With its ability to create new and unique content ranging from text, imagery, videos, music content and more, generative AI is revolutionizing businesses across industries ranging from manufacturing, healthcare, IT, arts and design, advertising, and marketing. Recently, some of the top experts in the field had come together to facilitate our understanding of business applications, limitations, regulatory considerations and use cases of AI tools.
“Unleashing a wave of innovation…”
Mr Gopi Kallayil, Google, USA
Every one of you has used AI today in a direct or indirect way, without realizing it. If you use Gmail, you’ve already used AI, because AI is being used to keep your Gmail saved from spam. If you shopped on a site like Amazon, you probably got some recommendations from the service, powered by AI. AI has already been embedded across many of the products and services you use. Like electricity, you may not realize it.
However, the field of generative AI has ignited the imagination of the world, primarily because there are some AI-based services that are now available to consumers. Bard or ChatGPT has allowed us to ask questions and it seems to converse with us like a human being does, with fully formed sentences and excellent answers. You can write code. There are other tools where you just give a text prompt and you can get a complete picture generated.
Three Major Platform Shifts
We have been privileged to witness three big technology platform shifts that have fundamentally altered human civilization. They have changed the way we interact with each other, the way we consume information and the way we conduct commerce. First was in the 90s when computers that were sitting on desktop suddenly could get connected to each other through the amazing thing called the internet. The browser that was created in 2004 allowed us to access all the information very easily. The second shift happened starting 2006 when mobile phones, especially smartphones started appearing in our hands and the world shifted towards mobile computing. Now the third one is taking place, which is AI, work on which has been going on for the last several years. Since last year, we could see it, feel it and touch it. This will create thousands of companies, entrepreneurs and new ideas. Problems will get solved.
Artificial Intelligence is any computer system that mimics or amplifies our incredible human ability called natural intelligence. AI can play board games and translate languages (like Google Translate). We can have very, very complex systems like ChatGPT or Bard that engage in conversation. Even more sophisticated systems are now coming up in diagnosing health conditions.
Will AI replace us? No, it will amplify our efforts. It’ll make us smarter and be able to move more quickly and solve problems on a scale that we could not. The analogy I want you to keep in mind is how humans did movement and motion. When we were young infants, we crawled around the living room of our parents. When we got a little older, we learned to walk. Then we learned to run. Then our parents put us on a bicycle and using our own muscle power, we could go farther and faster. The bicycle amplified our movement in motion. Similarly, AI tools we’re building, will amplify our cognitive ability.
Four Buckets of AI
What can these systems do? What are they capable of doing now? The future falls into four buckets.
- They can synthesize vast amounts of data.
- They can make predictions. They can look at patterns, come up with logic and tell us lots of things that will happen in the future, like the movement of a stock index or weather or copper production.
- They can create new content like how humans create using our right brain capabilities.
- They show capabilities of human conversation. When we converse, we listen to the words, look at the gestures and the facial expressions and then respond to them. The systems are beginning to replicate some of the conversation capability. However, they only do it with words. They can’t read emotions, they can’t see and understand our gestures. But in future, those capabilities may develop, looking at the way generative AI is going.
Let me put those into two categories. In the brain, we have the right side that does creative thinking and the left side does analytical, logical thinking. Similarly in AI, there’s an entire category of systems that seem to mirror the right side and that is generative AI. One of them is conversational AI, a system like Bard. It’s built on large language models. There are other systems that mirror the left side of the brain and they are analytical and predictive AI. For example, there is a feature in Google Maps called eco-friendly routing. It can suggest the most fuel efficient route, for different types of vehicles. We tested this in Canada and now it has been rolled out in US and eight countries in Europe.
Images, Music and Movies
Using Google Imagen, we can get images from text. If I want the picture of a chrome plated duck arguing with an angry turtle with a golden beak, I can get it in 20 seconds. Using Music ML, text to music creation is possible. Experiments are also going on to create movies from text, using a combination of Imagen and Phenaki. Now the quality is not so great. These are early days but those are solvable problems. In the future, this will get scaled up, so we can create lots of audio and video contents.
