Explore the impact of AI on our workforce and skills system, and learn from past waves of industry transformation.
Moderator: Misha Schubert, Chief Executive Officer, Science and Technology Australia
- Dr Enrico Coiera, Director, Centre for Health Informatics, Australian Institute of Health Innovation
- Dr Audrey Lobo-Pulo, Senior Manager for Public Policy and Economic Graph, LinkedIn
- Jeremy Howard, Founder and Researcher, Fast.ai
Text: TECHTONIC 2.0 – Australia’s National AI Summit – 18 June 2021. This session will commence shortly.
A countdown from 5 minutes begins. To the right, a collage of industry images, a woman piloting a drone, steelworks, a coal truck, computer processing chips.
The broadcast transitions to a webcam feed of a woman wearing glasses, she appears in a framed blue window and wears a white jacket.
Misha Schubert: Yuma, hello in the language of Ngunnawal people, the traditional owners of the nation's capital, the land on which Canberra is built. I am Misha Schubert, the CEO of Science and Technology Australia, which is, of course, the peak body for the science and tech sectors across this country. Thank you so much for joining us for a terrific discussion today with our expert panel wrestling with the question of how Australia goes about preparing our workforce for an AI enabled future. We want to begin by acknowledging the country on which we meet across Australia today and paying our respects to traditional owners right across the country and connecting ourselves as we do in Australia with the long story of that country and the deep knowledge systems across 65,000 plus years of continuing culture since the rich long story of Australia. And when we do that, we invite all Australians to connect with the story of the longest continuing cultures anywhere on the planet and what a source of inspiration and pride that is for all Australians today. We're going to have a terrific discussion today with a fabulous expert panel. Dr Audrey Lobo-Pulo is, of course, LinkedIn senior public policy and economic growth manager for Australia and New Zealand. She's also part of the World Economic Forum's expert network on the future of the digital economy and society, civic participation and the future of government. And she's a physicist with a PhD and has a masters in economic policy. So she's incredibly well qualified. She's also spent some time working in senior roles in the Australian Treasury, looking at deep impact of major trends that will affect our community and economy. Jeremy Howard, the founding researcher of fast.ai. He's also a data scientist, researcher, developer, educator and entrepreneur. Jeremy is, of course, the founding researcher at fast.ai, which is dedicated to making deep learning more accessible to more people around the world and is a distinguished research scientist at the University of San Francisco.
He's also the chair of WAMRI and is chief scientist at platform.ai. He's founded a number of start-ups ai companies including Kaggle and Fastmail and a series of entrepreneurial ventures. So he'll be bringing his expertise to this discussion, wearing all of those hats as well. And our third (UNKNOWN) Panellist is Professor Enrico Coiera, who is of course the director of the Centre for Health Informatics at the Australian Institute of Health Innovation. He's a PhD in computer science and artificial intelligence and has a research background in both industry and academia. So can speak to some of the big shifts we gonna see across our society, economy and workplaces with the benefit of both of those perspectives. So today we gonna have a really terrific conversation about, how do we prepare Australia's workforce for an AI enabled future? And hearing the minister's introductory remarks, of course, he noted some of that early work that is being done to anticipate and forecast the big changes that we're likely to see within jobs and across jobs in our economy and society. He referenced the fact that large percentages of the jobs, as we currently know them, might soon be transformed through the development of AI and machine learning technology. So I wanted to start there with that invitation for each of our panellists to give us their quick take on, how AI is already starting to reshape and change the jobs of Australian workers today. Jeremy, did you wanna kick us off, Jeremy?
Description: The broadcast transitions to four participants in separate framed blue windows. They each sit centred against a white backdrop. The man in the bottom right window begins.
Jeremy Howard: Sure. Yeah, there really has been a significant step change in capability around what AI can do in the last few years, driven by deep learning and neural networks, getting to a point where it can provide superhuman performance in a number of areas and is allowing us to do things that couldn't be done before. This is in the early days of significantly transforming what the job market looks like and and indeed transforming society. It's you know, Australia is still a bit behind on terms of that transformation. So I spent the last 10 years in San Francisco and we've seen it more rapidly there, but Australia is now starting to catch up. We've got a lot of very strong AI capabilities and competence here. And what I've been saying is AI moving into lots of existing jobs rather than necessarily creating new jobs, but people who are bringing AI into their organisations are becoming dramatically more productive and and really increasing their competitiveness.
Misha Schubert: Terrific opener. Audrey, you've got some insights as well from the sort of data we're seeing in job advertisements and people posting about the nature of their jobs changing, haven't you?
Audrey Lobo Pulo: Yeah, Misha, you know, it's been so interesting, some of the research that we've been doing at LinkedIn, we partnered with the Stanford University and put out an artificial intelligence index report. And the results for Australia are really very interesting in terms of the AI hiring rate, which is, you know, the demand for AI in Australia. We we're kind of like on average, we've seen a doubling of the AI hiring rate in Australia from 2016 to 2020. So it's certainly increasing. But what's been really interesting as well is being compared globally, Our AI penetration skills across different occupations isn't so high. We're actually below the average. So even though we're seeing a lot of growth in AI hiring rates, they're not, you know, across or like across a broad range of occupations. And so that's something to unpick. You know, as we sort of think about the AI national strategy, where is that AI skills penetration deeper than others? And what does that mean? And I'd love to sort of also, during the conversation, have, think about the equity side of AI and who's impacted by these skills and who isnot represented. We've also noticed a big gender gap. So that's that's another one to untangle.
