TONKLI, Mr Robert, Chief of Staff, Shop, Distributive and Allied Employees' Association WHITWORTH, Mr Sean, Researcher, John Curtin Research Centre Committee met at 09 :00 CHAIR ( Ms Briskey ): I declare open this hearing of the Joint Select Committee on Artificial Intelligence. I begin by acknowledging the traditional custodians of the land on which we are meeting this morning and pay my respects to elders both past and present, and I extend that respect to Aboriginal and Torres Strait Islander peoples joining us today. I remind all witnesses that, in giving evidence to the committee, they are protected by parliamentary privilege. It is unlawful for anyone to threaten or disadvantage a witness on account of evidence given to a committee, and such action may be treated by the Senate as contempt. It is also a contempt to give false or misleading evidence. Witnesses also have a right to request to be heard in camera. If a witness objects to answering a question, they should state the ground upon which the objection is made, and the committee will determine whether it will insist on an answer, having regard to the ground which is claimed. If the committee determines to insist on an answer, a witness may request that the answer be given in camera. I remind all senators and all committee members of their obligation, under the Behaviour Code for Australian Parliamentarians, to treat witnesses with dignity, courtesy, fairness and respect. I now welcome representatives from the Shop, Distributive and Allied Employees' Association, the SDA. I understand that information has been provided to you regarding parliamentary privilege and the protection of witnesses giving evidence. Do you have any comments to make on the capacity in which you appear? Mr Whitworth : I am a researcher with the John Curtin Research Centre who aided the SDA in this submission. CHAIR: I now invite you to make a short opening statement—as short as you can, because we've got limited time and I know there are lots of questions that members are keen to ask of you. Mr Tonkli : I'll abbreviate the opening statement, of which you have a copy. I'll just note that the SDA represents 200,000 workers in retail, fast food and warehousing. Women make up the majority of our membership. The majority of our members are under the age of 35, and most are on low incomes. For our members in these industries, AI isn't a future issue; it's already here, built into digital work systems. It sets their rosters, allocates their tasks and times of work, and, more and more, is also deciding whether they're performance managed. Usually, it's brought in with little or no consultation. Very briefly, our submission makes two key points. Firstly, AI is a work health and safety issue, not just a productivity or technology issue. Secondly, sovereignty and fairness at work can't be separated. Whoever controls the algorithm controls the decision, whether it's on a worker's roster or the systems running our defence and critical infrastructure, and our submission makes a key point on national AI sovereignty as well. As to recommendations in our submission, first, we must ensure that rights at work are fit for the age of AI, and we have detail on that. To support this, we're calling for a Fair Work Commission expert panel on AI. Finally, we need to ensure our national sovereign capability. That's the abbreviated version of the opening statement. CHAIR: I appreciate that, Mr Tonkli. I want to acknowledge some of the case studies that you included in your submission that really highlight the extent to which the deployment of artificial intelligence is happening, as you say, right now, in businesses and companies across Australia, and the real potential for the human impact of that on workers. What I'm particularly interested for the committee to hear you expand on is the gap between the algorithm, which can help drive efficiencies and help companies find those efficiencies, and what's actually happening—what's happening in real life, all the contextual factors that exist in a workplace, on a shop floor. Would you mind expanding, for the committee, on some of the experiences that are happening in your members' lives, in their workplaces, where that gap is quite large and it's having some pretty significant consequences. Mr Tonkli : In the submission proper, we highlighted three examples: one from a supermarket, one from a warehouse and one from a discount department store. We would say they're common in those sectors. For example, one would be the use of AI to performance-manage a worker. The example would be a check-out worker with more than 30 years experience, a senior person, whose scanning rate at the check-out was excellent. The error rate was very, very low, comparable to an automated system. However, on a particular shift, the AI surveillance camera, the software, picked up some errors. The way in which we understand the algorithm and the process operated was that it resulted in an automatic first and final warning. The person was never consulted. They were never told what the benchmarks or the requirements were. And we would say that the automated decision didn't take into account the context of what was happening on the check-out or the employee's excellent scan rate historically. So we've obviously challenged that. The point we make is that there has to be a human in the loop. We can't have a situation where a computer or an algorithm is making decisions without a human being—the person answerable to—because otherwise it doesn't pick up context. The algorithm might be wrong. We go into some of the other examples. The way algorithms are often developed is with work as imagined versus work as actually performed on the floor. We would say, if the views of workers who actually do the work were taken into consideration and the consultation were genuine, then that would be taken into account. The outcome both for our members, for workers, and for the employers would be a much better outcome, because otherwise, apart from disputes, you end up—again, with our ALDI example—with more and more