Google has been at AI for a very long time. In fact, Sundar Pichai, CEO of Google declared as early as 2016 that we are an AI-first company. Every few years, roughly a decade or so, Google shifts its focus to a new platform. When we started in 1998, we were a ‘search first’ company. Around 2006-2007, when mobile phones became widely adopted, we shifted our focus as a mobile-first company and built our products around mobile. From 2016, we saw where the world would be headed in AI and became an AI-first company.
At Google, we want AI to help three audiences: consumers; communities and companies. We want all our consumer products that you use work better. There are 15 products with at least half a billion people using them—Search, Gmail, YouTube, Google Lens, Google Translate, etc. Second, we are looking at big social problems and trying to solve them. With Alphafold, we are trying to solve the protein problem in the world. Protein is essential for animals, plants and humans. There are 200 million proteins in the world and only 4% of them are well understood. We used AI and over a four year period, documented every single protein and it’s available in our open database called alphafold. A million biologists, scientists and pharmaceutical technicians are using it to conduct the research. Third, we’re helping companies grow by improving the marketing significantly by AI. We help companies to become more efficient by using AI and manage contact centres 24 X 7.
Working on AI, there is so much of promise. But there are also a lot of risks and we understand this. We want to take a bold approach with AI, but also a very responsible approach. We work under seven core principles to guide all our work in AI. We’re very open and transparent about it.
Unleashing a wave of innovation, we will solve many problems that exist in the human condition. We will increase quality of life and also lift a lot of people in their economic level. This will be to the benefit of all, if we manage it all very, very carefully. So, as Google founder Larry Page would ask us, “Are you uncomfortably excited?”
“Generative AI will not replace you, however, a person using it will.”
Mr. Jaspreet Bindra, Tech Whisperer Ltd, UK.
The whole field of generative AI, in some sense, can be reduced to five fundamental facts, which I believe are the most significant today. AI as a technology has been there for a very long time, while generative AI is something which we know of today. The word AI was actually coined by John McCarthy, around the time India got independence. It’s more than 70 years old. AI in many ways is a philosophy rather than a technology. The technologies are the ones that sit under this AI umbrella—machine learning deep learning, neural networks, algorithms and so on.
The 5 AI Facts
AI has come out from the background and probably for the first time into the hands of ordinary people like us. That is why we say that this is something which is amazingly powerful. That’s the reason why 100 million users have ChatGPT in two months. The next fastest app was TikTok, which took about nine months. Facebook took four and a half years. There’s a wonderful tweet by someone which says that in technology, there are decades where nothing happens. And then there are weeks where decades happen.
1. This is Generative AI. Use it to generate, don’t use it to search.
Someone said that Google is a search engine and ChatGPT is a thought engine. Think of it as something which is creative. Generative AI is built on large language models. These models are trained on massive amounts of language, text conversations, paragraphs, etc. ChatGPT3 is two generations behind. We’ve had ChatGPT 3.5. After that, we have ChatGPT 4. When ChatGPT 3 was trained, all of Wikipedia was just 0.6% of its training data. So you can imagine the sheer amount of words and language that has been fed into it to train it.
A chat GPT or Bard pretty much makes up stuff as it goes along. It sometimes hallucinates. It’s not optimized for facts or truth. It’s actually more optimized for believability or plausibility. It’s actually what we humans do. That’s why we feel a little bit of kinship with these new, human like technologies.
2. Generative AI will not replace you, however, a person using it will.
Kent Beck is a legendary Software Engineer. He wrote that he was reluctant to try ChatGPT. In his tweet, he says, “Today I thought about that relevance.
The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x. I need to recalibrate.” Most of the white collar jobs can be replaced by AI. A report by Goldman Sachs says that AI will affect about 300 million jobs across major economies. About 7% of jobs will disappear. If you’re a low end coder, a journalist summarising your stuff and calling it news, then your jobs will go.