Misha Schubert: Absolutely. And we're gonna pull back to those, come back to them as we unpack this discussion. Enrico, did you have a few opening thoughts as well on that question of what we're already seeing now?
Enrico Coiera: Yeah, I might just talk about the sector I know best, which is health care. And I always like to remind people that this is not the dawning of the age of AI health care. We've been at it for about 20 years. If you'd gone into a laboratory in a hospital 20 years ago, there would have been lots of people in white coats running around. Today, you just go in. There are these quiet machines that hum which are driven by AI, albeit more old fashioned AI than we have today. So so this is not a new thing and it's already underpinning a lot of the business processes of health care. But what I am seeing now is lots of new tools appearing, seeing new business models disrupting the way people work and I'm also seeing new problems we have to deal with generated by AI and I won't spend too much time now just be going, but just to give you a sense, the FDA in the US about 12 months ago approved 80 new AI devices.
The EU was about 100 and you look over the last five years, about 300 commercially approved AI devices in the hands of people being used routinely. So it's already out there. In the UK, we're seeing primary care, general practice, which you'd think would be the most human of disciplines in health care, being heavily stressed in terms of business because of so-called digital first companies that start with AI driven symptom injectors. And just talking about COVID and the impact of AI there. You know, AI drives our social media algorithms that prioritises exciting news. An anti-vax sentiment is one of the most exciting things to share. So one of the big challenges we face right now in the middle of the pandemic to get people vaccinated, we're fighting against the algorithm, as it were. So new opportunities, but new challenges.
Misha Schubert: Absolutely. And that that really important role of combating disinformation, misinformation and showing that people have access to accurate, timely, scientific and scientifically grounded material is so important in a moment like this historically, isn't it? Let's maybe dig in a little bit more to that issue, that Audrey you opened up for us, which is that idea that perhaps we're seeing a sort of multi-track patent happening across our economy, which is that some sectors rocketing ahead with the uptake and incorporation of AI to transform jobs and their industry sectors and others perhaps not going at that at the same sort of pace. I noted that Data61 from CSIRO had estimated also that they're thinking we're gonna need an extra 160,000 additional AI specialist workers across Australia by 2030. So you've probably got those sort of two threads, haven't we, where we've got the need for a specialist AI workforce and the sort of need that we'll have as a country to train and build that specialised workforce, but then a need to prepare the wider workforce as a whole for the changes that they'll start to see in their jobs. Any of you wanna lead us with a few thoughts on that.
Audrey Lobo Pulo: I'm happy to step in. You know, I think, you know, it does feel like it's a bit daunting Misha, you know, how how do we sort of transition our workforce into becoming more digitally savvy? And where are these, where are these people coming from that are gonna fill up these gaps? But there is a lot of hope and there's a lot of heart and some of the research we've been doing as well. So one of our most recent pieces has been with the World Economic Forum, and we put out some results in the Future of Jobs report, which is recently released. But it suggests that, you know, transitioning into data and AI is not as difficult as people might expect it to be. We found that particularly in our data, most of the people that transition into that, into that particular skill set came from job families that had that weren't anywhere near that particular scale. It was actually easier to transition into data and AI skills than it was to transition into people and culture or engineering. So we need to sort of think about this perception that we have that transitioning into these these kind of skills is going to be really difficult because, you know, there are many steps that we can take, like micro learning and upskilling that may not be as daunting as what we think.
Jeremy Howard: Yeah, if I could add to that, I mean, Audrey's spot on here and I, I, I've possibly trained more of these new AI workers than anybody in the world. Our course has had hundreds of thousands of students. And it's been specifically targeted at people who, you know, they might be doctors or they might be paralegal or they might be activists or whatever and giving them AI skills. And, you know, when we started back in 2014, this was very speculative exercise.
People really thought you need a math or computer science PhD. You need, you know, years of study at, you know, at a university that specialises in AI and it actually turns out that's that's not the case. You do need pretty strong coding skills, but lots of people around the world and Australia have that. And actually that puts Australia in a particularly good position in terms of the skills we have. And we found with, you know, a few months of of training, which can be done entirely online, you can actually reach a world class level. And one thing I just wanted to mention is, whereas I agree with Enrico, that AI has been around for a long time in medicine. You know, I also wouldn't wanna understate the huge step tech change in capability that's been in recent years, which is actually why we need a high level of urgency in this country, because where we're not actually keeping up right now, we really could be. We have the skills and resources, but we're not really keeping up with a step change in terms of deep learning and neural networks. I don't see as much activity here and my best Australian students are generally going to work for American companies because we you know, we've still got work to do to really harness this resource in Australia.
Misha Schubert: Yeah.
