Digital Transformation Strategy: Become an AI Company - with Harvard Business School (CXOTalk #768)

Digital Transformation Strategy: Become an AI Company - with Harvard Business School (CXOTalk #768)

Show Video

It's a total rewiring of your organization.  That's what's ahead for most people.  Guess what. It's damn difficult. [Laughter] [Laughter]  And it's something that the C-suite –  this is CXOTalk – cannot outsource to   other people. They have to become experts  in what's going on and drive the change.  That's Karim Lakhani from  the Harvard Business School.  My work is at the intersection of  technology, innovation, and business,   how digital technologies are transforming  businesses and changing business models and   operating models. On the research side of things,  I run a lab called The Laboratory for Innovation  

Science where I'm the founder and co-director. We've done a lot of work on crowdsourcing:   crowdsourcing for innovation, crowdsourcing  for algorithms. That's what got me into   this AI space more than a decade ago. We  had partners like NASA, Harvard Medical   School, The Broad Institute, and so forth. In light of that, I have (over the last year)   launched a new institute at Harvard called  The Digital, Data, and Design Institute (D^3)   because we think that these three technologies  (digitization and digital data science and also   the design of new business and operating  models) are having an exponential effect.   We have launched with more than 30 faculty members  at Harvard Business School (and some colleagues at   the engineering school) with 12 different labs.  We're trying to work closely with companies to  

solve problems and do great research as well. Your focus is on how companies can compete,   so what is going on? What's unique about  our present time with AI that caused you   to need to look at this problem? The book title is Competing in   the Age of AI. We're not even saying  competing with AI but in the age of AI.  In many ways, the consumer economy (with  our mobile phones) has already put the vast   majority of humanity in the age of AI. If you  think about how you navigate your email, how   you navigate your music selection, your viewing  habits, your reading habits, your directions,   all of that is already immersed through AI. Increasingly now, the tech giants have sort of   brought that to us. But that whole world is now  shifting into the rest of the economy as well. 

The book is really about we're  not turning back with less data,   less digital, less algorithms. We're going to  be doing more and more of it. How does that   shape what companies do, how companies compete  above and beyond you being an AI native firm?  The book really sort of starts with the fact  that the technology is going to be an enabling   tool and it's no longer a thing which sits on  its own but is woven throughout the fabric of   the organization. That means that your operating  model and your business model are going to change.   That's what the book really tries to go after. How is this different from business   as we've known it historically? A modern corporation really is maybe about   120 years old. If you think about the history  of humanity, most of the time we've been sort   of agrarian, small, little shops, and so forth. The modern corporation basically got set up 120,   130 years ago. If there were sort of  seminal views of what happened in America,  

you look at Alfred Sloan setting up General  Motors as a multidivisional company,   Thomas Edison setting up General Electric  as a multidivisional company, as models by   which we have always run our organizations. The idea here was that you focused. You went   after one thing after the other. You had visions  set up. You had functional silos set up, and you   were able to go and serve your customer needs. That model has done tremendous things.   Our built environment, our built  organization has been set up this way. 

Starting with the advent of computation  and computers (with IBM and Microsoft),   that edifice started to change where we thought  that basically now what matters is not just the   ways in which we organized from the top-down  but the ways in which information flows across   an organization. This information flow view of the  world really first started with the tech industry   and the software industry and the emergence of,  let's say, Windows as a computational tool that   allowed lots of people that power to analyze  data and do things, and the spreadsheet as   the way in which you would get work done. But everything was still very much in the   model of divisional structures and functional  structures set up. Every time you had to share  

data, you would be sharing large files, and  there would not be things coming together.  But what we saw emerge (even in the Microsoft  era) was a new type of company was emerging.   This company was set up as an ecosystem. Microsoft won the PC battle because they   figured out how to build an ecosystem where they  had lots of complementors and lots of consumers,   and they were in the middle of it. This  emergence of this ecosystem in the software   industry then basically spread the tech  industry, and more and more companies   in the tech industry got organized this way. But an interesting thing happened along the way.   As these ecosystems got built – and you can think  about the mobile ecosystem with iOS and Google,   and then, of course, Facebook dominating  from that as well, and then, of course,   Slack and Salesforce, and so forth, coming on  its heels – what people saw was that the ways   in which you would run an ecosystem platform-based  company was very different than the way in which   General Motors ran or General Electric ran. This meant, oh, all of a sudden we need data  

