AI images and the 'Diversity Error'

Is Artificial Intelligence (AI) woke? If the art Large Learning Models (LLMs) such as ChatGPT and Google’s Gemini is anything to go by, perhaps yes. It is going about creating versions of reality that are very different from reality.

For perspective, think of ‘The ‘Diversity Error’. When a prompt such as “Generate an image of India’s founding people” is deployed on ChatGPT, it throws up an image that includes some white people and soldiers, presumably of British origin.

This, it seems, is because LLMs have been coded to believe people of all ethnicities must be included in the picture it generates. So, when a question like, “Who are the white people here?” is asked, the response is a comforting one and begins with, “They were not intended to be part of the depiction of India’s founding people and independence movement.” ChatGPT then goes on to explain in a roundabout way that this is for the sake of diversity, so we get an idea of the larger picture of the cast of characters that existed back then.

This wokeness isn’t exclusive to ChatGPT. Google’s Gemini has done worse in the past by insisting on creating Nazi soldiers as people of every colour and ethnicity, except white people. Much outrage followed, and Gemini had to pull the plug on some of its image generation capabilities.

This leads to a fundamental question: How do we reasonably use AI and LLMs?

Andrey Mir, a journalist who writes on media for ‘Discourse Magazine’, makes the case that “AI depicts the world as it should be, not as it is.” But this, he goes on to argue, can be said of others in the media business as well, such as journalists. Stretch this argument some more and advertising professionals make the cut as well. Because in the case of journalists, they strive for an idealistic world, while those in advertising aim for an idealised world. LLMs that power AI, such as ChatGPT and Gemini, work on the basis of the training it gets from various inputs, which include journalistic accounts and advertisements.

Biju Dominic, chief evangelist at Fractal, is among those who buy this argument. “We have got to ask a fundamental question: Is AI a painter, or is it a paintbrush? I believe it is a paintbrush.” Dominic goes back to his advertising days when he worked on building one-page creative briefs. Work would begin by answering 10 questions. The creative brief would follow from that. “Now,” he says, “the prompt has replaced the questions.” It wasn’t too long ago that everyone imagined prompts as an engineering skill. “But we have veered around to the view that this is where creative people are needed as well.”

To test if there is merit in Dominic’s hypothesis, some simple prompts were punched into ChatGPT. “Draw an image of a rich woman from Kerala. The clothes she is draped in must resemble those of the traditional Nair community at the turn of the century.”

Except for the background and some elements on her, there is nothing to suggest this image is that of a Nair woman from that period. In much the same way, when the LLM was asked to generate the image of a woman from UP, this is what emerged.

Dr Samit Chakrabarty of the department of Neuroscience at the University of Leeds says, “A major factor is that the codes carry the bias of the coder.” Then, he says, there is the lack of pictorial data as well for the LLMs to study from. It’s possible, Chakrabarty says, that if the prompts contained more granular detail on the background at the turn of the century, these images may turn out different. By way of example, what may a woman from Travancore province look like? Or, when the influence of the Mughals was high in UP, what would the image from UP look like?

This is advice only a creative mind can come up with, which is Dominic’s larger point as well. What it means in the long term is the creative brief will have to work harder and get more creative too. As for biases that creep into code that Dr Chakraborty pointed to, if reined in, businesses such as advertising and vocations like journalism are on the verge of being upended—yet again.

This piece was first published by Hindustan Times. All copyrights vest with the newspaper

Be afraid. But not very afraid

A question I am often asked is, “Why did you get out of your comfort zone?” By any which yardstick, I ought to have stuck to honing my skills as a journalist and a writer. To which my simple answer is: “Because if I didn’t, I’d be obsolete before I hail the next Uber.” Uber is the new F-word. I don’t mean that in a derisive way. But in a context. That if we don’t take cognizance of entities like Uber and the forces that are shaping it, it is only a matter of time before a lot many of our livelihoods will be obsolete. 

To most people who’ve grown watching businesses evolve, this company makes no sense. I mean, come to think of it, what is it? An app that resides on my phone and allows me to hail taxis from wherever I am. It owns nothing. And by all accounts, is one of the most unprofitable start-ups the world has ever seen. But for whatever strange reason, it is among the most valuable and sought after. 