Enrico Coiera: Yeah. I think actually, Jeremy, I do agree. If you look at where AI is most embedded, it's in those highly automated or process driven parts of health care, say, for example and what's very clear in the last two years, as I sort of hinted at, is that the new AI technologies we're seeing are getting involved in areas you'd never expect, like, you know, helping people with online screening and counselling from bot's, radiologists. Yes, you'd expect them to use the positions you wouldn't have expected necessarily them to abuse (UNKNOWN). So the real challenge is to understand what is the skill set that people want. If I look at the early adopters, things like the College of Radiology are obviously transforming how they train to be radiologists. And they're really a step ahead, they're global leaders, I would say, but the remainder of health care is not, is not yet engaged and probably doesn't even know what it should do. Should it learn to code? Should it go to one of Jeremy's courses? Should they rather learn how they...
Jeremy Howard: Of course they should.
Enrico Coiera: Yes, safe and effective uses? So there's lots of challenges. And I think just getting getting clarity on what are the skills I'm going to need to do my job is probably the hardest part. I don't see any challenge in terms of capability of the people to learn at all.
Misha Schubert: Well, and what is our best guess about the kinds of skills that will be foundationally important across the next decade in employment? What do we think people are going to need to add to their set of foundational core skills as workers to keep pace, to remain employed?
Jeremy Howard: The next 10 years is a long time and I think that's gonna change. I think at the moment.
I've been saying that people who learn to code and learn to actually train AI models themselves become extremely, extremely in demand and can make lots of money and make a huge impact. Having said that, I don't think that's gonna remain the case throughout the next 10 years. I think increasingly the technology is gonna be embedded inside what we do and there's gonna be a lot of kind of specialist work in in, for example, in kind of medical labelling enhanced by kind of AI feedback, for example. So I think that, you know, at the moment, I'd say people should be focusing on kind of code and actual kind of foundational AI skills. But, yeah, I think that's gonna become more and more embedded over time and the kind of the code side will become less important in a few years time.
Audrey Lobo Pulo: I'd love to add Jeremy on that, too. I think that human augmentation piece is gonna be so interesting and so important. What are the complementary skills and digital skills? Which sort of soft skills might be really important as we go ahead? And is communication gonna be one of those that feature? And how do we increasingly focus on improving that augmentation piece between that digital literacy piece and also that human side?
Misha Schubert: Absolutely.
Jeremy Howard: And as you mentioned earlier, Audrey, also, how do we ensure a diversity of people having these skills? You know, so that's been a real focus for us, is we we have had a number of diversity scholarships. You know, we've tried to increase the number of people from underrepresented groups. But as Audrey mentioned, both in Australia and around the world at the moment, this is not a diverse group of people and that's creating very genuine, direct problems in terms of the products and services being created, not working as well and solving the right problems as they as they could be.
Misha Schubert: We're gonna issue a call out for our fabulous audience to put their questions into the Q&A function in the in the platform so that we can try and weave, work our way through some of the questions you might have, and we'll try to get to as many as we can. One question that's come already through that is what can Australian industry or employers be doing right now to start to upskilling and prepare their workforces to stay competitive as we start to see these shifts come on at greater play?
Jeremy Howard: One thing I'd mention in terms of what not to do would be, don't outsource this. A lot of, I've noticed a particular tendency in Australia for bigger companies to want to hire in some big American company. This is not the right approach because it's, you know, this is something that's gonna become more and more of a core competence and one shouldn't be outsourcing core competencies. It means you're not learning. You know, in my opinion, the first thing to do is to identify your, you will have people in your organisation already who are actually probably pretty good at this, and you didn't even know it because they've been learning an online courses and practising in the evenings. So figure out who you've already got and connect them together and then give them what they need to to help push your organisation forwards.
Enrico Coiera: I'd like to reinforce that point so much, Jeremy, which I like to talk about algorithmic sovereignty. In other words, we do need local control over the capability to develop and modify algorithms. My big fear is that we will revert to the classic Australian model of being exporters of data as primary producers, and then we import value added product, which is somebody else's algorithms.
Jeremy Howard: Yeah.
Enrico Coiera: And again, in my world, it just doesn't work. If if I get an algorithm developed for X-ray interpretation in China or the US or Europe and I bring it to Australia, it's gonna need recalibration and I'm gonna have to fix it anyway. And so you're not gonna outsource those skills. So this is core to the way future work will be done. And this is not a commodity product to bought off the shelf like a spreadsheet tool, you really need the capability. So now is the time to start with a pilot project and obviously learning a piece of fruit. Keep going, learn, amd as Jeremy says, you will find champions.
Jeremy Howard: I mean, this is already happening. Isn't it Enrico, I mean, in the US, the top five largest companies by market cap are all digital. They're all software and hardware companies and they're all been made in the last few decades. You know, in Australia, four of our top six are banks. One of them is mining. You know, we've already been missing the boat on this on this shift. And I think AI is gonna really accelerate it.
Enrico Coiera: Yeah, I think one of the things I really enjoyed in the previous session, we're talking about things like manufacturing and aerospace and at least in health care, it's not a design space. This is an organic, evolved mess, which has just evolved over 200 years. And so therefore, it's not even about algorithmic design. We need to have the skills to understand where to place the algorithms into the workflows, how how to do no harm to the system that's around it. And those so-called translational implementations skills are so important and often there's such a focus on building the algorithm. We forget that fitting task of taking the work, taking working algorithm and making it add value to a business process is a skill and a special skill. And so my I guess my optimism is that's something that we're probably very good at here in Australia and it could become a national competence. It's gonna be hard to beat Apple and Google in machine learning algorithm design, but they're not experts in real world business processes, whether it's manufacturing or health care, that's our skill set and we should go for it.
