Spotlight: Gen AI in Automotive and Healthcare Industry and LLMs

Transcript

This transcript has been edited and refined for clarity.

Smruti Gandhi: Welcome to yet another episode of Spotlight, where we delve into all things related to technology. This is Smruti Gandhi, executive editor and event content lead for Information Security Media Group. Today, I am joined with senior editors from CIO.inc, Rahul Neel Mani, vice president editorial and community engagement; Shipra Malhotra, managing editor; and Brian Pereira, senior director, editorial. Rahul, Shipra and Brian, thank you always for joining me on Spotlight. Everybody has been talking about gen AI, and gen AI is a concept which is evolving rapidly and extremely exciting. The importance of gen AI lies in its potential to revolutionize various aspects of our lives, and drive innovation across multiple industries. Today, we'll talk all about the articles that you all have written recently on gen AI. Rahul, since you're the senior most, I will start with you. You have been talking about gen AI to the CIOs globally for a long time now. Recently, you also had a conversation with the head of digital of one of India's largest automotive companies. How are they using gen AI? Could you throw some light on that?

Rahul Neel Mani: Yes, Smruti, thank you for asking this question. Being the senior most or the junior most doesn't matter; the buzz in the industry starts and ends with AI and generative AI. Before I give a direct answer to your question, let me set the context of, or set the tone for, this particular Spotlight, which is focusing on AI. AI is the third most searched keyword today on the web, which is only after information technology and technology. If we talk about its subset, which is generative AI, the market seems to be buzzing. The market size, which was just $29 billion last year, will reach to about $670 billion in the next 6 to 7 years, by 2030, which is going to exhibit the growth of about 48%. I may not fully agree with what Sundar Pichai said in one of his statements earlier this year, that AI is one of the most important things humanity is working on today, it is more profound than electricity or fire. But, it is no less, that's what my comment is. I recently read a survey which Gartner did, with about 1,400 executive leaders across the world, on the usage and adoption of generative AI. Out of the total 1,400, 45% reported that they are in piloting mode with generative AI and 10% have put generative AI solutions in production, which is a significant increase from a similar poll conducted about 6 to 7 months ago in the timeframe of March to April 2023; 55% of organizations reportedly increased investment in generative AI since its surge into the public domain 10 months ago; about 75% respondents believe that the benefit of generative AI outweighs its risks, which means that enterprises shall stop being paranoid about cybersecurity or other risk concerns, and should focus on use cases. That's about the generic part of my answer. Let me now get on to the direct answer to your question. Automotive sector, in general, has been an early adopter of AI and generative AI, both from autonomous driving and predictive vehicle maintenance to areas like enhanced safety features and personalized driving experiences. AI has come up to speed to help in how vehicles are manufactured, operated and utilized by the customers. AI has also been applied to enhance the overall buying experience. AI now offers personalized experiences and recommendations. There are virtual showrooms based on AI, and there are areas that automotive manufacturers are working on to streamline customer interactions and further elevating the overall customer journey. I recently spoke to Bhuwan Lodha, who is the senior vice president and head of digital at M&M, which is a large conglomerate and has a large automotive division. It has taken a lead in deploying generative AI in its manufacturing operations. A typical use case that he mentioned was complete automation of robots and heavy machinery maintenance on the factory shop floor. Without getting too much into detail, by doing so, they have significantly reduced the downtime and bolstered the morale of the shop floor workers as they can now reduce the time taken to maintain a robot or make the robot functional once again if there is any problem, and remove the friction without seeking any external help. Additionally, they are harnessing generative AI to train its customer chatbots. This has helped in saving time for agents and delivering a very enhanced experience to their customers. The number of these use will undoubtedly go up. However, Bhuwan, while talking to me, gave a word of caution; he said that while the capabilities of generative AI are enormous, it is best used in a co-pilot mode as of now. The technology is yet to attain its full maturity for independent operations and, therefore, it is best used under human guidance. That's the long answer to your short question.

Gandhi: Thank you for setting the context for gen AI. Indeed, gen AI represents a transformative force with a potential to reshape the industries. If you look at the other side of it, it does enhance the customer experience, but what is the return of investment for their AI investment, from M&M?