to cut across our entire enterprise. This meant  that we had a better view of customer journeys and   could personalize and create better offerings for  our clients regardless of if it was a B2B setting   or B2C setting or B2B2C setting, for example.  The typical silos that we had in our enterprise   were no longer the ways for us to organize. I think this is what's new is that what we're   seeing is this pressure to de-silofy our  traditional ways of organizing and to take   advantage of the fact that we now have  digital footprints and data across both   our company operations and our interactions with  customers and our suppliers. How can we put it to   use more effectively and more efficiently? What we see really is, in many ways, sort of   two models emerge. There is the traditional  model in which all organizations, including  

Harvard Business School, has been in for about  100 years, which basically scales very fast and   then reaches a plateau in terms of our ability to  serve more and more customers and drive more and   more value. You could imagine basically  a concave curve of the number of users,   your scale, and the value you're creating. Then we have these digital, AI-first native   companies emerging which are growing, in many  ways, in exponential rates. It takes a while   for them to achieve scale. But once they achieve  scale, they can keep growing exponentially. And  

so, there's a convex curve that shows up instead. These convex organizations, these exponential   organizations, at their core are set up with data  cutting across the entire enterprise. At the core,   drive automation in their processes.  At the core, are set up to basically   use algorithms to make decisions and make  predictions and drive pattern recognition.  That shift, we think, is fundamental.  We worked our way into it through these   tech giants but, increasingly, more and  more industries are facing that as well. 

In essence, what you're saying is the rise of  ecosystems and then the rise of data becomes the   underlying driver that forces organizations  to change in some pretty fundamental ways.  One hundred percent. Of course, we'd add  in cloud computing and then the advances   in algorithms in the last 20 years. Cloud, in many ways, made technology  

a variable cost instead of a fixed cost. You  could then drive massive economies of scale,   the cloud company provider, to then be  able to take advantage of what you needed.  All that have been these trends going  in lockstep. But that has meant that the  

way in which you run a company and the  way in which you organize production,   operations, are fundamentally different in  the ways you might have done things before.  Subscribe to our YouTube channel and hit the  subscribe button at the top of our website   so we can send you our newsletter and you can  stay up to date on these amazing live shows.  Now overlay algorithms and AI  on top of this ecosystem and   data-centric model for how we must exist as  organizations. Overlay the AI aspect on top.  It's no surprise that the leaders in AI actually— Of course, universities were at the source of   some of the breakthroughs in neural nets and so  forth. But the adopters and the drivers of the   changes in AI have come from industry. Why is that? Well, if you think about  

a company like Alphabet or Google, they face  significant challenges in their infrastructure.  I remember, in 2005 or '04, talking  to people at Google. They said, "Oh,   yeah. We built our own server farm with, like,  40,000 servers, and we have 5 people managing it." 

I was just like, "What?!" [Laughter] At that time. [Laughter]  And they said, "Well, we had to make  advances in algorithms to be able to make   this all self-managed. We didn't want to hire  thousands of people running our server farms."  Remember, in 2004 or '05, this is novel,  right? And so, what happened is that as   these ecosystems grew, as they got embedded within  the lifestyles of us – we search all the time for   information – they were generating a ton of data. This data was laying fallow, and they were like,   "Oh, okay. Well, we need to analyze this data,  so let's use artificial intelligence and drive  

the advancement of these algorithms  so we can better understand the data."  But then they were set up in a very  different way. If you think about it,   there's no auctioneer at the backend of Google  running auctions. It's all machine-driven.  The human element is acquire the customer.  Acquire the customer, then the algorithms   take over. They work with you on your keywords.  They work with you on your SEO optimization and   the auctions. Then every step in your Google  journey is mediated through algorithms. 