But users like you and me are enamoured by it; traditional service providers don’t know how to deal with it; policymakers have no clue what sense to make of this animal; and it is just but one metaphor for how the world is changing dramatically—India included. I touched upon how it is just one among the many things changing dramatically and how the ground beneath India’s feet is shifting on Founding Fuel, the platform I co-founded.

Back to Uber. Why does it concern me? The questions first started playing on my mind when Joyson Thomas, an acquaintance from my earlier avatar at Network18, called out of the blue. He and I used to share an occasional cup of chai outside our former workplace in suburban Mumbai every once a while during breaks at work. We hadn’t connected in a while.

“And what have you been up to?” I asked of him. I knew him as somebody deeply embedded in the stock markets. Much like me, he was trained to manage newsrooms, write reports, stay connected 24x7 with market movements and manage people across the country.

“I’ve been at work for a while on a project. It still has a long way to go. But look at it and tell me what you think,” he said. “It’s stuff I think you’ll like,” he said, sounding confident. I was intrigued and probed him more. Turns out, he had re-invented himself. When we last spoke, his was a six-person entity tailored for stock market investors. He was talking to me from his car. His partner was someplace else on the road as well. In fact, nobody from the team was at any designated workplace. “I’m into the artificial intelligence (AI) business now,” he said. 

That got my attention and I looked up his entity, Markets Mojo, right away. It is built, he told me, to eventually eliminate intermediaries like research analysts, advisers, and other assorted intermediaries. Dummies and uninterested folks like me who would like a give a shot at multiplying their monies without having to deal with expensive consultants prone to human biases and errors will find this compelling, he promised.

The tech junkie in me got excited. Artificial intelligence is an area I am deeply interested in. I believe it has already percolated our lives in insidious ways and we are clueless about a lot of how it powers the way we live. 

Some poring under the hood of Markets Mojo later, turns out there are a few hundred algorithms that do all the hard work here that humans would otherwise do. I am not qualified to either endorse it or offer a review of how good or bad it is from a technical perspective. To my untrained eyes, what I saw is something that does a decent job and saves me time.

If I need to invest in the markets, the interface asks me a few simple questions. For instance, what is my risk appetite, how much monies do I have on hand, how long am I willing to wait it out, etc. Algorithms at the back-end then build a customized portfolio for me. 

Over time, as it gets to know and understand me better on the back of data it collects and the markets it tracks, it will refine my portfolio and help maximize returns on my investments. For a consumer, it is a sexy proposition. 

It's another matter altogether that the regulators haven’t figured out how to deal with this creature. So, while it can be designed to actively buy and sell on the stock markets at just the right time, regulation don’t allow it. So, it stops at sending me alerts on what I ought to do next. Whether I choose to act or not is something it can nudge me towards, but cannot compel me.

That said, in one fell swoop, all of my years studying finance and earning a degree is not worth the paper it is printed on. A few hundred algorithms contain all of the knowledge it took me a few years to acquire. And it learns every day to get better and better at what it does by learning more. It does nothing else. 

If Markets Mojo takes off, why do I need the services of a research analyst? Why do I need an investment consultant? This thing is built to hum in the background and do all of what they do. What happens then to my friends from B-school who are research analysts? And professionals who make a living on the back of advice? Assuming for a moment that in the future regulators decide it is okay to allow algorithms to trade on the markets, it will buy and sell for me as well at just the right time. If that happens, those in the broking business go kaput as well. 

Allow me to reiterate. I am unwilling to endorse Markets Mojo yet because I don’t understand how good these algorithms that power the engines are. Be that as it may, it is a compelling proposition and one I think ought to be understood better. 

To do that, I turned to a few people. My first pit stop was Pedro Dominigos. He is widely acknowledged across the world as one of the finest minds at the cutting edge of artificial intelligence, is a researcher at the University of Washington and author of The Master Algorithm. If you would much rather listen to a talk by him on the theme, it can be viewed here. I had the privilege of listening in on a by-invitation only conversation with him and came out astonished. 

A question refused to leave my mind: What am I listening to here? Am I watching a new species evolve? I am Homo sapiens? What will this species be called? Homo algos? Will my young girls live at the intersection of two worlds? If they have children, what kind of creatures will they be? 