Neel Mani: I agree that any technology, when it is deployed in an enterprise or a business environment, is needed to be evaluated on its return on investment. However, in case of generative AI, since the technology is still maturing and far from being on the top of maturity curve, we need to look at it in a little different way. While the early results are encouraging and CIOs and technology leaders are making most of this new tech, both understanding and deployment, it is a little premature to evaluate or calculate the absolute ROI. However, true is the fact that when we look at early implementations of generative AI, which is providing assistance to quick remedies to mundane tasks, generating software codes and automating the routine operations, the ROI is clearly visible. It needs to be considered in a slightly prolonged period of deployment, that how the ROIs are calculated. I would say that ROI depends upon long-term benefits derived by the cost of implementation versus the man-hours saved by an organization; the accuracy and the usability will also play a very vital role in calculating the ROI. ROI, to sum up, is not only quantitative but also qualitative. That's what my answer is.

Gandhi: Good perspective from you; thank you, Rahul. Shifting the focus from automotive to medical, Shipra, I want to ask you that you looked at the healthcare industry for gen AI, and have written about a U.K.-based firm for precision medicine. How is that company harnessing the power of gen AI? Could you provide some insight on that?

Shipra Malhotra: Thanks, Smruti. The company in question is a U.K.-based company called Exscientia, which operates in the areas of drug discovery, design and development, and it is a pioneer in precision medicine. The company is using generative AI to accelerate the identification and development of candidate molecules. The company creates these candidate molecules for both its internal pipeline of wholly- or partially-owned product candidates as well as collaborations with other leading biopharmaceutical companies. The company has already discovered a few molecules using its generative AI platform, of which six have already progressed into the clinical setting, which means that they have already shown some success to progress to the clinical setting. The sixth and the most recent generative AI-designed molecule to enter phase I trials was in May this year. It was created under the company's collaboration with Sumitomo Pharma. Why is all this important? We need to first understand the drug discovery process, and it has multiple challenges that include higher cost, very slow pace and high degree of uncertainty in the success of clinical trials. Exscientia is using generative AI to address all these three challenges. The AI designed molecules not only make the drug discovery process faster; in one of the cases, it brought down the timeline from what would have been 4 to 5 years in a traditional process to eight months. Besides, it increases the probability of clinical success because you have already shortlisted those candidate molecules that are most likely to be successful. There are three use cases of their AI platform in the drug discovery process. One use case is that they screen large libraries of potential drug candidates to identify those that are most likely to bind to a target protein; therefore, narrowing down the search for promising drug candidates, which is what I mentioned earlier, bringing down from millions of potential candidates to four to five or few, which are most likely to be successful. Design new drug molecules, which are tailored to interact with a target protein, improving the potency and selectivity of drug candidates, and predicting the properties of drug candidates to assess their potential safety and efficacy. I want to wrap it up with a quote from Exscientia's founder and CEO, Andrew Hopkins, who said that their AI platform allows the company to move rapidly from idea generation to new drug molecules ready for investigational new drug and clinical development. He also believes that this is the future of drug discovery and designing new drugs and new medicines.

Gandhi: I'm sure these clinical trials will bring down the timeframe required and help the patients globally, which is very essential now. You have also covered some startups in this industry who have used gen AI, apart from this company. Can you talk about those as well?

Malhotra: Why these startups play a key role, besides the other big names in the biotech and pharma space, is because these startups can innovate faster, they can build cutting-edge use cases around generative AI because they have the comfort of disruptive innovation with the ability to introduce alternate schools of thought and they don't have the legacy thought process. They can deliver leaner and agile solution discovery, and they can move and change faster than the established incumbents, which can be leveraged to scale faster by the healthcare companies. These startups are mostly operating in the areas of drug discovery, which we spoke about, and then precision medicine, pathology, radiology and diagnostics. In pathology and radiology, we have companies like PathAI, which is analyzing digital pathology images to assist the pathologists in diagnosing diseases, such as cancer, with higher accuracy and efficiency, and there are many other startups in this area. For instance, in precision medicine, we have a company called Paige, which is advancing precision medicine by detecting and classifying cancerous cells for more precise diagnosis and personalizing the treatment plans. Last is the drug discovery process, which I already spoke about, about Exscientia. There are other companies like Insilico Medicine, which is developing new drugs and treatments for cancer and other diseases. It identifies new drug candidates and designs new molecules specifically tailored to individual patients, making them more effective. These are just a few of the startups that I have named, but there are many others, hundreds, or rather thousands of these startups that are looking at very innovative use cases across the entire spectrum of healthcare.

