When AI Sets the Bargaining Table

Transcript

Eric Felsberg
Principal, Long Island

Hello, and welcome to We Get AI, our podcast dedicated to the world of artificial intelligence. My name is Eric Felsberg. I am a principal and one of the co-leaders of the AI team here at Jackson Lewis, and I'm joined by my friend, fellow principal and co-leader, Joe Lazzaroti. As we all know, AI is developing at a very rapid pace, and regulators are focusing on keeping up with this pace of development. On this podcast, we focus on keeping our listeners updated on the latest developments, demystifying the relevant issues and offering best practice considerations. Joe?

Joe Lazzarotti
Principal, Tampa

Hey Eric, good to be with you. Today we are blessed to have a guest with us, our own partner, Marvin Kaplan. If you want to do the introduction.

Felsberg

We're thrilled to have Marvin with us today. Now, some of you may know or at least have heard of Marvin Kaplan. Before joining Jackson Lewis, Marvin served as the chairman of the National Labor Relations Board. How are you doing today, Marvin?

Marvin Kaplan
Principal, New York City

Well enough. Looking forward to talking AI.

Felsberg

Likewise. We're excited about our discussion. With that, let's get right into it. Marvin, can you give us some insight and provide us with a general lay of the land as to the competing issues regarding AI when it comes to labor negotiations?

Kaplan

This is back to the world of the second coming of automation. These were issues that came up when factories started automating, the same things that are coming up here – job displacement, employees are concerned about layoffs, implementation of AI, retention, etc. They're concerned about how management will use the materials AI produces to assign work, evaluate performance, schedule employees, measure productivity, and recommend discipline. They're worried about workplace surveillance. AI computers can track GPS, take video, analyze video, use biometric technology- things that were not feasible from an employee or a normal manager standpoint.

The one you hear a lot about that is important in labor, but is more of an EEOC issue, is bias and discrimination, concerns about how the AI models are being produced and what they're actually measuring for. In the event of no human review, many employees are concerned that they are identifying individuals for work who are pre-programmed to identify things they want.

The other one that comes up a lot in monitoring as well is how they are monitoring your emails. When I was a Board member, I remember I was at a meeting once, and someone raised their hand and said, we have an AI model that can track union interest at an employer based on emails. I was sitting next to a Democrat member at the time, and she looked at me and said, did you write down what company that was? I do not want to know. It colors these lines of how much more can be done with AI.

The one that is less relevant to our discussion, but bears mentioning here, is the intellectual property in the creative work, which was at the heart of the writers’ strike. What work can be done by AI, what work should be done by individuals, and how does that displace? At the heart of this displacement concern among employees is this question: in this AI world, where do we fit in?

Felsberg

I would just add something to what you said: overall we're seeing a trend of increased emphasis on transparency. You mentioned that in your comments, but even when we look at some of the emerging laws on the state and local level, for example, it seems that more and more of the emphasis is on issues of transparency. Requiring notice and informing candidates and employees that they may be subject to some form of AI evaluation or monitoring, like you said. Your comments fall right into line with what we're seeing across the country.

Kaplan

Let me say there that is why, for some, the labor, the unionized workforce, is particularly attractive because if you're going to make these changes to your workplace and you're unionized, you have to bargain over them. That is an added protection in states where they have not implemented these transparency protections, etc. Transparency is built into your rights under the NLRA, which, in many cases, is a great selling point for unions, who say that if you're unionized, they can't implement a lot of this without bargaining with us first. That gives you the opportunity to be part of these decisions wherein normally or in some circumstances you would be completely left out.

Lazzarotti

Marvin, what prompted Eric and me to come to you with this was an article we read about nurses wanting more input on AI guidelines. Eric and I do a lot of work thinking about the companies our clients buy these services from, how to create governance rules, and what integration and deployment look like. From a labor perspective, what guidelines are you seeing? The article focuses on nurses, so maybe speak to that. I'd love to hear what you are seeing and what you expect to see in terms of how unionized workforces and unions advocate for employees, and what guidelines they are talking about, basically.

Marvin

Well, nursing and the medical field create their own complications. You're not talking about a flaw in a piece of paper. You're talking about an individual and their health. It is exacerbated by the fact that you're not talking about trade secrets about how paper is produced. You're talking about an individual's HIPAA-protected information. In those contexts, particularly in healthcare, it's more difficult because patient safety raises the bar, making the bargaining not only more sympathetic but also higher-cost and higher-reward.