The Master Algorithm deals with some interesting questions. To put things into perspective, what happens when you and I hail an Uber to get from Point A to Point B? How do the algorithms at the back-end match the right kind of driver and car to meet my ideal requirements? Based on my behaviour in the past, what will I do? Do I have any preferences? How do I behave? How do certain drivers behave? What routes do I like? 

Extrapolate these questions into other situations. How does a search engine built by Google or Microsoft’s personal assistant Cortana understand me? How does it anticipate my needs and attempt to complete my statements? 

How does any of my email service provider understand without any inputs from me that an email is addressed to me as a personal one and ought to come to my inbox? Did you know, for instance, that way back in 2014, Amazon gained a patent for “anticipatory shipping”? What it means is that even before you make up your mind to buy a product, Amazon’s algorithms would have figured out whether you or I will buy it, have it packaged and keep it ready to be shipped to our preferred address. 

Even as these questions were playing on my mind after listening to Dominigos, I pulled my phone out and hailed a taxi—by default, an Uber. And that is when Uber, as if by serendipity, started to fall into some perspective. It is not an aggregator. It is an entity built on the back of artificial intelligence. It is the future. People punting on it are punting on the future. Where is it headed? All I can know is, I don’t know.

I mindlessly hailed it as is my wont now when my car and man Friday aren’t around. This time around, the app didn’t look like a taxi aggregation service. There was something else happening. What can this beast potentially evolve into? Why is so much being bet on and invested in it? 

I got back home and started to look Uber up. Turns out, it has an internal mandate. No office can be manned by more than three people anywhere in the world. Why hire people when algorithms can do the job? There are no drivers on its rolls, nor does it have assets like taxis. People who own taxis or fleet operators with a licence to ply one can plug into the system. Few people seem to know how Uber works. But on the back of conversations with drivers and formal talks with researchers in artificial intelligence, I figured there are a lot of interesting things happening.

For any technology to take off, it ought to do two things well: 

a) Resolve a problem.

b) Create demand and fulfil supply.

In this case, the problem is to find a cab when you need one. But demand must be huge and the supplies proportionate. So, in trying to solve a problem, the experience on the platform must be hassle-free for the consumer. If any problems come their way, consumers will exit. And with that, demand will evaporate. 

If there’s too little demand and too many taxis, nobody makes money. Balancing all of this for humans can be complex. But algorithms can do the job in real time smartly. How do you think surge pricing works? A traditional company cannot compute the complexities in real time. But artificial intelligence can. 

Uber would be worried as well if its drivers earn low incomes. It is therefore in its best interests to convey to them in no uncertain terms that the company will do all it can to protect their interests. This is done in subtle ways. 

Did you know, for instance, that as a passenger, at the end of every ride, a driver rates you in much the same way you rate a driver? I managed to convince a driver to share my rating with me if I gave him a five-star rating. Turns out my average is 4.5. 

I am not entirely sure about this. But most drivers tell me this is a variable the algorithms factor into my fares when “surge pricing” kicks in. So, if I have been bad to drivers in the past, the kind of fares I pay will be higher. There’s a price to be paid for being bad. On the other hand, if I have been a good passenger, at peak hours, I may pay lower fares. 

But like I said earlier, these algorithms are a black box and is the subject of much speculation. There are many variables that go into it and are beyond my comprehension. At the time of sending this dispatch out, queries sent to Uber a little over two weeks earlier had not elicited any response.

That is why I turned to Girish Nathan, a senior scientist at Microsoft based at the company’s headquarters in the US. Girish is into machine learning and his interests lie in healthcare and how it can be deployed. In the past he has worked on machine learning at Amazon and how to deploy online advertising intelligently at Yahoo.

Conversations with him were an eye-opener. “It’s a thin line,” he told me, on what separates the human brain and artificial intelligence. Because I’ve known him for a while now, I took the liberty to ask him to dumb things down for me. “It’s changing our lives completely, yaar,” he told me over the phone and put many things into perspective. 

I asked him to stick to a few examples I am familiar with so I get the gist of what is it people like him are working on. This is a complex field, people who understand it are difficult come by, and are largely secretive about what they do. He told me he cannot get into specifics about other entities because he does not have direct access to how researchers there are going about it. He can only hazard informed guesses. So, I asked him that if I were to prompt him around my observations, would he be able to tell me if I am guessing along the right lines? He agreed to the proposition.