Gandhi: We heard, from Rahul, about the automotive industry for customer experience and from Shipra about healthcare where they are trying to reduce the downtime for the patients, which is very essential because healthcare is such a sector where the customers are most scared and want all the details earlier. Moving from both the industries to something called LLMs. LLMs contribute to the generative capabilities of AI. They play a very significant role in enabling machines to generate human-like text and create content, which is one aspect of the broader gen AI field. Brian, since ChatGPT and open-source LLMs are very generic, what is the approach that organizations are taking to develop LLMs?

Brian Pereira: Thank you, Smruti. Before I get into that, I just would like to compliment Rahul and Shipra for all their interesting use cases. Beginning with Shipra, I read recently that there are as many as 50 AI startups in India, half of them in Bangalore, and a chunk of them in the Mumbai-Pune belt. There's a lot of AI development work happening in India itself. Rahul says that there are concerns about adopting AI; I had a panel with some panelists from Indonesia last month, and their prime concerns were the accuracy of the output. They would like to focus on their datasets, for now, and are taking on low hanging fruits or some smaller steps and easier projects to have a proof of concept because we are still in the earlier stages of AI adoption. I might have another interesting use case coming next week. I was speaking with HARMAN a few days ago, and they have developed something called HealthGPT. They have taken a generic AI model and customized it because they thought that would be the fastest way to develop a model. This has already helped researchers in clinical research, and finding cures for breast cancer. Look out for that story, which is coming next week, and that's an interesting one. Smruti, you asked me about LLMs. I recently wrote a story about proprietary versus open-source LLMs. We all know that, in the enterprise world, 90% of enterprise software code base is open source. But, when it comes to LLMs, it's quite the opposite because organizations are hesitant to adopt open-source LLMs for a number of reasons. Yet, there are a lot of organizations which are contemplating the use of proprietary LLMs. Let's take the pros and cons of each. The open-source LLMs, like ChatGPT, Falcon 180B and Meta's Llama 2, are very broad based, and while comprehensive, they're not specific to a particular industry. There are licensing issues, and a company would want to do its due diligence when using an open-source LLM. However, proprietary LLMs give you a lot of control. But, the downside is they take as much as six months to develop or longer to develop it from scratch, and then you have to spend time on training the model with your datasets. It's also very expensive to develop a proprietary LLM. However, a large bank or a healthcare company might want to consider a proprietary LLM because it can be very specific. It's trained on its company data, and, therefore, the output could be more accurate. The other thing is that proprietary LLMs would give you more flexibility and a lot more control. Then you can also build in security and privacy by design and make a more secure model. A recent CIO survey by Recognize shows that 42% of organizations are contemplating to develop their own proprietary LLM. However, what we see is that we are now at the exploratory stage, and everyone's experimenting. As I said earlier, it's largely POCs and MVP. That's where we are on proprietary versus open-source LLMs.

Gandhi: Thanks, Brian. I did not touch about the security aspect in any of the questions. I want to ask you, how can organizations address the security and privacy concerns associated with the use of these large language models, as they may pose risks such as misinformation, data leaks or invasion of privacy?

Pereira: As I mentioned earlier, when you go the proprietary route, you are following something called DevSecOps, which means right from the beginning, you are thinking about security and privacy. They talk about security by design, privacy by design, and you can build that in right upfront when you're building your proprietary model. However, when you're using an open source, there are issues with open-source models because they are very generic, and they are scraping the internet and websites for data. Then there are no privacy checks and no security checks. There's also the issue of plagiarism because they are taking data from many websites without cross-checking or without attribution or taking permission. When you have a proprietary LLM, you have more control over it. You're thinking about this right from the beginning, and you have security teams working closely with the operational teams and thinking about security right through the process, and this testing going on right through. I think the approach to go would be DevSecOps, when you develop your own LLM, and that can ensure better security and better privacy.