Misha Schubert: Really important points about sovereign capability and about having that capability also in-house within employing organisations as well. Just further to that, let's unpack that a little bit more for out for our audience. In terms of the kinds of investments, Jeremy, what kind of length of foundational courses are you offering and what, people might get intimidated sometimes by thinking, I've got to put people through some long range comprehensive training program here too to stand up a workforce right there.
Jeremy Howard: Yeah. So our courses there, they're all free, they're all no ads. It's all like just an altruistic exercise. And to get to a point where you're competent so you can build useful models and also kind of assess other people's models, it's about 70 hours of study and depending on your background and to get to a point where you can kind of design world class models that are quite innovative, it's about twice that. And, you know, one thing I'll say actually is there are a lot of Australians who are actually significantly pushing forward core AI capabilities. So, you know, a lot of them are Australians who are at Google in the US or at Facebook in the US, to be fair. But, you know, there's a lot of Aussies who are actually genuine world class experts at cutting edge AI research. So, you know, I absolutely think we can compete there. So, for example, me and my students a couple of years ago won an international competition in AI to create the world's fastest AI training algorithm. We beat Google, we beat Intel. You know, these guys are not untouchable. They're pretty good. But Aussies are pretty good too.
Misha Schubert: Excellent Australian spirit of we've got, we've got you on later. Audrey, are you seeing shifts as well in the sort of entry points for how much of an investment people need to make to just getting started down the pathway of training their workforces?
Audrey Lobo Pulo: Yeah, I think I think one of the things we tend to overlook at the adjacent skill sets to AI and how we can sort of transition people from those across. My one of my big sort of passions is, you know, thinking about that gender equity piece. And what we found is, you know, women are very underrepresented, underrepresented globally in AI, but more so in Australia. So we're actually below the global average, which is, you know, really disappointing, but you know, it's something that we really need to focus on. Where women do have a greater representation in the digital side is in areas like genetic engineering or human centred design, nanotechnologies, and those technologies are very adjacent skills to the AI skills. So how do we how do employers and business leaders look to those adjacent skills and start employing people with those skills as part of that AI pipeline? So what are we doing on that front? So the adjacent skills are one of those. The micro learnings is another. During COVID, you know, LinkedIn partnered with Microsoft and offered a range of free, accessible, LinkedIn learning modules which you just gave people a bit of a foot up on that digital skills piece. And so I guess what I'm trying to say is you don't need to do a three year degree to sort of start on that journey. You can start in micro steps.
Jeremy Howard: Well, another area actually is very important is AI and data ethics, which is an area where, you know, there's a much broader range of people that are in that space. A lot of the best researchers in that area are black women, actually. And in Australia also, we have some of the world's top people in that space. So, you know, that's a kind of an interesting sub area, which is becoming increasingly important and I think Australia has very strong capabilities and we also have a, you know, relatively diverse group of people working on it.
Misha Schubert: Absolutely, and I was fortunate enough to go to the Women and AI awards at the end of March this year and see the breadth of terrific women leaders and entrepreneurs in the AI sector here. So if people are looking for diverse, fabulous female talent, to pres into service of further inspiring role models to build that pipeline and stronger diversity, there's a terrific wealth of inspirational figures out there. Another question from our audience. Is there more value in non-coders coming and learning AI or machine learning than in upgrading the existing software engineers to use and think of directions like deep learning directions? Is the question.
Enrico Coiera: I can think of my own experience, you know, running running a translational research centre, and the easiest path actually is to take the computer scientists who have got decades of technical expertise, and I can train them on a narrow task very quickly. They can become experts on a particular kidney disease and off they go very quickly, but takes a long time to take a health care person and train them to be as good in use of technology. So there will always be open arms to those who want to make that journey, and they are critical to translational people. But for most people, I would say it's creating a group of people who are sufficiently literate in both universes, the applied and the technology to be the bridge. I actually think when I look at, you know, what we have here locally, I can see, as Jeremy says, great computer scientists, also great people who who work in and again in the clinical space, but we have a hole in the doughnut, which is those translational people, and they are the enzyme. They catalyse everything. And so we we kind of need those folks.
Jeremy Howard: And it's important, I think, actually, not to think of it as a dichotomy of like coders and non coders. I know a lot of brilliant doctors and brilliant. I mean, I'm more familiar with the medical space as well, brilliant radiologists who can also code. And they're the ones who everybody wants to work with them because they actually understand the needs. They understand the constraints. They understand the operational integration. They know what problems are important to solve. They know when a solution actually doesn't make sense. And to take a coder and train them to be a radiologist (CHUCKLES), that's probably ten years, right? Or else taking a radiologist who can code a bit and making them into strong deep learning practitioners, more like 100 hours of work.
Misha Schubert: That's a really good point. And that piece of the EQ skills, of thinking about building the social licence and the uptake and confidence in technology, looking at the human use is really important in that. Another question inviting you all to sort of cast forward, well down the track, perhaps. One of our audience would like to know, they think technology is going to play an increasing role in the incremental changes we're likely to see within existing jobs. But ultimately, do you think that, with time, AI would be capable of doing all the jobs that we could ever require? Wow, we're going to put ourselves out of work in the midst of this answer, aren't we?