30% of jobs are safe and these are menial jobs. If you’re a farmer or a construction worker or a nurse who’s working on compassion, then nothing is going to happen to your jobs. A bunch of new jobs will be created. We still don’t know which ones they will be. Every technology has created more jobs than jobs lost. Net-net, there are always more jobs created. It’s up to us whether we use generative AI to enhance our jobs or become lazy and let someone else who uses it to take that job away from us. Generative AI hits the cognitive jobs or the higher end jobs first. That’s why, in India, we may have a minimum effect on jobs, while the developed economies will have the maximum impact from generative AI. It will vary across industry, it will vary across function. But the key thing is that it will not replace you. Someone using it could replace you. So don’t fear AI. You have to fear a human being using AI.
3. The horizontals are interesting but the verticals are useful
We see many horizontals in Generative AI like ChatGPT, Bard, and LLaMA. These are interesting and they do a lot of things. But the really useful stuff will be the verticals which are industry and domain specific.
4. Generative AI will be as transformative as were search, internet, email or phone
Every 10 years or so, a fundamental platform shift happens. We are at the time of a platform shift. It’s going to reshape technology, industries and big tech. There’s a new tech company called Nvidia. It’s a trillion dollar company, which emerged out of nowhere on the back of generative AI. We’re going to see some of these newer things reshaping the world.
5.General Artificial Intelligence is not equal to Artificial General Intelligence
We talk about super intelligence where technology becomes as intelligent as or more intelligent than human beings. There’s a lot of conversation happening in this. Elon Musk, Sam Altman, Geoffrey Hinton and others are talking about it. What we have today is not artificial general intelligence, at least for now. In AI, we have something called the alignment problem which is about how you can embed in AI values that are aligned with human beings.
My view is that within human beings themselves, the values are not aligned. We don’t agree on democracy. We don’t agree on what truth means. We don’t even agree on what facts mean. So the danger is not with a super intelligent AI, but with us not agreeing with each other. There are lots of ethical issues around AI. We’re talking about regulation, plagiarism, bias and fake news which are environmentally very destructive. For the first time ever, the creators of this Generative AI industry are the ones who stand up and say, “Regulate us.” We will see a lot of action in this space.
“This technology works on probabilities.”
Mr Sankaranarayan ‘Shanky’ Viswanathan, TCS
At TCS, I lead the generative AI initiative alongside a lot of emergent tech initiatives. One of the biggest questions our customers ask is: What do we do with AI? It’s a phenomenally good tool to give me a route from Delhi to Chennai, to tell me all the restaurants where I can eat and the temples I can go to. A lawyer was bold enough in the US to create an entire legal submission based on ChatCPT, but only to be thrown out, because the judge realized that some of the case references which ChatGPT had given did not exist at all. It cooked up an airline that did not exist.
This is a fascinatingly good technology and it is ripe. But the biggest question our customers in industries are asking is: What do I do with it? The natural inference is that it is good to converse. So the first immediate thing that customers think is it will replace contact centers. Generative AI is a good tool in areas where you can verify the answers and you are comfortable with the response. How can you control the response when you expose it into a wider ecosystem? These are the challenges that enterprises have.
Need for Training
This technology works on probabilities. You ask technology a question from what it has been trained to answer. If it is unable to find an answer, it keeps trying to build a language as a response, which is closest to the answer that you have. The probability might just be 17%. It doesn’t care, because that is the best response it could give. So, let us not worry that this machine will lie at any point in time, if you are able to train this model with sufficient data.
The second big challenge customers have is bias. We were operationalizing a mortgage with an insurance provider in the US. Ironically, we realised that a lot of insurance recommendations were more attuned to working class white men. Immediately there might be a regulatory body pointing out the inherent bias. The fact is that it is not an inherent bias. We need to understand and appreciate that the general corpus, the Wikipedias of the world, will have a lot more articles on white male than a tall, dark, handsome Tamilian sitting in Chennai. We must feed information and data to ensure that this bias is suitably managed and controlled.
Third, you have to be extremely prescriptive. You must ensure legitimacy when you prescribe. Again, it is all about probabilities. There are methods to ensure that the large language model starts to zero in on the right answer, rather than give any of the vague answers. There are a lot of training methods for that. These technology concerns which exist, can be mediated through processes, procedures and adjacent technology. So making it viable within your organizations is a viable prospect. We must understand, appreciate and embrace the beauty of the technology rather than anything else.
