But first, I’m joined by Alondra Nelson, former acting director at the White House Office of Science and Technology and senior fellow at the Center for American Progress.
Dr. Nelson, welcome to Washington Post Live.
DR. NELSON: Thank you so much, Danielle. It’s a tremendously important topic, and I’m really glad to be here.
MS. ABRIL: I want to start off with the study from Goldman Sachs that found some 300 millions of jobs will be lost or diminished by artificial intelligence. Some experts suggest this change will create a boom in productivity, but how do you see this playing out?
DR. NELSON: So this Goldman Sachs report from the spring, you know, was I think our first attempt to really get around the data, think about models and patterns that we’ve seen from the past and try to anticipate what this might mean for the future and global gross domestic product globally. So that report anticipated that there could be as much of us as a 7 percent increase and like the total market value of goods and services and that we might lift economic productivity over the next decade by more than a 1.5 percentage points.
To drill down a little bit more, the report anticipated that about two-thirds of U.S. occupations are touched in some way by the potential for automation, and that a percentage of these, about a third of these, could have, you know, work tasks sort of replaced or transformed in a pretty profound way.
So, you know, I think that there is–there are potentially kind of large structural changes coming and happening, that job displacement and disruption will be some of this change potentially. One of the things that I also found striking about the Goldman Sachs report, however, is that it was filled with qualification. So while it was bullish on the potential for what artificial intelligence might mean for work at its best, it was also filled with words about uncertainty and about possibility and, you know, that these things might happen and ways that I think economic forecasts are often not couched in such language of uncertainty. So, I wanted to be sure to highlight that, and that uncertainty was about when and how these tools might be adopted, about how quickly they might be adopted. And I would add as a kind of asterisk the kind of unreliability or the sort of uncertainty around the reliability of the tools and systems themselves.
MS. ABRIL: Yeah. That makes a lot of sense, and we’re seeing a lot of that play out, right, even employers who have moved to reliance on artificial intelligence and then realized they’re not that–they’re not as maybe trustworthy or ready to go as maybe people expected.
The National Bureau of Economic Research found automation technology has been the primary driver of U.S. income inequality over the past 40 years. Well, as this continues to ramp up all of this generative AI exploration, how do we prevent the problem from being exacerbated?
DR. NELSON: Yes. That’s the great question of our time, Danielle, and the upside of it is that we have an opportunity to actually do something about it. So, you know, the upside of things like the Goldman Sachs report, the conversations that we’ve been having in the reporting that’s happening at The Post and elsewhere, is that we recognize that we’re–you know, that there’s a transformation about to happen, and we’re in a moment when we can do something about it.
So on the inequality piece, I mean, if we think about–you know, one of the things, for example, one of the ways that AI is being used is with regards to home appraisals and whether–and the sort of costs of homes, and there’s obviously a racial and sort of ethnic gap in valuation of home prices, for example. So if we think about the inequality in that gap and we think about homes as, you know, it’s the American dream, it’s the way in which people stabilize families, it’s how we build wealth. So having algorithmically driven, you know, sort of inequality in that space, you know, actually erodes and works against, I think, some of our sort of highest policy aspirations and aspirations for our country.
And then we see these tools being used in other places in which, you know–health care, for example, in which the, you know, benefits of the sort of use of algorithms are leading to the sorting of particular sorts of recommendations for care to groups that already have good care, meaning that other groups are getting less care. And so we’re also getting the potential for widening inequality in the chasm that’s produced between some communities getting less and some communities getting more, and then, of course, there’s just the inequality that might be driven by mass disruption to the workplace that’s not steered by, you know, good policy, so layoffs, certain kinds of violations of worker privacy, of worker rights and the like.
But as I said, we have an opportunity really to create levers and systems and ways of thinking about this work and a moment of uncertainty about the tools and a moment in which these tools and systems are being introduced as well.
MS. ABRIL: So some of these more serious negative consequences aside, you know, what kind of positive effects might this explosion of AI have on the workforce?
DR. NELSON: Yeah. It’s a great question, and that’s, I think, the reason why we’re here is because we–you know, we hope for the potential of this.