physical injuries, not to mention the psychosocial injuries. In our submission we refer to a survey that we conducted of 11,000 members in retail, who were very clear that the psychosocial risks and effects, where more and more is demanded of them—and that's often exacerbated by an algorithm—are unsustainable and unfair. That's why we're calling for common sense, I think, and fairness. CHAIR: I think you highlight some really important ways in which we can try and get this right, keeping humans in the loop but also that genuine consultation and a sense of agency over how AI is deployed in workplaces, and the extent to which having both the workers' work experience and a genuine means to be consulted ends up delivering, potentially, a better outcome overall for the company and the efficiencies that they're seeking to gain and the opportunities that AI can provide around productivity and efficiencies—but in a way that's actually helpful for workers as well as the company. Are there any examples that you're seeing at the moment and might be able to provide as to where it is working well, where it has been deployed in such a way that there's been genuine consultation and it's in fact improving workers' experience at work or their lives? We hear stories about the extent to which it can help augment work and can help make work easier and better for workers. Are you seeing that at the minute? Mr Tonkli : Not really. I'm sure there may be examples out there, and I'd be delighted to hear of some. Unfortunately, we've come to you with examples, from the experience of our members, which aren't great. They're actually exacerbating issues in the workplace. You say there's a lack of real consultation, or genuine consultation. Consultation doesn't mean that we get what we ask for, necessarily, but it does mean that our concerns are listened to and the genuine experience of workers on the shop floor is taken into account. But then, hopefully, through that feedback, there's co-design—some of that feedback is incorporated into the model—and the outcome is better. Under the determination change and redundancy provisions in the Fair Work Act, often, with technological change, workers aren't consulted properly, because the decision has already been made at the time new technology is brought in. To try and get better examples, the Work Health and Safety Act model would be a much better model. Under that legislation, employers hold a primary duty to eliminate or, where not reasonably practicable, minimise risk to workers' health and safety, and this duty explicitly extends to psychosocial hazards such as excessive workloads, unsafe staffing levels and poorly managed change, which is our experience. But, under that act, the consultation requires that, before the decision is made, there has to be consultation. We think it's a better model for something that has as wide-ranging an impact across all sectors and across all work as AI is having and will increasingly have. CHAIR: I understand that the deputy chair doesn't have any questions just at the minute. Senator Sheldon has a few. Senator Sheldon. Senator SHELDON: I'll pick up on that issue about consultation. You mentioned that it's not only about disputes; it's about consultation and conversation between both the employer and the worker. When there are disputes, the Fair Work Commission is a place for conciliation and bringing parties together to have conversations, as well as the potential for arbitration. I note that in one of your recommendations—recommendation 4—you talk about an expert panel. Can you step us through that? What would be appropriate for the Fair Work Commission to do with regard to overseeing workplace AI technology, particularly as we're seeing technology ultimately rolling out now in all parts of the economy? Mr Tonkli : We are calling on an expert panel of the Fair Work Commission to deal with issues around AI. Australia has done this before. So, pre the Australian Industrial Relations Commission, when industries raised specific operational or industry issues, there were dedicated specialist panels or tribunals. AI is such an impactful technology. It's affecting all workplaces at the same time, in different ways, of course. It's already in use—again, in our sectors, in rostering, performance management and so forth—because it's having such a wide impact. We can only see it growing and developing. By developing expertise in the Fair Work Commission—they're experts, of course, in industrial relations—or having a specialist panel that can then deal with the specific issues of AI to cover things like algorithmic management, automated decision-making, data governance and the psychosocial risks that I pointed to, we think that would be a win-win. In terms of their expertise, it would develop and give consistency to what we think will be ongoing and emerging challenges in the workplace as AI is rolled out and developed. Senator SHELDON: You also mentioned statutory AI implementation agreements in your submission and the role that could play in not only getting the best benefit out of AI but also safeguarding the workforce at the same time and safeguarding business to have more efficient and effective outcomes. Mr Tonkli : I might ask Sean to speak on that one. Mr Whitworth : Yes, absolutely. With the AI implementation agreements, our submission argues that consultation in its current form is coming after changes have already been decided. We're putting forward that effectively it would be best practice for both business and the workers if that can be discussed beforehand—a best-practice approach based on how the work is actually completed and what the businesses themselves need. We argue that it shouldn't be able to be bargained away, as it could be through enterprise bargaining, and this would shore up the regulatory environment for businesses. It would make sure that workers are able to put forward their concerns and guarantee that these conversations take place, keeping any retroactive refitting from having to be undertaken. Senator SHELDON: There's a proposition about