If you're talking about a lot of hallucinations with AI – let's say you're using AI to help put together a plan for a patient, and it hallucinates something in its recommendation. The consequences for a healthcare hallucination are far greater than the consequences for hallucination and filing a brief before even a Court of Appeals. A Court of Appeals might sanction you and yell at you, but no one's dead. The safety, the patient safety stuff changes things. It has to be addressed delicately, and there has to be much more buy-in from a healthcare perspective. Making sure that the nurses and providers are comfortable with the AI tools, giving them the confidence to use them as effectively as possible while ensuring they understand that built-in systems are in place to protect them. The evaluation of those systems is particularly important given the potential dire consequences if it's not done properly.

With regard to outside the healthcare field, the issues are the same. They're just less significant if something goes wrong, but the concerns are the same. Staffing, jobs, security, electronics, surveillance, productivity monitoring. Those are all the same. They all want to be part of it. They want to know how the information is going to be used. If you're recording how much they produce in a day, how many widgets you produce in a day, how is that information going to be provided?

There is actually great value in some of this. Discipline is the one I always fall back on, as it takes some of the ambiguity out of discipline to have a model that gives consistent results regarding how discipline should actually be enforced or implemented. There is a huge benefit there, not only to employers, but also to employees, because the goal is to treat everyone the same under the NLRA or any other statute. Yes, there's a lot of issues involved there, but part of it is getting that buy-in from the employees early on, making them understand how this is going to affect their job, how it's going to be helpful to their job, rather than the Baba Yaga who's coming in to disrupt everything possible that they've been working on.

Lazzarotti

To that point, whether in healthcare or other industries, Eric and I often speak with clients about identifying use cases.  Because AI, a particular AI technology, a tool, can have a lot of different ways it can be used in a lot of different contexts. When you're in the middle of negotiating with a union over guidelines, those guidelines can sometimes look very different depending on how the technology is used. How does that work, and what are some of the more popular use cases that may become topics of negotiation?

A related question is: how focused is it on making sure we identify the duty to bargain? How do you deal with the fact that technology is changing so quickly and that use cases may arise after negotiations have closed, especially when it comes to rolling out new technology?

Kaplan

It all depends on your contract. If you're unionized, it depends on your contract terms. The current standard for management rights clauses: you might have specific language regarding AI in a particular provision, and that would control. Most CBAs do not have that yet, as far as I've seen. They are relying mostly on management rights clauses to say everything here was covered, although the current Board standard, in defiance of the D.C. Circuit, is a clear and unmistakable waiver, and that's pretty narrow. I have cautioned many in assuming that your management rights clause is broad enough to waive bargaining under clear and unmistakable waiver. I have long argued that until there's a return to contract coverage standard, those types of behavior should be taken with a significant amount of caution.

With regard to use cases, there are tons of use cases. The interesting thing is that the Board doesn't necessarily regulate AI. As you pointed out correctly, it regulates how you use AI and what your bargaining obligations are in the event that you use AI. If your use of AI affects your terms and conditions of employment, then you have to bargain. If for some reason you have a specific AI provision that allows you to implement without bargaining, you still have to bargain over effects. There is circuit precedent that, if you ever want to call and go to sleep because you can't fall asleep at night, I'm happy to walk through when you actually do have to effects bargain over management rights clauses. In general, the idea is to bargain.

Again, this is only in a unionized context, only a unionized firm or company. The one I've seen a lot is operational use cases. That's scheduling, staffing, and human resources use cases because you see them all the time at the EEOC. That's hiring, evaluation, promotion, and discipline. Those are the ones you keep seeing at EEOC, where they say the algorithm is inherently racist, and as such, it's discriminatory.

You're seeing more of the monitoring use cases. That's surveillance with cameras and productivity tracking. As those are used, their use is mixed, but you don't only have to bargain over whether you use them; you also have to bargain over how you're going to use the information they provide. This comes up a lot with discipline. When they are monitoring how often a truck stops or how long the stops are at a delivery, the question is whether, if they are not long enough, it could lead to discipline. Or, if the stop is too long, could it result in discipline? It's not only about its use; it's about what you're going to use it for. That is why these are getting so complicated: they cover multiple areas of negotiation and help you determine the proper use of that model. The NLRB will be intimately involved in this, but not to regulate AI; rather, to regulate how it's used in the workplace.

Lazzarotti

Maybe some practical takeaways. Neither Eric nor I are labor lawyers, and negotiations are not something I've ever done. Different types of negotiations, but certainly not labor negotiations. It's really fascinating stuff.