So obviously, you could imagine the creation of entirely new jobs. We saw with the introduction of, you know, the new technologies of a decade ago or 15 years ago, jobs like, you know, website designers and software designers and these sorts of things. So there’s entirely kinds of new fields, including creative fields, entrepreneur–you know, entrepreneurship, small businesses that might be created by the use of these tools. It’s certainly the case that worker productivity might be increased, and this might lead to an increase in wages, overall economic growth as we–the conversation that we began with, and an overall, you know, increase in living standards.
It might also lead to–you know, help us have safer jobs. So it might be ways in which there is work that has–takes a physical toll, repetitive tasks, dangerous tasks in the workplace that can be sort of offset to, you know, artificial intelligence systems and tools that make work safer, better, and more meaningful. So all of those things are, are potentials, but that–but we as a society, policymakers will really have to shape, you know, those benefits and make that potential possible and true.
MS. ABRIL: So I want to talk a little bit about your previous post as the acting director of the White House’s Office of Science and Technology Policy. There, you spearheaded a 70-page–70-plus–excuse me–page document called “The Blueprint for an AI Bill of Rights.” Can you tell us a little bit about that and what you were hoping to accomplish with it?
DR. NELSON: Sure. You know, so I entered the Biden-Harris administration in January of 2021. The president gave–you know, had big aspirations for what science and technology policy could do, that it’s supposed to improve people’s lives, expand economic opportunity, and that was really sort of the mission of the work. And the AI Bill of Rights was an attempt to think about, you know, as we were making this big structural transformation in society, the use of automated tools and systems, how can government create sort of aspirations and, as necessary, guardrails to guide the design of that. I helped guide the design of that, of those tools, of their development, their deployment, so that the rights of the American public are protected and so that the potential benefits of these really do–you know, are disseminated to all of the American public.
So The Blueprint for an AI Bill of Rights has a few pieces. It’s got five principles: that AI systems should be safe and effective; that there should be data privacy; that you should be notified when systems are being used; that you should have alternative options and should be able to opt out of the use of the systems; and that there should be protection against forms of discrimination, gender discrimination, age discrimination, accessibility discrimination, et cetera, through the use of algorithms and systems. And so that was the–those are the kind of principles.
That was accompanied by–most of those 70 pages that you mentioned, Danielle, are the sort of technological best practices, so how you put these principles into practice.
And then importantly, when this came out in October of 2022, as one of the Biden-Harris administration’s kind of first big, sort of assertions of its policy thinking in this space, it came out with initial executive action. So the Equal Employment Opportunity Commission and the Department of Justice, for example, released technical assistance around antidiscrimination. The Consumer Financial Protection Bureau confirmed that the laws that it enforces and regulates, you know, to protect consumers apply to the work that they do and other agencies besides. And since this time, October 2022, the executive branch has really just gone on to, I think, really expand the potential to use AI systems and tools and in ways that really expand opportunity and mitigate harms.
MS. ABRIL: So obviously, a blueprint is a great place to start, but we have yet to see actually any formal policy around AI at the moment. Are there any action items that the government could address right now as more companies consider operating or incorporating AI into their workflows?
DR. NELSON: So I would disagree a little bit, Danielle. I mean, part of what these executive actions were, were highlighting ways in which agencies, departments could use the regulatory authorities that they already have, and so while it’s certainly the case that this is–you know, the capabilities of artificial intelligence have, you know, expanded quite dramatically and that we need to be thinking in new and fresh ways about AI governance, which includes regulation and other kinds of levers that we can work on, you know, across sectors, including government. It is not the case that these things–you know, that AI is not regulated. So I would sort of just push back on that a little bit.
I would also say that, to go back to where we began, that kind of–that sort of line through or that vein of uncertainty that was in the Goldman Sachs report, you know, one of the ways that we address that uncertainty is through regulation. And so I would want to also offer that regulation guardrails are actually not staunching innovation, but also–but allow a kind of certainty in the field of play for workers, for consumers, and for industry, and allow people to have sort of rules of the road that are clear for everyone.
And so I think, you know, fundamentally with regards to the employment sector, work issues, and industry, the kind of uncertainty that we have, because we don’t have–we haven’t landed on what these sort of new regulatory tools and guardrails might be, I think is–will ultimately may prevent us from, you know, driving, leaning into the innovation, leaning into the opportunities that might potentially lie here for the workforce.