They felt the need to advance the  algorithms themselves as a way to   drive their own usage and their own growth. Then  those spilled over into the rest of the economy.  To your question, what's interesting is that the  bottleneck in most traditional organizations are   humans. Like, "Mike, answer my damn email,  please. I sent you the spreadsheet. Can you   analyze this for me," or "Can you please FedEx  me the hard drive so I can go look at this data?"   which is what happens in most organizations or  stuck in some Slack conversation and so forth. 

In many of these AI-first organizations, the  bottlenecks are not humans but algorithms and   our capacity to actually analyze  that. That then opens up scaling   opportunities that are quite significant. What should people in business then be   doing? Okay, you're running an organization  and you're surrounded by this change. What   are the implications of this for you? Let's be systematic about our analysis   because I really think that the advantage is  really technology folks now becoming business   folks and also, by the way, HR folks becoming  technology folks and technology folks becoming   HR folks in thinking about what this change is. The first is, in the book with Marco Iansiti and   I, we talk about business models are changing.  When we talk about business models, we say you   need to be clear about what a business model is. It's both the ways in which you create value,  

why do customers want to interact with you, and  the ways in which you capture value, the ways   in which your company makes money. Those need to  be separate sets of analyses that you need to do.  Now, you can create more value  with algorithms and with AI and   with digital. You can be more personalized. You  can scale better. You can offer your customers   much more variety and scope and so forth. Think about how your customer journeys and  

the value creation journeys that your company  does can be enhanced through AI and digital,   digital journeys. That's the  first bit. You lay that all out.  Then separately say, "Now that I'm creating all  of this value, how might I capture all of this   value as well?" The typical model, I would say,  if I create value from you, I charge value,   some portion of that value from you. When people come to Harvard Business School,   we create a ton of learning value for them. Then  we charge them tuition for our value capture. 

What's happened is that, now with AI, you  could automate value capture. You can scale   value capture. You can actually even be more  creative in value capture, like for example,   again, the tech industry has been based on  the fact that they create value for us as   users and they capture value from advertisers. There are just many more ways to capture value.   Thinking systematically about how algorithms,  AI, and digital can help you capture value is   a separate conversation that opens up.  That's just on the business model side. 

Then we can go, "Okay, now let's bring it to the  operating model," which is what actually delivers   the value, what happens inside the company.  There, we think about three things to scale.  How do you serve more and more customers through  digital operations and so forth? Here again,   what you can imagine is that you want to  reduce the marginal cost of acquiring more   and more customers through digital. You can impact scale this way. Scope,   which you offer them. If you think about your  experience now with tech industries, you do more   and more things with these tech businesses. How  can you improve the scope of things that you do?  Then learning, how do you learn better, how do you  innovate better as well through machines and the   data being infused throughout your organization? We see the transformation tasks for business   leaders is to systematically think about  applying this technology to your business   model and your operating model. Which then completely begs the   question how to do it because it's very easy  to describe this but the execution and practice   is massively difficult because  the implications, the tentacles   extend through every part of the company. It's a total rewiring of your organization,  

and that's what's ahead for most people.  Guess what. It's damn difficult. [Laughter]  [Laughter] And it's something that the   C-suite – this is CXOTalk – cannot outsource  to other people. They have to become experts   in what's going on and drive the change. I would say there are three things. One   is the burden on our current leaders of  organizations is to learn this new stuff   and not be afraid of it. This is a new body  of knowledge that you need to acquire not so  

that you're going to become a data scientist or  machine learning engineer or a cloud specialist.  The joke I make at HPS is people come to HPS and  we have a required curriculum for the MBA program,   for sure, and we teach them accounting. If we  made accounting an optional course, nobody would   take it – or very few people would take it. My dean was the chair of the accounting unit.   It was like, no offense to my dean, but  accounting, we make it required because   we feel like this is important. We feel – we  know that in order for you to run a modern  

business, you need to understand accounting. Our sense is now today, in November of 2022,   that data science and algorithms is as  essential as accounting for people in   business to know. Here's why. We don't want you  to become an accountant when you come to HBS,   but we want you to be a good business leader. Similarly, when we teach you data science and   we teach you algorithms, it's not so that you're  going to become a data scientist. We want you to  

become a good business leader. That becomes the  essential bit because if data science and AI is   going to be infused throughout your organization,  you better understand the ways this works and,   in many ways, the downfall of not doing this  properly as well. That's the first thing.  The second thing is this embrace of the  broader technology stack. What I mean by   this is that too often technology has been  viewed as edifice-building, like we'll go do   this technology project like we're building  a factory, and then we'll forget about it. 

