Arun Sundararajan is the Harold Price Professor of Entrepreneurship and Professor of Technology, Operations and Statistics at New York University’s Stern School of Business. His research studies how digital technologies transform business, government and civil society His best-selling and award-winning book, “The Sharing Economy,” published by the MIT Press, has been translated into Japanese, Mandarin Chinese, Korean, Vietnamese and Portuguese. Sundararajan has been a member of the World Economic Forum’s Global Future Councils on Technology, Values and Policy, and on the New Economic Agenda, and an advisor to or board member of organizations that include Fortune 100 companies, tech startups, nonprofits and city governments. He has provided expert input about the digital economy to the United States Congress, the European Parliament, the United Nations and dozens of government agencies and regulators globally. Sundararajan holds degrees in electrical engineering, applied economics and management science from the Indian Institute of Technology, Madras and the University of Rochester. He recently spoke to The Innovator about ChatGPT, generative AI and the future of work.
Q: What will the impact of ChatGPT and other forms of generative AI on the workforce?
AS: ChatGPT and, more broadly, the new generation of generative AI, are technologically similar to other kinds of enabling AI that we have encountered in the last 10 years, but have reached a new performance threshold. Primarily, this is because generative AI can actually finally generate realistic things that are normally created by humans such as poems, articles, art, movies, and songs in a seemingly unsupervised way. That said, to ensure minimal performance quality in any meaningful government or business application of generative AI, the current near-term state of the technology is going to require active human oversight. One of the things that has become quite clear in early use is that the technology is rife with inaccuracy, in part because accuracy was not explicitly designed in as an objective. Accuracy will increase over time, but for now, despite its advanced poetic abilities, it’s best to think of ChatGPT as a semi-skilled draft generator. It’s analogous to the way AI has been used in drug discovery for a few years to generate potential combinations of molecules that have the highest likelihood of leading to something therapeutic because humans don’t have the bandwidth to consider so many – but the humans have to nevertheless take over after this step before the drug is actually made.
What will this mean for the future of work? There is a long history of machines starting to get good at doing specific things at a much faster pace than humans, so this is not a new development. It’s important to remember that most work comprises a wide range of tasks, and so dramatic improvements in the capability of computers doing one task does not mean a whole bunch of people will be put out of work instantly.
However, there are segments of the work force – like the people who generate simple computer code that someone assembles into a more complex system – which are under immediate threat from generative AI. ChatGPT and its generative AI peers have gotten incredibly good at writing increasingly complex snippets of computer code. So the people who write simple code are certainly threatened. For visual art and music and video entertainment, generative AI will create completely new categories of AI-generated products.
But we are not being replaced by machines because the economy is not a zero-sum system. It’s an evolving system which grows with technological progress, and as the machines increase our productivity, this also expands the need for humans.
For example, a generative technology that can create computer code based on intelligent direction enhances the capabilities of non-tech savvy humans who need to make decisions based on data, and thus raises the overall quality of human output. Most of the business workforce depends on another human to make sense of their data. Generative AI will give people in the workforce the power to write simple computer programs to analyze their own data without having to rely on another human being. This dramatically expands the use of computer code for data analysis.
Similarly, in medical diagnostics, generative AI will allow an increasingly large number of people to be empowered to make assessments. This is not going to take away human work, but will instead dramatically increase the number of human ailments we have the bandwidth to analyze and deal with. In many parts of the world, access to quality healthcare via generative AI will not be used to replace physicians, it will instead make up for the absence of available physicians. Think about it – two hundred years ago we had no healthcare industry to speak of and today, and today it’s a double-digit percentage of GDP, but despite so much progress, we are still just scratching the surface of fulfilling the healthcare needs of humans. Generative AI can help fill that gap.
It is my belief that generative AI will have a similar augmentation effect on a wide range of other work processes. Machines don’t replace humans, they create new partnerships and allow up to re-imagine how work is done.
Q: Beyond coding and the arts what do you see as the business applications for generative AI?
The family of technologies enabling ChatGPT to be so good are called large language models[LLMs] or pre-trained language models. The LLM that ChatGPT is using is GPT-4, released very recently as a successor to GPT-3.5. These are massive models that has been trained on all information that is publicly available on the Internet to do one primary task – to predict what is the next word when given a set of words. ChatGPT is a separate layer on top of the LLM, trained on actual human conversations and trained to intelligently send the right prompts to the GPT models. This additional training and engineering investment induces the LLM generate a sequence of words that makes it seem like we are talking to a human being.
I believe there will be an explosion of other layers similar to ChatGPT on top of GPT-4. For example, applications on extremely well-defined topics that are specialized such as analyzing U.S. case law, figuring out gluten-free diets, and providing specialized customer support for particular products. What has been hard up to now has not been so much the ability to train for accurate support but more the inability to easily create human-like conversations. Generative AI overcomes this hurdle. Think of LLMs as having been taught some grammar and having been given a whole range of general knowledge. The layers on top then are teaching them to specialize in a particular topic or style of interaction.
