The AI Revolution-2: MIT Sloan's Rama Ramakrishnan on How AI Is Transforming Business Strategy

Rama, you advise many companies on AI issues, a question for you is what are some of the companies who are really ahead in this game and what you will recommend who are struggling and not decided what to do with AI? Yeah, yeah I think that's a great question. I think in my experience I see two kinds of companies very broadly speaking. One kind of company is very focused on looking at their existing processes, analyzing the processes using the notion of tasks that I referred to earlier and then figuring out what can we automate in this process to make it a lot more efficient, right? And that's a totally fine thing, there's nothing wrong with that. But I think there is another set of projects that people do as well, which brings me to the second kind of company, where these companies are saying yes, achieving efficiency improvements by automating existing processes using AI wherever possible is a good thing. But we also want to sort of have a different perspective, which is given that we have this amazing technology called AI, how can we reimagine our products? How can we reimagine our services? What new things can we do, right? What new things can we bring to market? How can we differentiate ourselves from the competition? And so on and so forth. And that essentially is an act of imagination or reimagination of your business. And that is frankly harder to do, right? Because in the first case, you can look at something existing in front of you and say, okay, automate that, automate this, automate that. But then if you have like a clean sheet of paper and say, okay, reimagine your business, like where do you start? It's very difficult. But I think the more progressive companies are realizing that it's not just enough to do just the productivity improvement using process automation, they have to simultaneously do this other thing as well. And I think because there's a real danger that an external entrant, maybe a startup or something, they reimagine your business better than you can. And then they come in from left field and maybe put you out of business. It could be a mortal threat. So you have to play very close attention to not just your internal operations and how to make them more efficient. You always have to be thinking about sort of this 3D chess game of what everybody else is doing, what new interns are coming in, how are they using AI to change the battlefield, the level, the playing field, and neutralize my advantages. And those are really important as well. So what I recommend to companies is that have a portfolio strategy. Make sure that you have some AI projects which are focused on productivity improvements, process improvements, process automation. At the same time, have these projects which you're sort of charging your talented executive teams to come up with new ideas, to differentiate existing products and services in new, interesting, powerful ways, which leverage perhaps your unique advantages in the industry, unique assets you might have, things you might own, things you might know, all those capabilities. I think by having this kind of balanced view, I think you're much more likely to survive and prosper in the sort of in the coming AI era. What about smaller businesses? Like this follows a similar, like you put them in these two categories A and B, or there is some should be different strategy? You know, I will confess that I have not advised a lot of small companies. So I don't have any great insight to offer. The only thing I will say is that I feel that a lot of small companies are really sort of like individuals in terms of how they can adopt AI and play with AI and explore with AI. Rama, talking about business, what about entrepreneurs? What are the skills they will need? What type of skills they need to hire? Yeah, that's a really interesting question. Obviously, we have seen this massive explosion of entrepreneurship in the AI space. But I think in general, everything in business is AI. I know, right? All the start-ups, everywhere you go, life sciences, manufacturing, healthcare, everywhere, media, everywhere is AI. Yeah, I agree. It's just nuts. But I think that if you think about entrepreneurship broadly, it has really kind of three phases to it. One phase is you got to figure out what people want. Number two, you got to build the thing that people want. Number three, you need to convince them to buy your thing and use it and you need to work with them to make them successful at using it, right? Very simplistically, we have three buckets of things you need to do as an entrepreneur. So I feel that building the thing, creating the thing, particularly if it's software, has become a lot easier now. A lot cheaper, a lot easier and so on. So that has become much better, much easier. But I think what that means is that the burden of being really good at figuring out what people want and figuring out how to convince them to use it and make them successful, the importance of those things have dramatically gone up, in my opinion. So because, let's face it, right? Previously, only, you know, qualified, credentialed, experienced entrepreneurs can do certain things. Perhaps they needed an engineering background to be able to write software and so on. Now, lots of people who may not have those kinds of backgrounds are nevertheless able to create prototypes and build the initial versions of products and so on. So what that means is that the space has gotten much more crowded now, right? For any particular piece of opportunity you might have or an idea you might have, there'll be like ten times more people or teams trying to go after the same opportunity because the barrier to entry has fallen down. So how do you differentiate yourself? Well, you differentiate yourself, first of all, by having a much better sense for what is worth building, right? Having a feel for the customer, having empathy for the customer, having a superior understanding of an industry and its evolution as it goes forward. All those things become much more important now. Similarly, being able to sell someone, to sell to someone, convince them to try you, particularly when that person is being bombarded by AI pitches constantly, right? How do you distinguish yourself, right? The human element has to come in somewhere that that person has to learn to sort of trust you, respect you, and so on. So you have to be able to really be good at sort of selling, empathic selling, then you have to work with them to make them successful, particularly sort of a B2B kind of situation. So I feel that entrepreneurs, it has always been important to figure out how to figure out what people want and then make, sell them, sell to them, make them successful, but I think the importance of those two skills has gone up tenfold, right? Because the middle thing, creating the thing has become much easier now. So that's what I would encourage entrepreneurs to do, is to sort of don't keep, take your eye off those two things and make sure that you're constantly getting better at those two things, because that's going to be the differentiator. Rama, let's also talk briefly about what are the risks with AI. We went through one situation here in my office, so we started to experiment with AI, you know, in editing some articles, and President Trump, he had taken over, it was about three, four months after he became the president, and one day, so what we started to do was, we'll write and have an article and thought, you know, for clarity and editing, let's run it by AI, ChatGPT, and it was ChatGPT, and we ran it, and we thought, this is great, saving us time, and we started to run, print that article. About three, four days later, we got a call from one of the readers, and the reader said, I have just noticed that in the last