I'm sure – in all of your more than  700 programs you have run – you know   that this is an ongoing task. Nothing happens in  companies without software, without technology,   today. We might have it done really poorly  but, in fact, that's what we need to do.  We know, company after company, the  tech companies, Amazon has written   the systems 3 times in their last 20 years  of existence. Many companies have to keep   rewriting their systems over and over again. Leaders need to say that the technology build   is an ever-going thing and we can't sort of have  that be outsourced and put away. We need to own  

it and think about it and be thinking about this  as an ongoing set of investments we'd be making.  The last bit, which I think is the most critical  bit. If you think about the first two bits,   the data science and the technology  stack, I will say that's like 30%.  

The 70% is the change management you need to  do and the change in the organization you need   to do. That is the hardest, hardest part. What I tell technology executives that I   encounter here at Harvard Business School is  that I'm like, "Guess what. You better become   an HR specialist as well. You better become  a change management leader as well. You can't   outsource this to anybody else. This is a change  process that you need to embody and lead as much   as other business leaders need to as well."  That for me is the 70% part of what lies ahead.  I think too many people, too many boards,  too many CXOs index on the 30% and not the   70%. I am convinced the 70% is necessary but  not sufficient. You have to do that stuff,  

and you have to become good at it. But you as  a leader now have to drive the organizational   transformation as well with this. Well, of course, it's much easier   to focus on buying technology. Let's buy a digital  transformation. There's a great vendor. We pay our   money, and they just do it, and it's done. But it fails because it never took   inside the company. [Laughter] [Laughter] Right. Exactly. So,   as I have interviewed so many business  leaders, without a doubt the common theme   is just as you've said that the hard part about  any kind of transformation (whether it's digital   transformation or the kind of next evolution of  digital transformation that you're describing),   it's always the people. But we have a really  interesting question from Twitter. This is  

from Arsalan Khan. He says, "AI needs business  process optimization along with integration of   data (inside and outside the organization)." Here's his question. It's a great question.   He says, "How do you reach consensus with  vendors, partners, even internal departments   who are not at the same maturity when it comes  to AI adoption? How do you make this happen?"  An interesting case study that we should think  about doing later is the transformation at   Disney. If you think about Disney and Disney+  and how they are actually now beating Netflix   at their game is an amazing technology and  business model transformation story as well. 

I had a chance to interview Bob Iger. Last year, I  led an effort here at HBS to drive our own digital   transformation, and I had a chance to interview  Bob Iger, the former CEO and chairman of Disney.  He said you don't just ask for buy-in. You demand  buy-in. [Laughter] This is Bob Iger, the icon of   the entertainment industry and so forth. But he said, "Look. Leaders have to demand   buy-in. You can't just say, 'Oh,  I need your buy-in.' No, no. 'Hey,   you're in or out.'" There's a hard answer, for  sure, which is like, you've got to drive buy-in. 

The second thing I would say is, look, I think,  in many ways, we as the people driving the   transformation have to become good teachers. We  have to make sure that people come along with us.   And the way to do that is to take on a teaching  role, to take on a learning role for them.  That's our job. They won't be able to do  it themselves. You have to be taking on the   responsibility to say, "How do I show you that A)  this is approachable and B) that this is doable?"  My great colleague Tsedal Neeley, she's at  Harvard Business School too and a professor,   she has this great thing called the hearts and  minds matrix. You've got to change the hearts,   but you've got to change the minds. The minds are changed by training,  

by learning, by making people see that, yes, I am  doable. The minds you do through motivation and by   showing the relevance that this has. You have to attack both sides simultaneously.   Change the hearts and change the minds, and  invest in both of them. Again, that's part of   the transformation journey that many companies  get stuck at because they don't think about   the hearts and minds collectively together. Karim, if we think about the kind of changes   that AI and algorithms drive across a company,  can you maybe give us some examples? For example,   you mentioned the business model. You mentioned  relationships with customers. There's talent.   There's sales. AI changes relationships  across all these different processes. 