The most promising area where I see generative AI being applied is for knowledge management within organizations. For the last 20 years, knowledge management has been coveted by organizations. In the early days of the Internet there was a lot of talk about intranets and how companies were going to index and organize all of their data. Let’s just say that the results have not been breathtaking to say the least – but if you can combine GPT-4’s capabilities with a specialized corpus of human generated documents that are proprietary to an organization you can create extremely effective supplements to human beings to the point where if someone is on vacation and you want to ask them a question you may be able to get those answers from a generative AI that has trained on a corpus of documents, emails and contracts generated by that employee. That is the world that large language models are leading us into.
Q: Are you saying that generative AI will create the equivalent of digital twins of employees?
AS: It could create static twins. These will not be a long-term substitute for human beings but rather more like a short-term band-aid. The world is constantly evolving. No company will want a version of an employee that only has last year’s knowledge.
Q: Will generative AI accelerate the need to reskill the workforce?
AS: We are currently preparing the work force in one early burst of education – in high school and technical schools and colleges – for careers that someone imagines will last 50 years. This no longer bears any reasonable connection to reality. The pace of change in the division of machine-human labor – what machines do and what humans do – has picked up. Over the next 20 years, tens of millions of Americans and hundreds of millions of people globally will find themselves needing a new occupation mid-career.
Granted, it’s not like society is sitting on its hands. Governments I talk to are spending substantial sums on reskilling and companies are investing in life-long learning programs. The real trouble is that 80% to 90% of transitions to new occupations will also require transition into new companies or organizations. So we need reskilling to be wrapped into a bundle of other forms of support, the way your college major is bundled with critical thinking, networking, recruiting help, a branded degree, and the rite of passage to adulthood. Similarly, you can’t just teach someone a new skill like data science mid-career and consider that your job is done and send them on their way. You need to wrap other things around that and imagine how they are going to actually make the transition. Can this be done in a viable time frame if they have a family to support? Can you connect them with peers that have made the same transition? How can you build their confidence? Can you give them mentoring? Placement services? A branded credential? Creating these new mid-career transition bundles is, to me, the most important public policy challenge of the next 20 years.
Q: How can companies help ease the process?
AS: First, invest in skill development that doesn’t assume the person will remain in your company and set up avenues for people to transition out with dignity. Next, help workers to continuously think about what else they can do given the capabilities they have. Companies need to build this into how they work, even on the factory floor. They should encourage workers to think about what they are good at and rotate them into different roles so that they are in the mindset to eventually transition either within or outside of the company.
Q: Why should business care about this?
AS: To me this is more than just a business problem. The polarization around the world over the last 20 years has been caused by mismanaging the manufacturing automation transition in the 1980s and 1990s. We did not create significant economic opportunities for laid-off factory workers. We did not prepare them for today’s world of work. As a result, there are large groups of people who simply lack economic opportunity and are unable to succeed. The divisions between cities and rural areas have led to far more than economic distress. It has led to large groups who are susceptible to social media filter bubbles and ideas of tribalism and revolution because they don’t see opportunity for themselves. This has destabilized the U.S. and many Western European countries.
The magnitude of the coming AI and robotics transition is at least twice as big as the 20th century manufacturing automation transition. So if we want our societies to remain stable, we really, really need to come up with a way for people to transition with dignity. It is such an important need and plays out over such a long period of time. The trouble is our political systems are based around short-term promises, and the polarization that resulted from the last wave of automation is ironically further stymieing our ability to address this wave.
Q: How should companies prepare for generative AI?
AS; First don’t panic. Stay calm. Think of generative AI as a source of opportunity. It is not a substitute for workers but a supplemental way of creating specialized pools of expertise for clients,and its technology can be used to improve knowledge management within your company or work force.
Next, accelerate the pace of preparing your company for the workforce transition. Any sensible organization should put transition into two buckets: transition within – training people for new roles in your company – and transition out – preparing people to leave the company and start new occupations elsewhere. Transition within is a good place to start because the shareholder argument is stronger. To help make the transitions out work better, I encourage companies to try and form networks or alliances with other organizations to help lower the barriers for each other’s workers who are transitioning out. I also encourage them to create peer networks, connecting people who have successfully transitioned with those in process.
I am optimistic that doing a good job of transitioning out will be favorably viewed by society in the years to come, much like sustainability investments which were once seen as a cost center are now seen as a positive by consumers. Workforce transition planning is about more than making the argument that “it is the ethical thing to do” or that “it is good for society.” Sure it is, but it will also be aligned with any company’s business interests in future so it is good to start doing the groundwork now for what will be a complex multi-year process.
This article is content that would normally only be available to subscribers. Sign up for a four-week free trial to see what you have been missing.
To access more of The Innovator’s Interview Of The Week articles click here.