making sure this conversation is taking place in workplaces and that there's a mechanism to do that. As you mentioned, Mr Tonkli, there's a specialist group within the Fair Work Commission. Is that within your submission? Mr Tonkli : Yes. Senator SHELDON: For examples of why that's necessary, we turn to a really powerful example. From my old Transport Workers' Union days, I note in the systems that were brought to Aldi, for example, as far as I understand from your submission there has been no consultation and no voice in how the system should work. People are being substantially injured, with injury rates going up. In my old trucking days, where they hadn't worked with people loading the vehicles, trucks were being loaded with heavy weights at the back of the truck, which I know is extremely dangerous for both the truck driver and other road users. Aldi putting people's lives at risk by not having a proper process is extremely concerning. Of course, there's a flip side of that. There are many examples like Amazon and how it's been approaching these similar issues as well, which traditionally have been seen as cutting edge. Can you run us through what the live experience is at Aldi and Amazon? Mr Tonkli : There's an example in the submission about Aldi, where they introduced new picking efficiency productivity benchmarks, which were generated through data analysis of the warehouse operations. It was work as imagined through the data. Management reported increased throughput of 16 per cent and reduced labour costs of 26 per cent. So you think, 'Wow, wonderful,' but what's the real impact to workers? We surveyed affected workers and found a very different picture, with significant and escalating levels of musculoskeletal strain, extreme stress, mental exhaustion, fatigue, longer recovery periods and workplace injury coinciding directly with those new benchmarks that were developed through the data. There were workers who had exceeded their productivity requirement for years who, in some cases, found themselves unable to keep up with the new benchmarks. As I said, it was generated from work as imagined. There was no consultation. We obviously challenged that. Unfortunately, there were injuries, and we had to challenge that. Amazon, I think is perhaps even worse in that they use technology in a way that maximises the output. For example, at their Kemps Creek facility, pick rates are driven by an AI managed goods-to-person system, and we say the overall pace of work is calculated using unsustainable performance metrics. Again, it's how you set the algorithm to maximise throughput. The system measures the amount of time it allows you to complete a single pick, and this is the maximum time a worker has to pick an item from a robotic pod, scan it and confirm it on the screen, and there's a continuous feed of items to pick, with the autonomous robots bringing inventory-filled pods directly to the worker at a stationary picking station. The point is that every second a worker is logged into the system, it's tracked, including time off task or idle time, and that creates a constant pressure to continue a really high pick rate, which is unsustainable. The high proportion of labour hire workers who are working there are at risk of not getting ongoing shifts unless they continue to meet an unsustainable performance rate, which, again, goes back to the digital work system that embeds the algorithm. It's about how you set that. Do you set it to maximum efficiency, or do you set it knowing workers have limits? Not everyone can work at the same limit. If you push it too far, people get injured. If people feel that their jobs are threatened, they might injure themselves because they want to maintain their employment. We think it's a serious issue, particularly as these systems are only becoming more and more embedded in our workplace. With the lack of consultation particularly before the system is introduced, we end up with these terrible situations where workers are getting injured. Senator SHELDON: You have retail examples of treating people like barcodes with a pulse, and then you've got this example of extreme fatigue, stress and long recovery periods. Employers need to realise that this is somebody's livelihood. They're raising a family. This is not a survival contest. There's a real need for change. CHAIR: Ms Chaney. Ms CHANEY: Thanks for your evidence today. I'm interested in your views on the role of the AI Safety Institute. I note you make a recommendation about strengthening regulator and AI Safety Institute capacity. I am interested in what you think the AI Safety Institute should be doing more of and what it needs in order to do that. You mentioned that the AI Safety Institute should conduct annual independent reviews of developers' risk management systems. Given that the AI Safety Institute has no visibility over those risk management systems, technically, I'm interested in whether you think it also needs to have powers so it can access and audit those safety programs. Can you give me a bit more on the role and scope of the work of the AI Safety Institute. Mr Whitworth : In response to that, our submission calls for expanding the AI Safety Institute's powers through establishing a national model assurance facility. The point of this submission has been modelled on what France calls its PEReN. Effectively, the regulators maintain their legal coverage of their particular sectors, but there is a standalone body that is capable of auditing and providing information across the board to the regulators. We have argued that they should have the power to call for training data. They'd conduct adversarial testing and continuous verification and ensure that the AI systems are increasingly intertwining with our critical systems. Our more standard businesses and institutions would have more of a view over those systems and provide advice on where those weaknesses are. Bodies like the Fair Work Commission and the ACCC are facing rapidly expanding responsibilities without matching increases to funding or technical expertise. For the