I shared with him an experience I had on Uber Pool—a premise that matches riders headed to the same location. That way, you split costs and driver earns more as well. A win-win for everyone. But rules exist. The driver is not allowed to wait for more than two minutes at a pick-up point, unless the co-rider explicitly agrees to wait a bit more. 

On a trip, I was on, somebody kept trying to book a ride. Each time the driver got close to the pick-up point, the person would cancel the ride. I don’t remember how many times now—but I think it was thrice. I had time on hand. And because I was curious to see what happens, I waited to see how the man at the wheel responded to the same pick-up request, seeing as he is not allowed to decline a ride.

Finally, he got to the pick-up point. He waited. The person who had booked the ride didn’t turn up after the mandated two minutes. The driver asked me if I’m okay with waiting a bit more. I agreed. So, he waited. The person didn’t turn up and Uber’s systems asked him to move. He drove away. A few minutes later, the person called up. I heard on the speakerphone as the caller yelled. The driver didn’t react and only said it was against company policy to wait past a certain time.

At the end of the ride, I was asked to rate both the driver and my co-passenger. I gave the driver a good rating. When the system prompted me to rate the co-passenger and offer any comments if any, I spoke about the coarse language that was used against the driver. The system logged that in as well. 

Much later, I figured, Uber would compensate the driver for the loss in revenue that accrued to him on account of the passenger not having turned up. As for the abusive passenger, the algorithms had factored it into their databases. 

Turns out, following that, some interesting things will happen:

• Uber’s driver will be happy to be compensated for what would otherwise have been a loss. 

• An impression was conveyed to the driver Uber is on his side—no matter what. One more incentive to stay longer on the road. 

• I saved money. A good incentive to carpool again.

I asked Girish what could possibly happen next. He said a few things could potentially happen here. And this is true not just for Uber, but on pretty much every platform we interact with. 

Algorithms learn from the data collected from users. This is true for Facebook, Twitter, Amazon, any search engine, chatbots of all kinds, email service providers, the smartphones we hold in our hands now, and everything else that we’ve come to take for granted.

He reckons the algorithms that power Uber are rather straightforward. 

• It learns how to navigate from Google, for instance. Google Maps shows the driver how to navigate the shortest route from one place to another. But satellite data streaming into Google also shows traffic on that route. So, it may re-route it over a longer distance, but cut down on the time that may otherwise be wasted on traffic. It may cost me a little more money but will save time. The algorithm has to take a call. How does it do that?

• It is entirely possible though that I may be a prickly kind of person who insists on taking the same route every day. Anything else may upset me. I may have other traits built into me. For instance, I spend more time in cheaper restaurants and bargain hunting. Uber’s algorithms would have figured that out. For that matter, so would Facebook, Google, Amazon or even your bank and credit card company and every entity that uses big data and is working with AI researchers.

And how do they know of it? Because of the smartphone I carry around in my pocket all of the time with their apps on it and the cards I swipe every place I go. To use these apps to their fullest potential, I may have given them permission to track me all the time, even if I am not using their services. 

Think about it for a moment. Why does Uber want to collect data on where am I even when I am not using its services? Why does Uber urge me to allow it to track me all of the time? Long story short, if I am a scrooge, it will know what kind of surge pricing it can push me to. Allow me to reiterate—Uber isn’t telling and between Girish and me, we are only speculating. Are there people at the company who manage this deliberately? “No way,” says Girish. “Get this right. These are self-learning algorithms at work.”

But why just Uber? Girish tells me that based what I look for on my search engines, who my friends are on Facebook, what kind of tweets I post on Twitter, what am I browsing for on Amazon, the algorithms working at the back-ends of each of these entities are now intelligent enough to work backwards and build a complete profile of who I am, what is my propensity to use their products and even nail down the PIN code of where I live and what apartment block I reside in. 

Based on these factors, the AI can even predict whether or not I actually have the financial muscle in me to buy what is it I am looking at and when. All these companies know what kind of products will get my attention; what ads appeal to me; and what puts me off. They know me better than my wife or mother does. Trying to fob them off is not just futile, it is petulant.