For someone who is thinking about how to get to the bargaining table on stuff like this, what might someone think about to be successful with addressing how their business is using AI and trying to get as much of that into the agreement, or to be able to take advantage of the technology as much as they can and what they think makes the most sense? What's the strategy? How do we make sure we're dealing with this effectively as an organization?

Kaplan

The answer is true among most things in bargaining. You're going to have to address it. It is unlikely that you're going to be able to secure a management rights clause that gives you singular authority to change terms and conditions of employment regarding AI. Unless the union is asleep on the other side of the table, it is not a viable strategy, in my opinion. You can try for it. It's like trying to shoot for the moon, land in the basement. Take the shot, but it's unlikely they'll let you have it.

That leads to my view: be honest with the union and involve them in the AI selection and testing implementation. Have them understand how it works, be straightforward about how it will be used, seek their input on appropriate use, and be clear about how it will affect staffing levels. The more you can get through the initial issues before implementation, the better the rollout will be. If you are trying to blindside them, you're going to get stuck in negotiations and won't be able to implement. This is why I tell people all the time: you can't implement until you have bargained in good faith through this. It's not like you can say, we're going to put all this in place, and then we'll let you know on the back end how it's going to affect you. It just can't happen that way in a unionized workplace. You're going to have to go through all this.

That's where I suggest getting the parties involved early in dealing with these issues will go a long way toward getting their investment. I have long argued with people on everything that getting an individual invested in what you're doing and having them understand the value of what you're doing and why it's a benefit to them is far more useful in the long run than trying to jam a change down someone's throat. Because all you'll end up with is resistance. If you've already decided and said this is what's right for you, I don't know that I've ever been able to win with, no, this is the right thing to do. They've all looked at me like I've lost my mind, and you can't imagine anyone else will react the same way when you tell them this is what's right for them without explaining why or getting them invested in the work.

In many circumstances, it's true that AI, how it's used in the workplace, could be of great benefit to employees. I go back to, and this might be a broken record, discipline. Consistent discipline would be a great thing in the workplace. It would be better for me. I would be less busy if employers implemented it consistently, and in most cases, it's not intentional. It's a lack of information. AI is that one resource that can provide that information and consistency. All of these things together show the importance of AI and how it can be used as a great tool. But if it's implemented without understanding, because of concerns about job loss, staff reductions, et cetera, it is automatically seen as negative rather than considered in terms of how it might be positive.

Felsberg

Marvin, it's interesting what you're saying because Joe and I hear it more anecdotally: AI's coming for my job one day. It seems we have to get over the initial hurdle of showing this isn't about displacing a job; to your point, it's about enhancing a position.

Going back to something you were saying before about being transparent, making sure we understand how these tools are going to perform and what we intend them to do. Are you seeing more experts getting involved in the negotiation arena around things like validity? Is this thing actually doing what it purports to do? How do we monitor for things like bias? How do we understand what types of data it's processing? Are you seeing it, or is it more of a layman's discussion, short of involving AI technical experts?

Kaplan

I have not seen the experts come to the table. I will say, though, that from the companies I've seen, I've been impressed with the information they have available to provide to a company intending to use it, and that can be distributed to employees to make those points. Information explaining how the model was built, how it will work and why it's not biased, et cetera. In those circumstances, I would argue that's probably more helpful because it is information you can use to digest what you're getting.

Too often, you run into this with the courts too: individuals can be hostile to witnesses or experts they believe are there to sell a narrative. You bring someone in from the AI company and say, Yes, this expert from the AI company says their AI system is unbiased. You see the same issue at the Board, where the Board will often say, this manager is biased because he works for that employer; why should he be given greater credit than anyone else? AI companies would benefit from spending time building a resource that can be provided to show, this is how we did this, this is how we did it, to ensure it's not biased. Then it falls to the employer to write up its own policies outlining, for example, that a 20% reduction in productivity is not discipline-worthy. This is when we would warn you and give you coaching and training. The AI system would identify ways we can improve efficiency, and we would meet with you to discuss them.

That's the thing – it's not just the AI model. Again, it's the implementation that is important. It's the second part: yes, we are going to use this product, but we're going to use it in this way, and when we identify an issue, this is how we intend to use it. Not leaving it open, as yes, we're going to use this technology, and it is going to affect how we handle discipline. It is going to affect how we do discipline in a specific way.

Lazzarotti

This has been great. We could probably sit here, Marvin, and talk about all this stuff for hours. I really appreciate the insight. I hope the folks listening do as well.

For the listeners, we appreciate you guys listening in. If you have any questions or any thoughts for future episodes, please email us at AI@JacksonLewis.com.