the talks are very technical.

From Technical Talks to the Human Element of AI

But today, I also thought that we should talk about the human element and talk about us, how we can leverage what AI brings to, and how the lay of the land is shifting, how the The AI world, how the industry requires more and more domain experts like all of you in this hall today.

So that's my LinkedIn QR code. If anybody would like to, feel free to connect. And if you have questions, all of them are more than welcome.

A Case Study: Mercor and the Rise of Domain Experts

Has anybody heard of a company called Mercor? This company.

these guys these are three kids from california last week the company got evaluated as a 2 billion dollar company unicorn he was talking about it what are they selling are they selling ai or

they're selling something else anybody amazing product market fit the product market fit is phenomenal though there is no product per se there's a platform but it is something which is serving all of us anyone who has heard about them would

What Makes the Platform Different

like to share so essentially these folks they have a company which has a platform where domain experts like all of us could create a profile and you could provide your services on hourly basis so right now this is from the job side so

you have different options available as you to sign up but what is the importance how it is different from any other job portal all the AI companies

Why AI Needs Domain Expertise

right now require us who are sitting here either when you are developing the models or when you have deployed the products based on large language models multi -model models you need domain experts to help you train or help you align the systems and that is why they are what they are they are evaluated at

whatever 2 .5 billion and they've just raised 350 million youngest billionaires beating Facebook CEO Mark Zuckerberg because there is a need there's a product market fit what gary was talking about now with this context i would like to draw your

Agentic AI in the Real World

attention to what i was talking about so this is based on my own research and my own understanding and helping certain fortune 500 companies in silicon valley on adopting agentic ai how many

of you are using agents so a lot of us are using agents right in some other form some of you may be developing them some of them some of you may be using them at your workplace or your personal

Model Drift and the Role of Experts

environment you would have seen as agents come into production they start drifting because they finally are based on generative models generative models drift so we need

domain experts to guide and help the development team or the product team a line and that job cannot be done by anybody else your data scientists machine learning engineers your product teams cannot do that you need subject

Healthcare Example: Doctor–Patient Transcription

matter experts for example if let's say you have developed a transcription tool to have the conversation between a doctor and a patient recorded transcribed and submitted to your electronic health record using

generative AI who could be the domain expert who could help us align the output any any guesses who is that whom should you hire as a domain expert more

data engineers machine learning experts or more business people sales people marketing people any thought no for doctors it's a medical yet so no one was for sure if they would know the what they're typing but who else exactly yeah you need medical healthcare professionals who understand the context that is what i'm trying to draw your

attention to and that's why that company has become what they have become and that is where

Introducing the AI Connector

i would like to draw your attention for next 10 to 15 minutes to this concept called ai connectors where we would like domain experts to come forward and contribute to finally doing what they want to do in their careers and move the needle in terms of AI revolution right and this also helps making AI more aligned and safe so all

Adoption, Impact, and the Alignment Gap

of us are using AI either at workplace or in our personal domain adoption is quite high there is market evaluation that it could surpass almost a third of the fourth gdp in the world in next five years in terms of the market value a lot of us are using it for our

knowledge work undoubtedly good impact there are issues of safety alignment there are gaps in performance when you only use ai or you use a combination of ai plus human expert so what we're

The Agentic AI Paradox

trying to solve here is this typical paradox which we have with agentic ai which is a cause of concern and what is it it essentially is that if as you see there's a massive investment which is going on in agentic ai or ai per se but then the organizations are not able to take it forward

there is a gap you deploy these systems but you do not get the outcome what you were expecting to because the output of the systems are not very much aligned what you had thought because either we didn't do any kind of alignment or we didn't do proper monitoring or we didn't observe it properly we didn't govern it properly

and that gap is something which is only be fulfilled by humans domain experts so So this is what the dilemma is.

Two Extremes: Full Autonomy vs. Full Supervision

If you give full autonomy to agentic systems, they can be very unsafe. They can also cause a lot of harms.