AI Safety Institute, the capacity there would be that they are able to assist in expanding technical expertise across our regulatory environment. Ms CHANEY: That's part of their role now, but they've got 25 employees and $7 million of funding a year. Would you see a need to uplift that resourcing to meet the challenge? Mr Whitworth : I'd say the current numbers are insufficient to address everything. To address the role that AI has now taken and will increasingly take in our economy, our critical infrastructure and our defence systems, we need a much stronger capacity to review the current systems and to provide our regulators with the technical understanding they need to police the AI sector. Ms CHANEY: I have a question on job displacement. There's a lot of uncertainty about what the job impact will be from AI. Given that in some scenarios it has the potential to lead to significant job losses, are you satisfied the government has done enough planning for how it might respond to any significant job losses in those different scenarios? Is it okay for the government to be looking at the historical data and responding to that, or should it be doing some scenario planning on where this could go and how we might need to think about the problem differently in different scenarios? Mr Whitworth : I'd say referring to the historical data is appropriate based on current circumstances. AI is coming onboard in a scope and at a speed which is unseen compared to things like the Industrial Revolution. Our submission calls for development of a national AI skills academy so that workers are able to study portable and stackable microcredentials that will keep them up to date with developments in artificial intelligence, the idea being that, rather than having to travel or commit to a TAFE or university degree, you're able to build up microskills that you can take from workplace to workplace, making it more accessible to casual, part-time and full-time employees who aren't able to necessarily take that break from work. Mr Tonkli : Planning is important. We don't yet know the impact. Various reports will say different things. There was a KPMG report that said the impact of AI will—and this is speculative—lead to greater productivity and therefore more jobs in the retail sector because people have more money to spend. That would be great if that's the case; we don't know. We know the data is not showing that quite yet, but we have to plan. Part of that is that skills development is key, and part of that is our call for just transition. If the worst-case scenario is that there's a loss of lots of jobs, which gives rise in some scenarios to a great productivity increase or benefit, then there has to be a sharing by those who are at work in that benefit but also by those who have been displaced. Ms CHANEY: What are some ways that that sharing mechanism could work? Mr Tonkli : We've called on bargaining with retailers for additional leave. Paid leave could be one example of sharing that, as part of the productivity benefit, and a pay increase in this difficult time. It could take a number of forms, but it wouldn't just have to be money; it could include paid leave. Ms CHANEY: Thank you. CHAIR: Senator Pocock. Senator DAVID POCOCK: Thank you very much for your time and your submission; it is an important one. Do you think, when it comes to AI and data centres, that big tech companies and other vested interests have had a disproportionate impact on government policy? Mr Whitworth : I would argue that they have definitely had a significant input, having the background and expertise for their work and their sector. I would say that it's not been total input; there have been significant examples of community pushing back and representatives listening to that. I would argue that there have been non-government organisations conducting similar work and trying to pull the conversation from being solely about the benefits provided by industry towards a more societal conversation about the impacts on living areas, water, electricity costs and such. The main point we take with the data centre conversation in this submission is making the argument that having data centres housed in Australia is not enough to guarantee that our data is protected. Under legislation like the US CLOUD Act and various national security laws in China, data stored in Australia can still be compelled and called back to their home countries. So, unless Australia were to develop its own data centre and sovereign compute capacity, that would always be an ongoing issue. In terms of whether that would play with Australian industry having the same concerns—if we were to develop a home industry, would they have an outsized say over the future of data centres?—I don't necessarily know. That would be a conversation for then. But, from this immediate point in time, I'd say it would be in Australia's best interest to develop our own data centre, cloud sector and sovereign capacity to make sure that our critical data is protected. Senator DAVID POCOCK: Finally, and I know we're right up on time, I want to ask about something in your submission. We're being told that we need AI to lift productivity—more output per hour worked by Australian workers. Aldi says that its new warehouse system raised throughput 16 per cent and cut labour costs 26 per cent—an increase in productivity. Did any of that reach workers as increased pay or fewer hours worked? If not, how do we ensure that Aussie workers benefit from productivity gains, not the big companies? Mr Tonkli : It's a fantastic question that goes to the heart of our submission. They've obviously realised the productivity gains through the implementation but at the expense, we would say, of workers' health and safety. There has to be a better way, and hopefully this is what the committee will help achieve. CHAIR: We are right on time. Thank you so very much for your appearance and evidence provided to us today. We really appreciate it. If there's anything additional that you want to provide the committee, please do so through the secretariat..