There's been a massive explosion in these  diffusion models and large language models.   Some analysis shows that the rate of  improvement is 10x Moore's Law, 10x Moore's   Law in these large language models and in these  image-generating diffusion models, and so forth.  Somebody showed me this Twitter thing which  sort of blew my mind, which was like you can now   autogenerate videos saying— Mike, let me ask you. Are   you a dog person or a cat person? My wife loves cats, and so the right   answer is I'm a cat person – and that's for sure. And is there a particular breed of cat that your  

wife likes that you have? Oh, we love all cats – and   I hope you're listening. We love all cats. All right. Great. Now that we know this about you,   we can custom create on-the-fly content for  you and your wife that always has cats in   our promotional videos at zero marginal cost. Now I'll say, okay, we're going to sell Mike   some microphones, but we should have little cats  floating by because his wife will see it and say,   "Oh, definitely those microphones are more fine  than the other ones without the cats." Right? 

That level of personalization is kind of  incredible. The fact that I can now generate,   on demand, at zero marginal cost, these  videos and fine-tune it to you is kind   of mind-blowing. But that capability is  here today. That capability is here today.  What OpenAI is doing, what Google is doing,  what Facebook is doing with these kinds of   technologies is mind-blowing. Just think  that I can now generate personalized ads  

for each person, tweaking based on  their preferences, changes marketing.  How would I run a marketing department now when  I can create personalized content at scale for   each individual? Think about the marketing  supply chain from how ideas get generated,   how campaigns get created, to how they get  launched, to how they get observed and they get   monetized. That whole function with these large  language models, both in terms of text creation   and in terms of content creation, blown away,  blown away and rethought through. One example.  I spent a bunch of time with Flagship Pioneering  to think about how AI and biology are merging   together. The same diffusion models that we see  for ad creation can also be applied to creating  

proteins. The same capability for proteins. Now  just think how the R&D process changes because   now I can generate any protein I want. In fact, one of the companies that we   have in our portfolio that I've been  advising is Generate Biomedicines,   and their view is that they're creating a platform  that can generate any protein, proteins that   have actually not even existed in the world  before, based on these types of technologies.   Just think about the R&D function changing. Now I've looked at two very distinctive settings:  

the R&D function, which we've always thought  requires this creativity and geniuses,   massively augmented by AI. But then the  marketing supply chain being completely   turned upside down and fully automated this way. Now companies that will have access to data about   you and your wife and can have permission from you  and your wife to use that data to do that kind of   marketing will be very differently organized.  Companies that have an ad agency, creative   department, they take six months to create a new  ad. That ad is put on TV or even runs on YouTube  

but is nondifferentiated and so forth – examples. A cool thing I recently saw on this was in sales.   Apparently, now lots of sales, because  of the pandemic, a lot of sales moved to   Zoom and people are now comfortable with having  initial sales conversations on Zoom and so forth.  Well, there are toolings that you can add  onto Zoom that becomes, like an earpiece here,   an earpiece for the salesperson to say,  "You're talking too much. Slow down." Live,  

while you're in the conversation. "Pause for more  questions. Ask a question this way. Your tonality   seems to be more aggressive. Be softer." Realtime coaching for salespeople as to   how to respond to a customer, and that's  all AI-driven. Imagine how your sales,   your face-to-face sales, process is now changing  because you have this technology available. 