Gandhi: Thank you, Brian. We spoke about different industries where gen AI is used. We spoke about LLMs. In this field that you're working, what are the AI tools you all employ to support your editorial work? In what ways do they assist you?

Pereira: Right now we are all dabbling around with ChatGPT 3.5, and I'm thinking about taking the version four as well. I use it for creating summaries of my stories. I use it for asking questions for interviews when preparing for interviews. It’s quite helpful! I don't use it for writing my story entirely because I prefer to do that myself. I'm unsure about the sources from where it's picking up the information, so I'm very wary about plagiarism. However, I just use it as a co-pilot, as you mentioned earlier. The other tools that I would like to experiment with is the SEO AI tools, like Surfer SEO, SEO.ai. I also want to look at the other ones, like Bard, Jasper or Chatsonic, which I will be experimenting with this month. If it helps me to write a better story, with better insights, better questions for interviews, why not?

Gandhi: Of course, a virtual assistant is always better. Shipra, which AI tools do you use to assist your work?

Malhotra: I mostly used Bard AI and ChatGPT for assisting with my editorial work. One of the areas where I've seen usefulness is when we have an interview lined up. Looking at some story ideas, looking at questions can be helpful. Besides the questions that we already have in mind, it can throw up story ideas and questions, which probably we did not touch upon. Another area where I've used ChatGPT is looking for story ideas, new innovative ways or aspects to cover on a particular technology or on a particular subject. I'll give you an example of the story that I spoke about, which is Exscientia AI. I started with asking ChatGPT about what are the different use cases of generative AI in healthcare, and it threw up a whole lot of use cases, and I picked up drug discovery. Then, I prompted which are the startups and companies that are currently using generative AI in the drug discovery process. I got several names out of which Exscientia was one of them. If I had gone to Google Search, and prompted the same questions, it would have thrown me tons of search results, and I would have gone through each one of them, which would have taken me at least 1 to 2 hours, if not more, which the ChatGPT platform did in 1 minute. That kind of helps bring down the time. But, of course, post the results that I got from ChatGPT, I need to verify, cross-check those results. But, it brings down at least 1 to 2 hours of time, which would have gone into the basic research and gone into going through all the search results that Google would have thrown up.

Gandhi: ChatGPT seems to be the favorite virtual assistant. Rahul, what about you?

Neel Mani: If all of us in this frame may have heard of a tool called Grammarly, that tool exists since 2009. We all have been inadvertently using this for a very long time without realizing that this is an AI assistant. They have now come up with a tool called GrammarlyGO, which is based on generative AI. I have just started experimenting with that. That's number one. Second tool, which I used in my personal capacity, not so much for work, or rather not at all for work, was Midjourney. It's a great tool for creating and generating images. This gives you images based on text prompts. It's a fascinating tool. I used it for multiple purposes, both for creating non-technology as well as technology images. As far as work is concerned, I have been very conservatively using the free version of ChatGPT and Bing and Microsoft Edge Chat, which is powered by Open AI. Right from seeking assistance on writing blurbs to writing emails to creating survey forms, using it sporadically for these purposes. The results are quite encouraging. I would say they're cool; they're satisfactory. Of course, as suggested by the experts, I also don't use it blindly, it's a generating tool. We have to be very cautious about it. But, for faster responses and quicker turnaround, these are absolutely great tools to use.

Gandhi: Great! Thank you so much, Rahul, Shipra and Brian, It was a pleasure to catch up with all of you on Spotlight today. Thank you for sharing your views here.

Neel Mani: Thanks, Smruti. It has been a very good conversation on generative AI. Generally, we end up spending more time by asking questions to our experts, but this time we are answering questions.

Gandhi: Thank you so much. This is Smruti Gandhi, signing off. Until next time.