MS. ABRIL: Actually, that leads right into my next question. You know, given this regulatory environment and the way that you put it, that there are some, you know, guardrails and certain ways in place for this, does that mean companies should lean into this technology? Should they be waiting? Is there a certain moment when it’s right to jump versus not?
DR. NELSON: I think that’s the open question, right? You know, we’ve already seen a few instances that companies who have been really kind of leaning into this. So we might think about, you know, The Economist reported about Samsung, Samsung, you know, semiconductor factory–or semiconductors–excuse me–and sort of entity, you know, that was allowing its employees to use one of the chatbots, I believe ChatGPT. They were putting proprietary strategy documents, proprietary code into them, into the chatbot, and were really worried about that being sort of adopted, taken into the training data. So, there were kind of issues around the security of proprietary data, around IP issues that companies have to think about. You know, there are–there’s been reporting about the use of artificial intelligence tools and systems in hospitals, you know, nurses you know, sort of doubting their kind of confidence and their skill set in nursing in ways that might lead initially to, you know, worsening of health outcomes for patients. So I think each kind of sector, industry, entity is going to have to think about their own, you know, kind of risk calculus around the uncertainty of these tools.
And I think–you know, I appreciate that sometimes people think that that regulation and that governance really, you know, adds friction to the innovation that they’re trying to drive, but we’re really in a moment that, you know, the innovation that folks are hoping for is really going to require that we have some kind of shared rules, rules over the road.
MS. ABRIL: Understood. I want to switch over to a question from the audience. Colin Owens of Massachusetts asks, “What is the responsibility of government in ensuring people are retrained to meet the needs of an economy that removes some jobs from the industry?”
DR. NELSON: Yes. So this is an excellent question, and I think that there is a responsibility for government, and there’s lots of the government can do. So I think the range of things include programs like the Department of Labor, for example, has something called the Trade Adjustment Assistance Program, and it focuses on job loss due to technological disruption and focuses on, you know, creating other opportunities for people, other options, upskilling, reskilling, and the like. And so I think a lot of our conversation around work, labor, you know, and AI is just about the disruption and the sort of sense that it has to happen and it’s going to happen all around us, and that there’s nothing that we can do about it. But smart governance, smart regulation can, you know, lean into sort of policies and programs and initiatives that actually help to mitigate that disruption.
We might also, of course, think about things like social safety nets, which state and local governments have been experimenting with, you know, modernizing unemployment and insurance, expanding eligibility for those kinds of benefits, increasing the levels of those benefits.
And then there’s, you know, the thing that we fight about quite a lot: tax policy. You know, one could think about using–government could use tax policy as a way to direct AI development in a positive direction. It could be a general tax to reduce kind of the domination and the sort of consolidation of power in the AI space. You could think about tax credits as a reward for the development and adoption of AI tools and systems that augment labor or augment workers and have a worker-centered approach rather than displacing workers and the like.
So I think there’s lots of policy, innovation, and creative ideas that can happen in this space, and as I said earlier on, the upside is that we have an opportunity to do something about it because it’s very early stage.
MS. ABRIL: So we only have a couple minutes left, but I got to squeeze this one in. You know, obviously, we talked about inequity. We talked about possible, you know, job changes and what that looks like, but what is your biggest concern when it comes to this evolution and more companies, more people using these generative AI tools? Like what are the biggest concerns that you have here?
DR. NELSON: I think that my big concerns are probably principally two: one, that we are facing a big transformation and that we don’t have all of the voices and stakeholders at the table that we need to have, and so this can’t be a conversation for only people in the technology sector or only in the business sector. You know, what is so powerful and potentially exciting about this new advanced AI is that it will impact so much of our lives in ways that are potentially beneficial, the way we live our lives, the way we do our work, the way that we communicate, and lots of things besides, and so that means there needs to be a lot of different voices at the table. And my concern is that we won’t have all of the voices at the table that allow us to do policy innovation, regulation, to do, you know, forecasting, and just to get ready for the transition in a way that takes up the sort of–the opinions and the perspectives of lots of different folks in the American public.