few days, whenever you are mentioning President Trump, you are calling former President Trump, and we went back and checked, and all the articles had a former president, and we're trying to figure it out, that why it is, so again, we went back to ChatGPT, and put that, you know, President, he's a, Trump is the current president, he said, no, he's a former president, and then after spending several hours, finally figured it out, we are using the wrong version of ChatGPT, which was not updated yet, and this was a huge thing, it was caught, thanks to our reader, and after that, we became very careful, you know, that, okay, you can use all these things, but you really need to check. So, what are some of the risks even an entrepreneur or anybody can face using this AI or ChatGPT? You know, just on that, what happened to you with this incident, I think that it's worth remembering that all these models, the large language model, all the chatbots, they have something called a training cutoff date, so they, you know, they take a whole bunch of data up to a certain cutoff date, and only the data is used to train the model, right, and so in some sense, the model out of the box doesn't know about anything that happened since that date, okay, so my guess is that the training cutoff date was earlier than when President Trump assumed office the second time. Yeah. Now, that said, there are ways to get around it, all these systems now are able to use search, web search to win answer to a question, so if you ask it to do something for you, you can actually insist that it do a web search before it answers the question, okay, then it will do the web search, and then it will use the results to give you hopefully a more sort of current... I am glad you mentioned this, and I think this is very smart, so if we do anything, we should keep asking the chat GPT or AI questions till we are satisfied fully or to get the better answers. Yeah, I think actually you have to understand that it uses a whole bunch of... that it has a notion of a training cutoff date, it doesn't know anything beyond that as to what happened in the world, but you can get around it by using search, it uses a whole bunch of other tools, so you can get a little smarter or more knowledgeable about how to use these systems, if you have like a clear understanding of how what's going on under the hood, and so in fact when we talk about how people can personally get ready, I think there are foundational things people need to know, and I'll talk about this when we come to that. Now going back to the question of risks, right, there is the personal risk that you are actually fed information that's just wrong, okay, like you might be given health advice that's wrong, right, so I mean you have to somehow find a way to fact-check it, like one of the things I commonly do is if I'm asking a question for which the answer is important to me, like consequential to me, then I will ask the same question to Claude, to JGPT and to Gemini and so on and see if their answers roughly match. So what has been your experience so far? You know so far I've been pretty happy with what I've seen, I feel that two years ago it would make a lot of mistakes, if I fast forward to say right now it hardly makes any mistakes, in fact I find I'm hard-pressed to find examples where it makes mistakes. Now so what that means is that it's getting better, clearly, but it also means that maybe when it does make a mistake it maybe it becomes much harder to catch, and so it's a bit of a delicate balance, you know, it's sort of like if you have like a Tesla for instance and you're using its self driving feature, maybe the last thousand miles it didn't observe driving and nothing bad happened, and so you sort of like get a little relaxed about this, you're not so vigilant anymore and you take your hand off the wheel or whatever and then you have a crash, right? So I think as these things get better and better, they do lull you into a sense of false security about how correct the answers are, so I think it is, there's no shortcut to this and it's kind of... So how will you rate those three? I think they're all comparable, I personally happen to use Cloud a lot, but again, Chargipity is really good, I just like Cloud this and I know Gemini is very good particularly for, you know, multi-modal images and videos and stuff like this, really good, so I use all of them, I'm not sort of, you know, ideological about any of them, I use all of them, but I think, you know, trying to, if something is important, ask many of them, fact-check it, cross-check it, and then maybe just do your own Google searches to collaborate, corroborate what it's telling you, when it's really important to you, ask it to do a Google search and give you citations, ask it to give you links that it's used to come up with the answer, and if you think the answer is interesting and you want to actually depend on it like a health care advice or something, take that link, go to that link yourself and read it and make sure it matches what the thing is telling you. So now, Google was the first one, we are all familiar with Google for everything searching, so how do you compare, like, if I ask a question in Google and then Chargipity to anybody else, is there any difference, or which is really the very authoritative source? That's actually a very profound philosophical question, because at one level, let's say you look at Google before the advent of these chatbots, right, you would ask a question, you would get a whole bunch of links, you go click on them, you read the page, and somehow we believed though what we read on the page was the truth, but the reality is that that's just a page created by somebody, that person may or may not have put down the truth, or maybe they think it's the truth, maybe they put something else, we just don't know, but because it came from a human, we somehow automatically believed, yeah, I can believe that, right, so I guess all I'm saying is that even with Google pre-chatbots, assuming that what you read on a web page, thanks to a Google search result, is basically an act of faith, you're assuming that the people who created the page knew what they were doing and put down the truth. Now, fast forward to today, these systems are trained on the internet, they are trained on all the things that are on the internet, the good, the bad, and the weird, right, everything is part of the training. Now, they can also do a Google search in response to your question. Now, the additional element of uncertainty here comes from the fact that they take all the Google search results, all the internets that they've been trained on, and then they proceed to give their own synthesis or answer from it, and in the process of synthesizing whatever they have seen and giving you an answer, they may introduce their own errors, right. That is just the facts of what's happening, so it's very difficult to know what's authoritative. My advice is that I, like a lot of other people, frankly, are using these things as an alternative to search. I don't want to be clicking on five different pages and trying to figure this out, it's just so much easier, but if the final answer that you're looking at, if it's something you're going to act on, as opposed to just some mild intellectual interest, if you're actually going to act on it, then, and if it's important to you, then do your own research to fact-check all these things, right, and what I mean by research, do your own Google search, meaning you're once again relying on the fact that maybe what Google finds and gives you on those web pages is kind of the truth. One thing I always do, by the way, just as a quick addendum to this is, if it's healthcare related, I'll always ask these things to search reliable, reputable academic publications and give those citations back to me.