(CHUCKLES) Or not.
Enrico Coiera: From my perspective, I think I look at the business of healthcare. And there's so much unmet demand that even if we're getting better at meeting that demand, we're just going to have more work to do. I can see no future in which we meet all people's healthcare needs.
So, I don't think that's going to happen. And all of the models I see people comfortable with, say, for the next 15 years are going to be hybrid AI-human jobs. Some people call them centaurs. I don't know if I like that term or not. I keep on imagining we're going to build a push-me-pull -you and it's a disaster. So, I think the human side of business is critical. Absolutely, AI will do things that humans aren't good at. But, you know, I've yet to see anything in the pipeline that would suggest to me that we could sit down and have a conversation about end of life, or what are the best treatment options for you, if you're going through IVF - those aren't necessarily machine conversations. So, I don't think those soft skills will ever disappear. I actually know work will change. But what I often do when I talk to medical students, I'll say, look, you'll graduate in a few years time. You'll probably complete your specialised training in ten years. And then in your early 40s, you will be at the peak of your career. That's 20 years from now. I cannot tell you where technology will be. I cannot tell you what your job will look like. So, the one thing I'd tell them is to be open to radical change in what you do and how you do it. Don't expect you won't have a job. Just expect you might have some very different jobs over the next 30 years.
Jeremy Howard: There's less than one-tenth the number of doctors we need in the developing world. And estimates are that it will take about 300 years to fill that training gap. So, the idea that we're going to run out of things for people to do, in the medium term at least, seems pretty unlikely. There'll be plenty of jobs which we're going to need less of those people than we have now. But it doesn't suggest in any way that we're going to run out of things that humans can help with.
Misha Schubert: Audrey, any additional thoughts on that one?
Audrey Lobo Pulo: Yeah, this is one of the topics that I really love. And I think there's a lot to be said for AI. But having said that, there's also a lot to be said for humans, like our ability for compassion and empathy to recognise inequality, and feel strongly about making the world a better place is something that is hard for a machine to do. And so, I think there's always going to be a space for humans. And one of the things we do best is evolve. So, the Industrial Revolution brought on technology and we don't need to do a lot of things that we used to do many hundreds of years ago. And so, we'll find the next thing to do is what I kind of feel.
Misha Schubert: Indeed. Another question from our audience. For companies who perhaps have few employees who are AI-inclined and might not have a large amount of resources to develop an in-house AI team, what might be the best approach to harness AI capability without sacrificing sovereignty? Would a collaborative model with external AI companies potentially headquartered in Australia be a pathway they should be thinking about?
Jeremy Howard: I would say do it anyway. Lots of small companies have AI capabilities. Just one person goes a really, really long way. And if you are going to partner with other organisations, it's even more important that you have at least one senior person in your organisation who very deeply understands the technology, because honestly, there's a lot of snake oil out there.
Otherwise you're going to get taken for a ride. And you're not actually going to be building up this competence. All you're going to be doing is teaching some external consulting firm how to do things better. And they'll be using your data to make things better. And then they'll be selling that to your competitors for less than it cost you to build it. So, I don't think that's a strong, competitive way forward.
Misha Schubert: Additional thoughts on that? No? Will AI drastically change the overall composition of our labour market? What should we be preparing for is the next question.
Audrey Lobo Pulo: Gender equality.
(CHUCKLES) If AI is going to start dominating the future of jobs and women are vastly underrepresented, what does that mean for women's jobs and what does that mean for wage equality? And how do we bring about that balance? So, absolutely, AI is going to compound the issues that we have unless we start doing more to encourage under-representative groups to also partake in this world of AI.
Misha Schubert: It's a really important point. We've already got a pretty persistent gender wage gap across the economy as it is. And when we look at the STEM discipline - science, technology, engineering and math - technology and engineering are the places where change and where progress towards something closely resembling equality has been slowest and hardest. So, there's a there's a big task here, isn't there?
Jeremy Howard: There's also this widespread, false belief that the reason for that is because there's not enough of a pipeline. So, a lot of people are very focused on teaching girls at school to code and stuff like that. But when you actually look at it, a lot of brilliant technical women leave the workplace because there's just a lot of issues with toxic jobs and with that whole career. And so, fixing this is not just about teaching girls to code, but it's about fundamentally ensuring that the workforce is valuing and fully utilising all the people that are there.
Misha Schubert: And it's clear-eyed business case to do that, isn't there? It's not just a important principle. Enrico, did you have some further thoughts on that, how we really make further progress on the diversity pace within the workforce?
Enrico Coiera: I think it has been said that we really just need to improve the nature of the cultures of the organisations that we have, and just really root out these old biases and perceptions. And computer science, historically perhaps, had been really male-dominated. In the healthcare space the gender balance is very different but the skill sets are also very different, too. So, maybe we see on our shop floor less imbalance, but I don't think across the whole thing it's the case at all.