And I think my second concern is that we will go about this transformation with a sense of inevitability, that these things are already preordained, and that, you know, we have a sense of disempowerment about what might happen. And I think, you know, these technologies are really powerful, but it’s also the case that there is a role here for workers, for consumers, for citizens in addition to, you know, for researchers, for industry, for the entrepreneurs that are so important for our society as well. And so there’s nothing inevitable here, and that we have in this early moment an opportunity to really create the potential that many of us imagine that AI systems and tools can have in American society and global society.
MS. ABRIL: Well, I’m sure there will be much more to talk about here, and we will continue to follow this story.
Alondra Nelson, thank you so much for joining us here on Washington Post Live.
DR. NELSON: Thanks very much, Danielle.
MS. ABRIL: And thanks to all of you for joining. Please stay with us for the next segment of the conversation.
MS. KELLY: Hi there. I’m Suzanne Kelly, CEO and publisher of The Cypher Brief, a national security-focused media organization.
It is my pleasure to be here with you today to talk about the impact that generative AI is having specifically on the workforce. It seems like we’re seeing headlines every single day about how this is changing our lives, but how do those headlines translate into how organizations can embrace AI’s transformative potential with confidence while considering the ethical responsibilities and, of course, always keeping humans at the center of that picture?
Joining me today to talk about this is Beatriz Sanz Sáiz. She is EY’s global consulting data and AI leader. Beatriz, welcome.
MS. SANZ SÁIZ: Thank you, Suzanne. I am delighted to be here with you today.
MS. KELLY: I’m delighted to be here with you as well because I’m really curious about how you’re seeing AI-driven generative technologies. I’m very interested to hear specifically your perspective on the nature and the potential of AI in the workforce today.
MS. SANZ SÁIZ: Sure, Suzanne. Look, the nature of GenAI is transformational. It will not only improve productivity or reduce operating costs and optimize back-office processes, but it will transform the way we work.
As GenAI becomes more integrated across all industries, many jobs will evolve to leverage the capabilities of AI systems. Think, for example, health professionals. They will be able to leverage the power of AI to generate more concise medical reports, or think about the world of marketing, like marketers will use GenAI to assist with generating and initiating written content, or a student, he will be able to leverage AI personal tutors. That will have huge impact, helping them to better understand the rationale and steps required for problem solving.
So, look, in essence, Suzanne, if it’s hard for the younger generations to understand how the world works before the internet, like education, research, connectivity, well, the impact of GenAI will be even bigger. A new form of intelligence has emerged.
MS. KELLY: I think you’re absolutely right about that.
You know, we spoke, and you did a great job of kind of laying out some of the opportunities that are presented with AI. How do you see the evolution of work next and beyond in the context of AI?
MS. SANZ SÁIZ: Well, Suzanne, in my view, there is no next and beyond. There is only now. Look, in the last few weeks, we have seen an explosion of companies managed by a handful of people, just because they have leveraged AI and new types of AI. So the concept of the enterprise will be fundamentally reconfigured. So from people, process, and technology, towards technology harnessed by data, executing workflows and supervised by humans, we will see a simplification and standardization to the limit. So the impact of GenAI in the software development industry, for example, is decreasing because of IT and, therefore, barriers to entry for the small and medium enterprises.
So I truly believe that AI will not replace humans, but actually companies that embed AI at the core will displace the ones that don’t.
MS. KELLY: So let’s talk about some of the ethical considerations, then, that go along with this. As data and AI leader, what do you think the ethical considerations are that are going to need to be addressed as AI becomes more prevalent in the workplace?
MS. SANZ SÁIZ: So I’m glad you made this question, Susan, because, look, as important as AI will be to transform the business and the operating models, it is critical to advance the risk frameworks in a world that evolves from deterministic to probabilistic. So principles around ethical AI and transparent AI, they need to be revisited.
New technical standards are needed for governments enterprises and citizens to confidently and safely adopt this new form of intelligence. So we are already working with policymakers, industry experts, and software developers to revisit the ethical frameworks as we move into a world that is probabilistic, but to be honest, there is still a lot of work to be done.
MS. KELLY: Yeah. I’m curious to know how you see the impact that AI is going to have on people who are just now entering the workforce and then also on students as they enter universities, because the world that they are going to be embracing when they come out of university may be very different from it is even today. How do you see both of those things?