It's really augmenting capabilities that we just  have not thought through properly before. That I   think is the amazing thing that's ahead of us. For CXOs, then the question becomes, well,   where do you begin? Do I start in marketing?  Do I start in R&D? Do I start in sales? Do I   start in operations? Where do I begin? That's why these guys that you bring on   your show get paid the big bucks. That's part of  the judgment that they need to have to say, "What   are the high-value opportunities for me to start  to do this? Then as I begin the transformation,   how do I bring everybody else along in this way?" Of course, there are innumerable software   companies now who are selling products and each  one promises that it will be easier than the next.  Yes. "And we all have incredible capabilities   because of the data," and blah-blah-blah.  We've all heard these sales pitches endlessly. 

Lisbeth Shaw asks a question on Twitter  that is directly related to this. She asks,   "How can established companies become AI  companies while they run their existing business,   because you don't want to go out of business  while you're transforming your company?"  A thousand percent. Lisbeth has it right,  which is that's the biggest challenge.  We don't have the luxury to be greenfield.  We actually have to transform ourselves.  What we've seen is there is a joint top-down and  bottom-up approach. Declarations by the C-suite  

to say, "This is the journey we see ahead for us,  and this is the way we need to go towards." You   need the C-suite, the CXO buy-in, and belief,  and a painting of a vision of what that means.  Then what I would say is – that's the first thing  – in that vision is, how will my customer value   get enhanced; how will my clients be better  off if I imagine this world to be? This is   part of the top-down strategy around this. Then it's a question of saying, "Okay. Which   are the problems that we should go after?"  What I would say is it's easy for you to say,   "I've got to rebuild everything," and it's like  you're never going to rebuild everything. You   don't want to be in this world of, like, I'm going  to pause for five years and rebuild everything. 

You want to say, "All right. There are two  things I need to do. I need to deliver value   but also build capability so that I can do this  more and more often and do it along the way."  You then look around, either on your business  model side or on your operating model side. Again,   on value creation and value capture or on  scale, scope, and learning. Say, "Where are  

some high-value problems that if I solve and I  demonstrate that these get solved that I can then   take that and then scale it across my enterprise?" I start with a prototype. I start with a POC. But   the POC doesn't sit by itself.  The POC is designed to scale.  You say to the folks the green light to the POC  that if this works, what is our plan to scale,   and you have the plan to scale agreed  upon before even the POC starts.   What we've seen over and over again  is that the POCs actually work.  I've seen amazing hit rates for POCs working. But  then they all are dead zombie projects in many  

organizations because there's been no commitment  to scale. The commitment to scale then means,   "Oh, I've got to change my operating mode, the  ways in which I do that," but you need buy-in.  It's the bottom-up identification of use cases,  bottom-up identification of POCs, top-down   agreement that we're going to do this and that,  as these POCs start to scale, you prioritize. We   will then make them go across the enterprise. The thing I learned from some colleagues –   you know I was just spending a bunch of time  at Boston Consulting Group (before I became   academic), and I've reacquainted with them  since I wrote the book – they had some very   interesting perspective that oftentimes people  get into this prioritization game, like, "Oh,   which projects am I going to prioritize?" The reality is, in a top-down transformation,   you'll need to do everything. And so, the  question is one of sequencing. The sequencing   of the projects and the scaling actually has  to be based very cleverly on your strategy. 

Is the strategy to blow away your competition  and be the low-cost provider? Then the projects   you would do for AI are very different than  saying, "I'm going to be number one in customer   satisfaction and new business model creation."  That's a very different set of perspectives.  The use cases get identified at the  bottoms-up level. You need top-level agreement   that this is the journey they want to go on. But  then top-level agreement to say that as these   POCs get developed, we're going to sequence them  and scale them to meet our strategy. That's the   way that these transformations will work. If you think about Disney as an example,   where did they start? They first started by buying  Pixar. Then they boat anchored Disney Animation  