Misha Schubert: Getting to critical mass is also important. Transformations in cultures happen when you hit that critical-mass tipping point - and so, recruitment exercises that actually actively go out. To build the critical masses there inside, I think, are often most effective and have most durable results. What sort of changes are we going to see - I know, Jeremy, you brought up that issue, the discussion around the pipeline of school students and young people into the foundational skills of math and coding. But what sorts of skills are we going to need our school-level education system to start to really emphasise and extend people's learning in, to prepare that generation and beyond for the bigger changes ahead?
Enrico Coiera: I think, again, for me, talking about health care - but they just ENRICO COIERA: I think, again, for me, talking about health care - but they just need to teach people to be safe users of the technology. It's already the case that you can download AI-driven symptom checkers to tell you what might be wrong with you. But people need to be aware that these are limited, flawed technology. If you're feeling crook, go and see a doctor. So, there needs to be, I think, an understanding of how wonderful these things are, but also to make it transparent about how you decide to base your life decisions, based on these prompts you're getting. We often say things like, but AI is already here, it's shaping your Twitter feed and shaping the results you see on Google. But that's all behind the scenes and non-transparent. And the fact that two people can have different Twitter streams, just based on the implicit models Twitter has of you, is problematic if you don't know that's happening. So, I think this is probably a whole-life thing we need to be talking about, even how we teach our children about safety, tolerance of each other. There's a big push right now around diversity and safety. This should be another thing that is just bread-and-butter, as part of education.
Jeremy Howard: And in the end, a lot of that is about understanding the actual capabilities and limitations of the technology. And one of the things we found - at least for me, I don't know if this is a shortcoming of me or a general issue, but I find it very difficult to get students to really understand that, if they don't understand a reasonable amount of the actual technical foundations of what's going on. And if I think back to my schooling at high school, the kind of stuff I was learning and I think it's still the case is at a fundamental technical level. It isn't necessarily that well-aligned. I'd love to see high school students in math learning more stuff like linear algebra and statistics and machine learning and stuff like that. So, then they can go to university and study topics that are going to directly help them to understand AI principles. I feel like even our school curriculum was created for a different age in some ways, and maybe some of those important, foundational, technical skills are missing, still.
Misha Schubert: We've also seen a bit of a comparative slide in our performance at that sort of thing, at the seismic levels at which we measure math and science skills across the school years, relative to some of our big economic competitors across the world, as well. So, there's clearly further work that we need to do together on that front, as well. Another question from one of our audience members is what role could vocational education and training and that sector play in skilling this emerging or existing workforce. Thoughts on that?
Enrico Coiera: Yes.
Misha Schubert: Excellent choice, better than the alternative.
Jeremy Howard: I mean, that's where most of this is happening, right? Audrey's talked a bit about kind of micro-credentialing and micro courses. Like, courses are all online. Nearly everybody taking them are at work. I mean they're not necessarily taking them at work, but in the evenings, on the weekends. I think increasingly a lot of the skills people are building up are after they finish school. And certainly, this is a skill which is very well adapted for that.
Audrey Lobo Pulo: I wanted also to give a bit of a shout-out to GovHack and some of the other initiatives that are out there, where young people can form groups, get together and work on a project, use their coding skills and learn from each other. So, it's great to have that cultural piece within the country as well, where people come together and focus on the digital side and the data side, too. So, I think that's a great positive.
Misha Schubert: Yeah, and that idea that you're constantly dipping back into your professional education and continuing education, rather than - I've acquired my formal education and now I'm off into the world of work, and not referencing back to that upskilling as a continuous process.
What business collaboration opportunities do we see, is our next question, in the potential application in the community sector and not-for-profits? So, things around the use of AI technology in engaging communities, citizen learning, democratising, access to technology.
Enrico Coiera: It's such a big area, isn't it? Look, I think most people in this space now are used to discourse around ethics and biases. If we're talking about social justice, people need to be aware of the risks in terms of algorithmic decision making, which I'm sure I don't have to explain to anybody here. So, I think, what are, to the community, niche issues within that community probably need to be brought out more widely are like citizen juries, people debating what it means for them, what we should do. So, our challenge probably is that there's still not enough of a core group of folks in the AI space to do all these things. But I agree with Audrey and Jeremy that the training required to be sufficiently competent to go and help is not that onerous. It's just endless, ongoing micro-credentialing, so we can do that. In academia, I think there's huge willingness to engage. I've got colleagues in legal faculty who are interested in the impact of AI, on ownership of IP, who could own an AI, in terms of justice, et cetera, et cetera. So, I think that the whole social impact of these technologies is unfolding slowly before us. And it's very hard right now to even know what all the things will be that we'll be talking about in five years.
Jeremy Howard: And I love it when I, quite regularly, receive emails from professors who are in some totally different area. And they'll just email me and they'll say, like, oh, I'm in geosciences or material science or law. And I took your course and started doing deep learning and here's the things that we discovered. It's really cool to see people retraining, even at this very senior research level. And you're now seeing AI find its way into research in all of these different areas, as well as into industry.
Misha Schubert: It sparks a thought about that's happening obviously in a kind of organic way at the moment, on a piecemeal basis. But do we need to do more, in more concerted ways, to smash these silos of discipline, operational knowledge, and bring people together across disciplines to see how they can embed this expertise in all the disciplines?