MS. SANZ SÁIZ: So for the people joining the workforce, look, I believe this will be a game changer. I think we will start to see roles fundamentally change and new career paths emerge. For example, today, we may be able to do one project, but by leveraging AI, we’ll be able to do several projects in parallel and deliver them faster.
So meanwhile, AI will facilitate the development of more cross-functional skills by providing accessible and interactive learning resources, and this will allow apprenticeships and new employees to develop more effective skills and explore various roles and domains. So people joining the workforce will be able to use AI in the onboarding process and exponentially improve their learning curves.
I think for students, look, I will ask them, “Are you ready to get a chip that will make you smarter?” Look, by this amazing power feature, this could be something that the next two or three generations will see. So I will tell them to be open-minded and focus on the good that AI can do. It’s our power to ensure that we use AI for good. So, I mean, be open to a new form of education. The world of education will completely transform, and students will be able to use an AI power tutor, as I said before, to access unlimited amount of knowledge. So we may not need to spend the next 25 years gaining knowledge when we will have easy access to it. So really the world is at the beginning of a huge leap, Suzanne.
MS. KELLY: I think you’ve given us a lot to think about, really fascinating context around this. Beatriz Sanz Sáiz is EY’s global consulting data and AI leader. Beatriz, thank you so much for being here.
MS. SANZ SÁIZ: Thank you.
MS. KELLY: Now back to my colleagues at The Washington Post. [Video plays]
MS. ABRIL: Welcome back. For those of you just joining us, welcome to Washington Post Live. I’m Danielle Abril, tech at work writer at The Washington post.
I’m now joined by Sal Khan, CEO and founder of Khan Academy, and Michael Howells, the president of Workforce Skills at Pearson. Sal, Michael, welcome to Washington Post Live.
MR. KHAN: Thanks for having us.
MR. HOWELLS: Nice to be here.
MS. ABRIL: Sal, let’s start with you. Artificial intelligence is already a part of our everyday lives. How can education help prepare the workforce for the possible changes ahead?
MR. KHAN: Well, the first thing is the workforce is already using these tools. I’ve told our own internal team, every leader at every corporation is essentially already sending the signal to folks–is if you’re already not–if you’re not already using these tools in some way, shape, or form, you’re probably not doing your work optimally anymore. These can really streamline a lot of tasks, and the tools that leverage generative AI are only going to get better and better. We’re really at the–not even at the top of the first inning just yet. So given that it’s going to be in the workplace, it is going to be hurting students if they don’t get exposure to these tools in some way, shape, or form. Now, it introduces a whole series of risks around students using it maybe when they shouldn’t be using it for doing their homework and cheating. So there’s going to have to be some guardrails in place and some policies in place on when we encourage you to use these tools in order to do things more efficiently and to do it in a more streamlined way and say an academic setting and when you aren’t allowed to use these tools. And we need to make sure that you yourself know how to write, you yourself know how to create, you yourself know how to do your homework assignment.
MS. ABRIL: Yeah, not relying too much on the AI to just do all the thinking for you.
Excuse me. Michael, there have been estimates that some 11 million people in the U.S. will need to retool how they work with the changes AI and the related technology is going to bring. Who should be thinking about this, and how should they be thinking about getting training?
MR. HOWELLS: Well, I think similar to in education, everybody should be thinking about this, and I think most people are. It is very interesting to look at how new entrants into the workforce are thinking about this, how students who are considering going into higher education, for example, are thinking about this. Everybody is aware of the fact that the nature of work is changing really, really rapidly. And we think about this at the task level. It’s not so much that whole people are going to be replaced by AI, I think, as we’ve heard from many of your speakers in the course of this series. New roles, new opportunities are going to be created as these technologies roll out, as is always the case.
We think about this as tasks. What are the specific tasks that can be automated, that can be augmented, that can be made more efficient, and how can I as an individual understand how that’s going to change the nature of work and how I can prepare myself? And so we think about this very much as really a question of how to utilize AI tools in order to empower individuals to make informed choices, how to personalize learning experiences, how to figure out what are the right opportunities for you as you progress through your career.
And we’re just starting to see, you know, the way in which generative AI can improve some of those services. Of course, AI, in general, has been used extensively in personalization and recommendation tools for a number of years. It’s only just now making its way into learning services and products that people consume every day.