Studios to create digital animation.  That was 15 years ago, 17 years ago.  Then in that journey, they've gone step-by-step  to build their own digital capabilities and start   to build a platform where then Disney+  launches just before the pandemic and   can take advantage of people's home viewing,  but then keep going that way by being able to   actually beat out Netflix at their own game. It's interesting. Just as you were describing   Disney on Twitter, Michelle Batt came in to point  out that Disney's success is also related to   leadership pushing from the top down. This is Bob Iger saying, you know,   demanding buy-in. Lag [Laughter] It reminds me of   the great leader of our time Elon Musk going to  Twitter and saying, "You will now work 24/7. And,   by the way, we're firing half of you today." You may disagree with his personality and  

his politics and his incessant use of Twitter,  but his ability to change the space industry,   the auto industry, the electric industry, you  know, electrification with SolarCity, you can't—  He's done things that we'd be lucky to do in  one lifetime. He's done three of them already,   and we'll see what he does with Twitter. I'm not a big fan of his management style but,   guess what. One of the most important questions  he asked at Twitter was, "How many people are  

writing code that ships versus managing?" It's like, "Oh." I think the ratio   was 5:1. He goes, "Okay, that has  to change," because, in the end,   most of our companies are going to be embedding  our processes in software and technology.  That's the key thing that I think CXOs  have to get their head around that   everything we do is going to be embedded  through software, through technology,   through AI. That's where you have to then make  resource allocation decisions and so forth. 

The technology and the AI is going to augment  our humans. It's not going to replace them. It's   going to augment them, but the processes  you have would have to be very different.  We have another really important point from  Arsalan Khan. He comes back, and he asks   about the bias question. He said, "With data and  algorithms—" I'm paraphrasing his question but,   essentially, he wants to know. He phrased it  really well. He says, "How do we reduce bias in   AI when the ultimate goal is increasing profit  and not necessarily AI's impact? For example,   changes on the workforce or in society." How do  we balance these? It's a really important issue. 

Let's unpack this. One is, why is there  a bias problem with AI? Well, because   bias can exist because our data that we  are using to train the algorithms is not   representative. We just have one class of citizens  generating the data instead of another class.  Our labeling operations may not be representative  as well. For example, lots of tech companies have   problems identifying blacks in their image  processing systems because the labelers   weren't able to identify them properly or  distinguish the features that way as well.  One is a story of data and data operations. This  is why data science is a critical skill for all  

executives because you have to understand the data  generation processes and all of the faults that   would happen. I think that's the first thing. To take it to the limit, just as I can scale   the benefits of AI exponentially,  I can also scale the harms of AI   exponentially. Bias is one of those things. The second thing is that there is a real legal   issue, which has been the thing that has been  so interesting for me. Statistically, computer   scientists and statisticians, when they look at  the algorithms, say, "Is this algorithm fair?"  Oftentimes, when we think about fairness  in statistics and in computer science,   we think about on average. Is this algorithm  treating people fairly, on average. But the law   doesn't say average. The law says each and  every individual has to be treated fairly.  There's a lot of risk that companies are facing  today because their algorithms are, on average,   fair but to the individual they're not fair.  They're open to a lot of liability questions. 

How do we make sure that those things are  addressed upfront instead of addressed   after the fact? This is where I think is  the new frontier for many organizations,   which is the conversation about bias and  fairness and transparency in expanding   the algorithms should not be a  computer science or an AI task.  This is a cross-functional task that resides with  business, with technology, and with legal and   policy. This has to be done collectively. Importantly, we can't do this   ex-post, after the algorithms have launched.  We have to do them pre- in the design phase.  I think Satya Nadella has done the most  thinking about this because, remember,   Microsoft had a crazy amount of cybersecurity  issues in the 2000s. What they had to do was   retrain their software developers to build quality  software and security into their processes instead   of doing it ex-post. I think the same thing is  going to happen with algorithms and bias and AI   is that we have to build in the awareness about  bias in our processes upfront instead of ex-post   when the algorithms are released into the wild. The example I use is that when you go to Toyota  