Jeremy Howard: We do. One of the things that I really noticed with our students at whatever level, whether they're professors or undergrads or whatever, is this current approach is very fragmented. And so, often these people will be the only person in their university department or in their organisation who is learning these skills. And so, they don't necessarily have mentors or people to bounce ideas off or champions at work to help build new products and services. It's tough when there's a new technology coming through, the early adopters. They need community and they need champions. They need mentors. And so, this current fragmented approach, that's one of the issues, I think, that comes out of it.
Enrico Coiera: It's interesting to watch smashing things, but COVID has been a big, big smash through many things. And two years ago, the idea that we would have nationally funded telecare was a no-go zone. Now it's essential for delivering care in the world of COVID. And what we're seeing is telecare plus AI is actually disruptive of work in the healthcare space. I can see patients with the AI app there helping me. It's autotranslating what we say and creating notes as we go. I'm getting diagnostic hints. I'm getting treatment hints. This is a totally new way of work. So, I think maybe we're moving into an age where there are crises, whether they are climate crises or another pandemic, where we just can't get by without doing these sorts of things. And so, maybe the old world, where it was a decision to get engaged in AI 'cause it's cool - now, it's no longer a decision. It's a necessity. And whether that penny has dropped or not now, I'm pretty sure that in 12 months' time, it will have.
Misha Schubert: Do we need a formal, structural vehicle to bring some of these emerging capabilities together and really help, particularly for small businesses that want to build this capability to connect to this world of expertise in a more concerted way?
Enrico Coiera: Maybe that's part of the goal behind the national AI centre that's being proposed. And the capability centres, you can imagine - if it was the healthcare one, then that would be the place where industry could come to meet with academia. We already have organisations like the Australian Alliance for AI and Healthcare, which is industry and health services providers and academia. So, there are already attempts to bring things together.
But in the end, resources are necessary to make these things cook, so whether they come from industry or government. But you're completely right. We are too small a country. We're too fragmented. If we don't come together, we're not going to get that critical mass we need.
Jeremy Howard: One of the really cool things in this new strategy is this Next Generation AI Graduates Program. It's small, but just to be one of those couple of hundred people, they're going to be a cohort. That's going to be a community of people who are learning and building and working together, but from all over the country and from lots of different industries. OK, 200 is not 160,000 but I mean, they'll see this. And I think once you - big communities form from small ones. We have an online community, through fast.ai, of about 40,000 AI practitioners now. And it started with one person - me, you know? And then one person joined and then two people joined and it grew from there. So, yeah, I think it's something that - we all need to do our best to connect people and then to create spaces to allow those people to connect to other people. And hopefully, we can build up a really strong community in this country.
Misha Schubert: Yeah, learning as a cohort often does do exactly that. It accelerates that sense of a community of practice and learning, doesn't it? And Audrey, it strikes me that some of those social media, professional networking tools are also really coming into their own now, in a way that connects people with expertise and interests that may not have found each other a couple of decades ago.
Audrey Lobo Pulo: Yeah, absolutely. And just to add to what you were saying, Misha, and Jeremy, too, about having those cohorts. I think one of the other pieces in this is a level of transparency about what's the path forward, right? So, a lot of the times you don't know how much effort or work it's gonna take to start doing those skills. And so, one of the projects we've been looking at in some of our research has been - what are the career pathways you could take to transitioning to AI? So, you might already have seven out of the ten skills you need. And it's just those extra three skills that you might need that give you the foot up in progressing to that next step. And just knowing that and just being aware that I'm more than halfway there already is that sense of encouragement that can really nudge people to make that transition.
Misha Schubert: It's a really good point. We are heading rapidly towards the end of our fabulous discussion. And time is against us, but just a couple or more quick questions from our audience. One question I would like to get your thoughts on - what are the top emerging jobs that are starting to feature automation or AI skills to a reasonable degree?
Audrey Lobo Pulo: From a LinkedIn perspective, you can find out top skills on our website. But I think what's more interesting is how those skills are gonna be used. So, are we using our skills for a green economy? Where are those skills being employed and which are the sectors that are really going to benefit from AI?
Jeremy Howard: One thing I think that feels really underappreciated in Australia - so, I just got back here after ten years in the US. I got back in February. In San Francisco, I mean this seems really obvious but, a really big job right now is an AI engineer, a deep learning engineer. And at Stanford, for instance, by far their biggest, I think it's the biggest course they offer in first year is their AI course. Everybody wants to take it and they have to stream it because there's not enough room. And then I see people coming out of those courses first year out. So, I used to try to employ those people at my medical AI company. Salaries started at 200K US. There's not enough of these people. Everybody wants them. And I'm just not seeing young Australians being told this. People are still being told like, oh, you're super smart, you did well at high school, you should go into law or medicine. Law and medicine are great careers, but why aren't we telling them there's this huge, pretty newish industry in AI, where you're gonna make way more money than medicine and law in your first year, and have huge impacts and then impact all these other fields, as well? I don't think that's something we're saying loud enough to young people in this country.
Misha Schubert: I feel like Enrico might like a rejoinder to that.