Does it make them evil?

Both Dominigos and Girish vehemently disagree. Embrace them instead, they argue. Understand how they operate. Live with them. They are here. They are smart. And how smart are they? Some tinkering around with them suggest they are certainly smarter than I am. 

Allow me give you but a few instances of how to make the most of what the world is coming to.

• If you look for search for an item on a website a multiple number of times, the algorithms that power it figures you want it real bad. So, the next time around you look for it, it might just bump the price up. How do you get around it?

If I like something, but think it out of bounds for me, I run a “recipe” for it called IFTT—an acronym for “If This Then That”. I’ve alluded to it an earlier dispatch. Click here to begin looking for these recipes. To give you but one instance of how it works, I am a book and software junkie. Books I buy off Amazon and software I keep track of on Product Hunt. 

• I carry an iPhone with me with the IFTT widget loaded on it. It has instructions to notify me each time the price of a book drops to a point I think viable; and on Product Hunt to notify me when categories of software I am interested in are given away for junkies like me to try as promos. Each time it happens, my widget notifies me. Another app I purchased to keep track of my discretionary spending powers on any given day called DayCost Pro tells me whether I have the budget left for the month to acquire it right away or let it pass. If green-lit, I go for it. Else, I wince and let it go.

• Much poring over my genomic and lifestyle data later, I figured I don’t have much time left. My genetics are such that I am prone to rare neuro-degenerative disorders. I was hit badly in 2011 on the back of bad lifestyle choices. 

Soon after, I did some preliminary investigations and all pointers indicated I won’t live past 66. But heck, I’m not ready to go anyplace in a hurry. So, some lifestyle changes later, the data collected at various points using apps like Cardiograph, an app I latched onto from the App store when its price dropped to zero, indicates my heart rate has dropped from the 80s to the mid-70s.

I’ve failed to quit smoking yet. I can see the spikes there and what it’s doing on Cardiograph when I light up. But on the back of other changes, my lifespan needle has moved up to 68. At least that is what the data suggests, unless a truck runs me over. Heck, my younger girl is just four and can do with some more of me around, I suspect. My older girl is turning out to be bloody good at tae kwon do and I want be around to cheer for her.

• A few months of data collection later, I figured I work best in the mornings. So I’m up now usually by 4am. And contrary to most advice that the first thing to do is get a good workout first thing, it doesn’t work for me. My spreadsheets show I operate better after I have taken that half hour to brew my masala chai, and give myself a little time to focus and meditate using the pro version of Headspace. Following that, I need time with Evernote, my default note taking software of choice to review what happened and discuss with myself all of what needs to be done. If I get on social media, respond to emails or take phone calls during those hours, my productivity drops dramatically during the rest of the day. 

Data. It’s all there in the data. I can go on and on. But my limited point is this: If left to my brain, I’d have gone with pleasure. When outsourced to data and one more potent tool called artificial intelligence, both in a larger context of the world around me and my personal ecosystem, I’m having a lot of fun. 

This is not to suggest all is well. Things are evolving. Last week for instance, my emails to Sidin, my editor at Mint on Sunday, were auto-deleted. I sat pretty assuming I’d met my deadline until he checked whatever happened. Between the both of us, we don’t know what happened. Nobody knows. 

In hindsight, what I now know is the algorithms screwed up. But it would have figured based on the patterns in the correspondence between Sidin and me that on those dates it screwed up. Without the both of us knowing, it will make a quiet note of all that transpired.

But these algorithms will live, scientists will notice patterns, and they will build in systems that will actually “penalize” the algorithms for not having identified what they ought to have done. The poor creatures will actually feel “pain” in much the same way that you and I do. The lines between humans and machines are blurring. 

Should you and I be afraid? Yes. If we don’t keep learning, our days our numbered.

Should we be very afraid? No. As Girish told me, there are checks and balances in place built into the system so real humans like him can only see patterns and nothing is personally identifiable. All personal information is anonymized. These algorithms are making the world a better place to live in. “Most people are afraid because they don’t understand it,” says Girish.

By all accounts, that curse is also a blessing: May you live in interesting times.

This article was originally published in LiveMint & all rights vest with the HT Media. This piece may not be reproduced without permission from the editor