So for example, there's an example where an autonomous system continues to do what it's supposed to do and it does something called emergent behavior where it goes beyond the the set parameters and causes a lot of harm to the company you must have heard about this incident

where entropic when they released lot 4 .5 during testing the engineer had asked the system to shut down and the system refused to shut down saying that it would not shut down have you heard about this particular incident yes so those are the self -preservation behaviors of these systems

where come what may they will try to achieve their objective, which can cause a lot of harm and which can be very unsafe for us, our organizations, our society and humanity at large.

That is one extreme.

Second extreme is it's humanly impossible to have so many subject matter experts to supervise these systems, right? We are doing nine to five, right? Five o 'clock. You packed up.

All of you are very, very interested in AI. that's why you're sitting here but otherwise we would have gone home we would have like done our second part of our lives spend some time with family and whatever gymming or whatever you wanted

Real-World Risks When Agents Run Unchecked

to do but agents are at work at night as well when you're sleeping you don't want your let's say if you work in a bank or an office you are the cfo as gary was saying and you have five agents deployed you do not want your agentic ai financial assistant to sign off a check of five million at night though the authorization for that particular agent was only $500 because

either it was jailbroken or it got tempted and it did whatever it wanted to do just imagine if you have these agents deployed and organization level these are the scenarios which can come up which can cause lot of losses so we need

Why Traditional Human-in-the-Loop Falls Short

to strike a balance and these are the problems which cannot be fixed with the traditional human in the loop mechanism which we have been relying on for generative ai previously

Defining the AI Connector Role

with the traditional machine learning approaches right now towards that what we envisage that there is a re alignment of a new role which i would call as an ai connector and a new operational framework which is most of it is based on what we exist we do right now but a little shift so that we can we can pivot that also allows us all of us to to pivot to be ready for ai so that is an

ai ready career move and what exactly it can consist of so seeing all of you as domain experts most of you are like i just want to ask with a raise of hands how many of you have more than 15 years of experience so we have almost 20 percent people more than 15 more than five years of experience five years and i assume the the younger lot is less than five so we have a pool

of experts who have spent 15 to 5 years of their time doing that particular domain right nobody knows that domain better than you if you are in ca you are an auditor or you are a doctor nobody Nobody knows it better than you.

Human Judgment at the Core

No machine, no data scientist, no ML engineer can actually replicate your knowledge. That is what we want to actually bring in here, that your knowledge gets utilized to orchestrate the whole agentic system. It strikes the chord between what is ethical, what is not.

It helps us train because what is happening is, the moment you put agentic AI systems to production they start drifting they start hallucinating there are biases which come in which to somebody who has not worked in that domain they will not come to know they for them it looks very glossy it looks very fluffy looks very

Catching Drift and Hallucinations in Context

human like English but it may be drifted totally it's like going back to the example which we are talking about transcription tool let's say the doctor is talking to the patient about cold and cough and the solution starts

fabricating outputs which are talking about something totally different but looks very medical or looks very aligned until it's a doctor or a nurse or a medical professional reads it we may miss the mark and you may diagnose the the patient incorrectly or a wrong history may get

embedded into the EHR right that is the whole idea about this particular concept where we bring in the experience and try to curate it as the orchestrator of human and ai workflow and what

Core Competencies and Training Pathways

could be the core competency which we are proposing and this would come as a course or as a program through university so you will have a option of actually getting certified by university if it moves ahead the way it is planned so essentially domain expertise almost 40 percent

of experience which individual has some understanding of how to deploy and use different AI tools and as Gary was mentioning very rightly to use AI to create value to make money or save money or increase efficiency we do not have to

Using AI Like Electricity—Safely and Effectively

build AI we just have to use it as it's repeatedly said AI is the electricity of our age so for this electricity we don't build it we don't go to a hydro or a nuclear or a thermal power plant to build it we just have a subscription we pay for it similarly for ai we have a subscription we pay the apis we we pay through some other subscription but

we know how to use electricity safely if we don't use it safely it can cause lot of issues we know it can it can be very very severe for our property for our lives so we strike that balance Similarly for AI we need to know what is the safe way of using it that is what

should be here and then as Gary also mentioned we need to actually start thinking from a systemic view we should think systemically instead of thinking of point solutions or system thinking to solve the problem.