But I think, you know, just looking at the way that my children learn, for example, they’re already pretty smart in how to use these things, and that’s only going to get better over time.
MS. ABRIL: Sal, I’m going to jump back to you. What are some of the best ways leaders can help the workforce adapt to the changes that AI brings?
MR. KHAN: We’re still in the very early days. It’s very simple. You know, even the original question around reskilling or retooling, on one level, that sounds like a really heavy thing, like does everyone have to become an expert in generative AI, et cetera? The simple answer is no. Even today, a lot of people talk about, you know, retooling and reskilling, but the honest truth is if someone has a solid ability to write, communicate, present themselves well in the workplace, solid critical thinking, kind of a middle school, early high school level of mathematics, science, civics knowledge, if they actually know the material, they’re actually already quite potent in–to get a career in the white collar workforce. Now, you know, we ask them for things like college degrees, et cetera, et cetera, but we know that most of the classes you take are–they’re really just litmus tests for can you do the homework, can do what you’re told, but it’s really if you retain that knowledge, you’re already ready.
And obviously, we assume, if you’re taking–getting into a career track, white collar job, that you also have basic skills using, writing a doc, spreadsheets, PowerPoint presentations, things like that. This just adds one more tool to that toolkit. So my biggest recommendation for folks is they should be–if they’re playing with it already and if they’re already using it for certain tasks, they’re already way ahead of the curve, and I think they’re going to be in good shape. And what you’re going to see over the next year or two is a whole bunch of productivity tools that make it even easier to use, and they’re really just going to be seamless with what we’re already doing. You’re writing a document on Google Docs or words–or Word. You’re going to be able to use generative AI for certain parts of it. You’re doing workflow on any of your existing pieces of software. Your email is going to be triaging your email based on importance, based on the feedback you’ve given it. So hopefully this revolution is actually going to require even less, I would say, bespoke training. It’s just about being out there, using whatever tools are out there, and being familiar with them.
MS. ABRIL: And, Michael, I know you mentioned that you kind of believed, obviously, in what some of our other speakers have said about sort of this revolution changing jobs or creating more jobs. But I’m curious if you share the concern that AI could widen wealth inequality as fewer workers are required to do the same tasks with this kind of automation.
MR. HOWELLS: I think the question as to whether that’s a risk or creates an opportunity really depends on training and investment in the kinds of–through life upskilling that we all now need to embark in to make sure that we remain competitive in the labor market.
And what we find very encouraging, actually, when we look at the data that our own machine learning creates, when we look at the way in which trending demand for skills in the labor market is projected into the future, we actually see growth in those core human skills that we’ve been talking about. It’s one of the–one of the happy consequences in many ways of the increasing automation and augmentation of certain kinds of work that’s been made available through robotic process automation, for example, as that the value of what people can bring into work becomes clearer.
And the way to capitalize on that is obviously to make sure that we’re really investing in developing those core human skills that are where people bring the most value into productivity, the most value into the outputs that any organization, whether it’s a company or an institution, is working through. So there is huge opportunity in this to really understand and invest in the value that people can bring in work if we make those tools available to people in the right way and if we make sure that people are informed by what the data is telling us, so they can make the right choices for themselves.
MS. ABRIL: Yeah. And so speaking of investing in this kind of tools and understanding, Sal, this year you launched an experimental artificial intelligence guide that mimics a writing coach or tutor to help students learn in school. What are some of the highlights of that, and what are some of the challenges you’ve seen with that program?
MR. KHAN: Yeah. And I’ll add to that. It could be a tutor or writing coach, but also on the teacher side, it can act as a teaching assistant, develop lesson plans, rubrics, even eventually help assess papers and things like that. When we went into this, we were initially a little bit cautious. We started partnering with OpenAI over–almost exactly a year ago, so well before ChatGPT came out, and this was when GPT-4 was on its way–when it was coming, and I was skeptical at first. I had paid attention to GPT-2 and -3, and they were impressive, but they didn’t have a good handle on knowledge, hallucinate a lot of facts, et cetera. Obviously, everyone knows when ChatGPT came out, which is based on GPT-3.5, as impressive as it was and is, it still had a problem with hallucinations. It still had a problem with math errors.