Enrico Coiera: I don't disagree at all. To be honest, I'm sure if we trained everybody we could, just in machine learning, deep learning, we still wouldn't have enough. The challenge is that there are other roles, too. So, I would say to somebody who wanted to go and do medicine, look, you're going to do medicine, but you're going to become a computational doctor. I've got a wonderful professor here at Macquarie with me who is probably the world's first computational neurosurgeon. He doesn't just do surgery. He's busy building AI-based models. And so, the opportunities for these folks are just enormous. And I think, Jeremy, you said earlier on in the discussion that it still is early days and there are a lot of pioneering roles out there. It's different to what it was 30 years ago when AI was just a niche thing and you weren't so much a pioneer as a weirdo. And that was the generation that I came from. Now, it's cool and exciting. But what I'm seeing through the courses that Jeremy and others have put forward or the way in which state universities say this is now becoming much more regularised, it should be easy now for folks who are excited, who are starting out, to go in that area. But also, it should be easy for a medical student or a nursing student to get those skills also, to become the pioneers they need to be.
Jeremy Howard: And it's not just the skills, right? The resources to grow, like Enrico's talking about, 30 years ago or, I'm trying to think, 25 years ago when I was training neural nets, it was in the big banks. We needed millions and millions of data points. And we literally had to purchase special silicon hardware to run them. It cost millions of dollars. Nowadays, you can train a world-class neural net for free. Google has something called Colab where they give you GPUs for free. As we've discussed, the training, you can access for free. The algorithms we have now, using something called transfer learning, allow us in many areas to get brilliant results with 100 data points. So, the constraints, they've just collapsed.
And I think, to Audrey's point, if we said to people, here's where you are now, here's what you still need, people would be very surprised to discover how close they are to actually having strong capabilities.
Misha Schubert: We are so close to being out of time. I just wanted to invite us all to end on a high note. If we can nail all the terrific lessons you've exhorted us to think about today - that securing strong sovereign capability, not outsourcing, but doing even modestly from the beginning, from in-house capability ourselves, transforming our sense of how do we encourage the workforce at large to dip in our shoe in micro-credentialing and smaller bite-sized chunks of AI, applied education skills throughout a career and a working life, and build that momentum that will give us the network of people who are experts in this field, if we can nail all of that, what are going to be the upside benefits to Australia? What does that unlock for us as a country?
Enrico Coiera: I would hope that we become an agile and very resilient nation. I am optimistic in the sense that we have the capability. I worry that externally, there are a lot of things happening globally, climate change, the political space is a bit messy at the moment. And so, for this country to be sovereign and to be able to be agile and respond, it's going to have to do this. This is not a nice-to-have. This is essential to the survival of the nation over the next 50 years. And if we don't do it, we will be an outsourced nation. So, I'm excited but also I'm a bit scared that, there's a bit of inertia around. So, Jeremy will get us energised. Thanks, Jeremy.
Jeremy Howard: Oh, absolutely. I would be happy to, because I will say as an ex-expat, I've decided to come back to Australia despite all the attractions of San Francisco, I had high hopes coming back here, but actually coming back to Australia after ten years, the feeling for me has been, bloody hell, this is a good country! It really is. And yeah, there's a lot we need to do but we are fundamentally living in a bloody good country. So, I think if we can sort these other things out, we will be the envy of the world. We're one of the very few countries that actually had a competent enough COVID response to largely keep it out. This is not an accident. We've got smart people. We've got people who are passionate, who care, and a culture of fairness and generosity. So, that's why I'm here, because of what Australia is. And I do think those things would be very hard for another country to add on, right? But the stuff we're talking about, we can absolutely add on to Australia. And I think - yeah, I think we can have it all.
Misha Schubert: Audrey, do you want the final word?
Audrey Lobo Pulo: I echo what Enrico and Jeremy said, and I'll add this, and this goes to a comment that you made, Misha. We have some incredible women in AI in Australia. And gee, it would be nice to have a really strong AI presence where we are equally represented in gender. So, that's my vision for where we would be going in Australia. So, great things to come.
Misha Schubert: Terrific. Thank you all for such a fabulous rousing exhortation to round out this discussion. I want to thank our expert panel for a really rich and terrifically expert, insight-laden set of observations. Dr Audrey Lobo-Pulo, Professor Enrico Coiera and Jeremy Howard, thank you all so much for your time. Thank you to our audience for listening along and thanks for active participants in this conversation, as well, in making it a real conversation. To end this session you just need to click on that livestream button that you can see down on the left-hand side of your screen, and that'll take you into the next session. Enjoy the rest of your afternoon. Thank you for joining us today. Bye bye!
(ELECTRONIC MUSIC PLAYS)
The participants leave the chat and Misha ends the stream. The broadcast transitions back to the opening splash screen.
Text: TECHTONIC 2.0 – Australia’s National AI Summit – 18 June 2021. This session has now concluded. Thank you for joining.
To the right of the text is the collage of industry images.
- Techtonic 2.0
- Welcome, opening and keynote address
- Panel session: AI applications in manufacturing
- Primer on artificial intelligence
- Stream 1: Putting the AI Ethics Principles into practice
- Stream 2: The next wave of AI technologies
- Stream 3: How to AI-proof our workforce
- Stream 4: Using AI to deliver for citizens
- Panel session: Future opportunities for AI in Australia
- Closing address and remarks