Evaluation, Ethics, and Governance

AI suffers a big way from misalignment so we need a very very articulated idea of how to evaluate the outputs both from the business and technology point of view ethics and

governance will remain a fundamental concept if you want to ground AI to make it more aligned to human values and obviously you would like to have that idea of communicating it to the larger clients or your own stakeholders what Gary also mentioned earlier and with this combination this particular persona

Orchestrating Human–AI Collaboration

of us or future AI leaders would be able to strike the balance between what the business needs and what technology can provide similarly to to orchestrate this particular concept this framework of human centric AI orchestration which which will allow human and AI collaborative work can be a great game changer.

Autonomy Levels and Outcomes

And you would have seen this, read about it, that there are different level of autonomy of agentic system.

So it could be, like it can start from level zero, it could be level four, where initially it is totally controlled by human.

From Driver Assist to Full Autonomy

so it's like when we are driving a car which has certain indications which is not autonomous but you generally get some indications if you are going off road if you are departing lane but when it comes to level 3 or 4 it can be akin

to autonomous driving level 3 what full -scale full self driving on Tesla would be it today that you are leveraging autonomous systems but there's a human behind the wheel and level four could be a robotaxi where there is no human it's totally

independent autonomous it's running on its own but you have the arterials in place you have systems in place which will make sure that it does what it's supposed to do right and when we have

these four level of five level of autonomy you will have to see where it fits how it fits so we

Safety, Efficiency, and ROI

did some some trials and we found out that if you are able to have this kind of concept where there human and AI systems collaborating together you will be able to make sure that reduction of critical errors is quite good and efficiency also increases

a lot so we are going ahead with both safe solutions which are very very efficient and which can definitely give you that return on investment and keeping systems very very safe and aligned which would be the requirement

Technology Must Serve Business Goals

of any business and as Gary was mentioning earlier I mean we are not here to solve a problem only with technology because business will only pay for either cost cutting or revenue generation right so you have to align the technology to solve a problem it cannot be just a technology demonstration

Solving the Paradox: A Balanced Path Forward

and with this in mind once again a recap of how this paradox can be solved

because you will find that more and more market hype would be there for deployment of agentic AI but if agentic AI is to be deployed in a true sense we will require more human and AI collaboration and human supervision

Scaling with Human–AI Collaboration

alone may not be sufficient because we do not have so much of human capital available and this also allows us to repurpose lot of our human capital to a new role and as we know that definitely there is a huge market of which AI can

transform it is also allowing us enabling us to reposition ourselves as future leaders leveraging AI and this dilemma will remain and this will get a little more complicated as AI safety becomes an issue as we move forward so

The Middle Path: Safe, Aligned, Ongoing Deployment

that is why it is important for us to have more domain experts coming forward to take on the responsibility of aligning AI to what would be the the middle path so that we continue to deploy it and continue to train it the way it should be if you leave it to total autonomous mode it could be quite unsafe and

The Orchestrator Model

misaligned again these are the few solutions which we propose and this is one of the model we call it orchestrate and the human becomes the orchestrator as it happens in any symphony orchestra where as the orchestrator you are able to leverage different positions different notes

and conduct a symphony and that is what humans would be doing and this is the repositioning which you will see happening as we move into next few few

The Decade of Agents Depends on Experts

years because earlier they said 2025 is a year of agents but now they have repositioned it as the decade of agents so you will find a lot of autonomous systems coming into use but for their success it will be important for them to

to be very much dependent on this kind of collaboration with domain experts.

Next Steps and How to Get Involved

And these are the two QR codes.

Join the Waitlist and Expert Pool

If anyone would like to join the initial wait list whenever the course is released, so the one on the left is for the website.

And on the right is for the pool of experts who would be joining.

So once we have more details, we'll be sharing it with you.

Building Partnerships and Opening Access

Right now there is a website which is just as a landing page available and we are in process of syncing up with a few of the universities and other institutes to create a course so that we can spread the word and help more and more individuals, more and more professionals to see this side of AI where they can leverage their experience and for that they don't have to have a data science, computer science, AI degree.

Closing and Q&A

they can definitely be the expert what they are and still help us use AI to solve that business problem so with this I'll stop if there are any questions please feel free thank you