When we saw GPT-4, it still had some of those issues. They had been a lot that–it was mitigated to a large, large degree, but what really impressed us is GPT-4’s ability to be steerable, so to speak, and be able to take on these personas of a tutor or a teaching assistant or a writing coach and do it based on some of the best practices out there. So that’s when we started working pretty hard at it.
When ChatGPT came out, at first, I thought it was a little bit of a bummer. I was like this thing is so imperfect. It wasn’t designed for education. The main headlines everyone’s seeing is kids are using it for cheating. The school systems are banning it, and I was afraid that the baby was going to get thrown out with the bathwater. But when we launched Khanmigo–and we did it coincident when OpenAI released GPT-4 in March–it was actually a blessing that ChatGPT was out there because people were struggling with it already. They saw the power of this tool. They knew it was going to be in the workplace, but kids were using it to write their essays.
They were using it to do their homework, and so we were now able to offer something that was special purpose built for education, put some guardrails around it. All of the conversations are accessible by parents or teachers. Second, AI monitors the conversations and then can actively notify parents or teachers if something’s going on. We’ve invested a ton above and beyond the GPT-4 layer to make sure that the math and the hallucinations–the hallucinations are as low as possible and that the math is as good as possible, and that’s been a pretty interesting journey.
But we now have been testing it for about four months in mainstream schools, and the feedback has been, honestly, better than I expected. Teachers are saying, “Hey, that’s been able to answer questions that my students were even afraid to ask me, but I didn’t even–if they did, I wouldn’t have had the bandwidth to get to all of those students.” We’re having teachers say, “I spent 10, 15, 20 hours a week doing things like writing lesson plans, grading papers. This is already dramatically reducing that.”
And it’s not in any way minimizing the value of a teacher. What we’re seeing is if you can create alongside–just as if you have a great teaching assistant who’s creative, that doesn’t make your creativity any less important. When you are able to riff with another creative party, when you’re able to create together, you actually create something even better. It actually makes both parties more creative. We’re seeing that happen in the classroom where already amazing teachers who would make these great lesson plans, when they’re able to riff with Khanmigo around, hey, how can I make this lesson hook even better, or how can I improve this rubric, it’s energizing them. And it’s saving them time at the same time.
So we’re pretty excited. We’re working on now ways that–you mentioned writing coach, where a teacher can develop assignments and rubrics with the AI, with Khanmigo, assign it through Khanmigo. Students can do the writing assignment with the AI, not having the AI do it for it, and then the important element is the AI can report back to the teacher and say, “Hey, look, I’ve been working with Sal on this. He’s struggling. Michael has finished his outline, but he still hasn’t gotten quite this far yet. And, you know, Mary just copy and pasted something in here. That looks a little bit shady. She might have used ChatGPT for that. You might want to talk to Mary about that.” So we actually think not only will it be able to mitigate the risks of cheating, we think it could actually supercharge kids’ ability to write more, get more feedback, and give more information to teachers.
MS. ABRIL: I’m curious what those challenges are, though, and, you know, when you give AI sort of that ability to make those judgments, obviously, like you said, the error rate, you guys have gotten it pretty low. But I wonder, are you seeing some unexpected errors, some unexpected challenges come out of this?
MR. KHAN: The biggest one is actually because the AI can be very engaging, it can be a distraction sometimes. And it’s not like a bad distraction. Like we have activities where students can talk to a simulation of a historical tech character or they can talk to a literary character, and what we’re seeing in those situations, that could be a really powerful thing at certain times, but you shouldn’t be doing that if you have a writing assignment or if you have a math assignment to work on. So we’re looking at ways that teachers can turn some of these things on and off. That’s been the main issue.
We’re also working on memory so that you don’t have to keep telling the AI what you’re interested in or the type or the tone that appeals to you so that it can remember what matters to you, so to speak, in previous conversations.
MS. ABRIL: Michael, I’m going to jump back to you now. Earlier this year, Pearson published a study in partnership with Google that looked into what skills workers are prioritizing to land their future job. As you see a new generation enter the workforce, what’s your advice to them as they navigate a new space with AI?