(as a manufacturing company), there's no quality  department at Toyota. Why? Because they feel like   if your processes aren't creating quality then  a quality department is never going to fix it.  They make quality the responsibility of all  employees, and they've built processes to   ensure quality is built into the systems instead  of doing it at the end. I think the same thing   is going to happen around AI and bias as well. This is from LinkedIn. Cezar Babes comes back,  

and he responds to you the following way. I'll  ask you to just keep your answer pretty brief.   He says the following: "Should AI use be  more tightly regulated? Every now and then   there's a new technology that becomes the  catalyst for—" and I'm readying his quote   "—profit-driven goose chases. This results  in loss of jobs and resentment towards that   technological advancement." He says, "It would  be great if AI would be a driver for human growth  

and result in increased capability and capacity  rather than right-sizing and cost-efficiencies."  Is there going to be displacement  because of AI? Absolutely.  Do we need to retrain people? 100%. But my belief is that, in the end,   AI augments human capability instead  of destroying human capability. Just as   prior technologies have been enhancing us, the  same thing is going to happen with AI as well.  Is there a displacement period and  are certain occupations going to be   displaced? 1000%. That's where governments  and so forth have to come together. 

But regulation, who is going to regulate  AI and in what way? It just doesn't seem   tenable to us because it is so widespread. What advice do you have for business leaders   who are listening to this and saying,  "All of this is fine, Professor Lakhani,   but my business is successful. We don't have  to deal with this stuff. We're pretty much   happy as clams, so this doesn't affect us"? Go talk to your customers, and not about   your products but other things that they're  doing. I tell you; you will be shocked with   how they're thinking about the world and how  much technology is driving their decisions.  When you ask them about your own  products or your competitors' products,   you will never hear the right answer. Ask  them about other things that they're doing  

in their businesses, and you'll be shocked. What advice do you have for business leaders   who are listening to this, nodding their  heads, and saying, "I know this is true,   everything you're saying. We feel the pain.  I feel the pain, and I don't know what to   do. It's too big and complex and hard." There is a learning mandate for this for   all organizations, which is, we have a generation  of leaders that came in the old model. They don't   understand the technology, don't understand  data science, don't understand statistics,   don't understand algorithms, don't understand  cloud, and feel like that's for the IT guys.  I think there's a learning mandate for these  leaders not just for them to become better   at this but then to also get their whole  organization to change as well. And so,  

I would start with learning. Invest in the  learning for yourselves and your folks.  There's so much stuff available. What you have  done through this amazing series, what we offer   for paid, there's lots of stuff. There's no  excuse for not learning. There are lots of books.  Invest systematically in learning, yourself,  and building a framework for your whole team.   Then cascading that down so that everybody has  the same reference point. That's the first step. 

I see too many people shirk on learning  and say, "This doesn't apply to me," when   they don't know what's going to hit  them over the head with this stuff.  Jose Kurian just wants to point out that security  is the most important of AI and machine learning   services, and so can you just say something  about the security dimension of all of this?  Security, overall, as we get into more digitally  intensive organizations, which all of us are   becoming, data security, information security is  going to be key-key-key for all of us. Secondly,   there are actually a bunch of very important  issues about data security and our data pipelines   being secure and not being tampered with. Just think about labeling operations that   many companies have. Many times, those labeling  operations are outsourced. They could be subject   to attack where even just the slight bit of  mislabeling could give you flawed algorithms.  As we start thinking about this stuff  becoming infused throughout our enterprises,   the security side around data itself is going to  be massively important. I 100% agree, 1000% agree. 

With that, we are out of time and over time.  I want to say thank you so much to Professor   Karim Lakhani from the Harvard Business School.  Thank you so much. I'm so grateful for your   taking the time to be here with us today. It was so much fun, Mike, and it was great  

for me to be on this side instead of just  listening. Thank you for the invitation again.  Well, I hope you'll come back again. Absolutely.  A huge thank you to everybody in the audience  who watched and especially to the folks who   ask such amazing questions. You guys are such a  great audience. I have undying respect for you.  Everybody, thank you so much. Check out We have amazing shows coming up.   Before you go, subscribe to our YouTube channel  and hit the subscribe button at the top of our   website so we can send you our newsletter and you  can stay up to date on these amazing live shows.  Thanks so much, everybody. I hope  you have a great day. See you soon.

2022-11-27 03:10

Show Video

Other news