Need for Guardrails
Generative AI’s response has a bearing on the way you interrogate or interact with it. It is not feasible for you to go and educate everybody in your organization to do things in the right way. So sufficient guardrails must be put within your enterprise or organization to ensure that such deviations do not cause serious harm.
We must use it for creative purpose. Most of our customers right now are using it for its conversational capability because it converses very well. The second big usage that we see is anything which entails knowledge search. For example, in your HR department, a lot of people ask you about the PF or LTA conditions. If you give all the information to AI, you don’t need a person answering these questions. There is a phenomenal opportunity to leverage anything related to document extraction and summarization. It does it in a phenomenally good, secure and predictable way. No challenges at all. But that is just the tip of the iceberg.
The biggest opportunity generative AI is going to unravel is in data analytics. An investment advisor looks at a huge amount of charts and then gives his recommendations. A shop floor mechanic who looks at a huge turbine problem comes and looks at all the preconditions and parameters and says what you have to do to fix that. What we have done so far is, we have automated the intelligence being presented to people, so that the right action can be taken in a phenomenally good way. The most important thing is to understand that the space between our ears is what defines how actions are being taken. That is what I call tacit knowledge.
While plumbing is the same, you still want the best plumber to come to your house. Because you know that experience and tacit knowledge exist. To fix your turbine, you want the person who’s seen this problem multiple times to come and solve the problem. The reason is, this inherent knowledge is never captured in a company. We must be able to formalize, define, capture and document knowledge. More often than not, our industries are still knowledge driven. They are run by experienced people who know what calculated action is to be taken.
Knowledge as a Capability
What if we are able to disrupt this knowledge value chain? Generative AI or large language models are not about predicting the next best word like a WhatsApp message. Language is the best form of documenting knowledge. Language is the form of communicating knowledge. Generative AI, when you feed data, insights, intelligence and more importantly, the actions which have been taken in the past, has a unique ability to discern and distill knowledge and save it. It will be able to cross the knowledge gap and then discern and determine the right action to be taken.
Just visualize this. You capture knowledge across your workforce, create a knowledge corpus as a capability and make it available for anybody in the world in your type of organization. That to us is the biggest difference Generative AI will make. For example, a shop floor person goes and stands in front of the turbine and tries to figure out the problem. It is not about just his knowledge. It is about the collective knowledge of every shop floor mechanic, from what happened in the past and it is available to them in a digital fashion.
So the biggest shift that generative AI will create is that knowledge as a capability will be made available to every industry in a similar category. Don’t just focus on collaborative automation or knowledge extraction. Generative AI provides a unique opportunity for you to reimagine your enterprise. It is not about automation or productivity improvement. You can completely reimagine the way your shop floor will work using multiple technologies together.
The Convergence of Three
There are three very important technologies which converge to make this a reality. The first is compute. The compute capability of chips will become so pervasive that they will be existing in a watch, in a cell phone, and in an IoT device. Each of these chips can potentially run one large language model on its own and deliver specific outcomes.
The second big technology is 5G. 5G is not about downloading YouTube videos fast. It’s about having have pervasive systems of connectivity, to bring intelligence to life and collaborate with any of the systems of knowledge.
The third big disruption which is happening already is the convergence of cyber and physical systems. What we mean by cyber physical systems is that most of the robots are built for robots to exist on their own. Humans will collaborate with technology and systems and robots.
Last but not the least, digital twins will give real time avatars of everything that is happening around you. That gives the context; Generative AI gives the knowledge. Cyber physical systems ensure people and machine work together in a completely different way against the ubiquitous connectivity enabled by 5G. We must thus look at multiple, concurrent technologies together to reimagine our industries. Generative AI is the biggest piece of the jigsaw puzzle, primarily because it is not about automation. It is about a knowledge driven enterprise. Knowledge is going to be enabled as a capability within your respective industries. These are, therefore, absolutely exciting times.