MR. HOWELLS: That’s a great question, and I think–well, what that research showed us, which was really encouraging, was the point I made earlier, which is that the hot skills today and the even hotter skills tomorrow, which our data and our analytics show us, but also the work that we did to survey thousands of learners around the world, thousands of employees around the world, showed that they had understood this as well. It is the value of human skills is really where the future is taking us, and some of that is because those machine learning-based tools, those new technologies that are coming on stream now are so accessible to you, without necessarily a particularly high level of technical competence, that they can augment work in a way that liberates people to really make the biggest contribution they can through what people can uniquely do.
So trusting in the value of people and understanding your unique attributes and potential and capabilities as a human is really the best advice that I could give to somebody who’s coming into the labor market now.
But I think one of the biggest changes–and this is something which if you spend time with any young people in work today, you’ll see this very evidently–is that you really want to take control of your own career. We’re way past the time when people join the labor market and they think this is my career and I’m going to stay in it for the next 30 years or 40 years. We’re way past the time when people think that management of my career is going to be provided by, you know, a well-resourced HR department that’s going to look after me and guide me through this. There are great HR departments and companies that really care and invest in their people. Of course, there are. But as in all areas of the consumer world, the key really is agency in yourself, is the ability to take control of your own destiny and make the right choices, informed by data, informed by the kinds of insights and recommendations that these AI tools can provide to you.
So my advice to everybody is to really learn about what those tools can do for you, really embrace them, become smart about their opportunities and their limitations–and there are limitations–and recognize ultimately that what you need to do is invest in the value of yourself in a way that works for you, and it’s going to take you where you want to go through your career.
MS. ABRIL: So we only have a few minutes left, and I want to get two more questions in as quickly as I can, just because we did get some audience questions. So I want to make sure to get those to you. Sal, let’s start with you. Philip Ray of Germany asks, “How will AI affect public K-12 education in the United States? Europe? Africa? Asia?” Any thoughts on that?
MR. KHAN: Well, our hope is that it can start to fill in some of that gap to provide more support, more personalization. I think when you get out of the wealthier countries, there’s some real cost issues. Obviously, our mission, free world-class education for anyone, anywhere, the marginal cost of delivering a webpage is fractions of a penny. But right now generative AI is actually quite expensive. It’s about $10 to $20 per user per month, if you’re using something like GPT-4. So we’re trying to figure out ways to get that as accessible as possible. I suspect in the coming year or two, those costs will go down by at least an order of magnitude.
MS. ABRIL: And, Michael, to jump to you, we have a question from Stanley Cohen of West Virginia, and he asks, “How will or should the educational sector, high school and college, train for this upcoming transition to AI-based careers?”
MR. HOWELLS: Oh, that’s a great question. Well, think, you know, the great thing about so many higher ed institutions, of course, is that they do amazing research into this area. So they are leading the way in many ways and helping us to understand how to respond to these changes. I think the key is in–I think there is a role for governments here and educational policymakers in helping to make sure that they’re keeping pace with these changes and they’re making the kinds of resources, the guardrails, and above all the funding needed to allow the sector to respond to what are going to be really seismic changes in the nature of how people educate themselves. But I think the trend that we’re going to see running through all of this–and this has come out a lot in the conversation already–is the way in which these tools are empowering individuals to take more control of their own educational careers and then their professional careers, and that is not something that we’re going to be able to put back in the box. You can’t un-invent the calculator. It’s a fact.
And so really, I think the key is going to be investing also in the training and support to professional educators to make sure that they’re empowered and understand how to work best with these trends in a way that optimizes all of those upsides there, which are to, I think, liberate teachers to provide more of the really high-value inputs that they as teachers, as coaches, as people who can provide the right kind of personal support and interventions that students require, which, you know, automation of some of these traditional tasks that have been taking up a lot of their time might enable them to do.
MS. ABRIL: I think that’s a great note to end on. Thank you for all your insights, Sal and Michael. Unfortunately, that’s all the time we have. So thank you both for joining us.
MR. KHAN: Thanks for having us.
MS. ABRIL: And thanks to all of you for joining us as well. To check out what interviews we have coming up, please head to WashingtonPostLive.com to register and find out more information about all of our upcoming programs.
I’m Danielle Abril. Thanks again for joining us.
