Subscribe
Share
Share
Embed
The Garden State Law Podcast discusses all things legal in the state of New Jersey. Presented by lawyers from Saiber LLC, the Garden State Law Podcast highlights trending topics in litigation, transactional law, real estate, and more.
00:00:03:21 - 00:00:12:20
Speaker 1
Welcome back to the Garden State podcast. I'm your host Tim Lough. Joining us today is Saiber employment attorney Vincent Cirilli. Vince, thank you for taking the time to join us today.
00:00:12:27 - 00:00:13:27
Speaker 2
Thank you, Tim.
00:00:14:00 - 00:00:24:25
Speaker 1
I'm particularly excited about having you here today because from what I understand, you have acting experience. You've had some theater experience. So I'm expecting some big things as well.
00:00:24:27 - 00:00:26:15
Speaker 2
No podcast experience.
00:00:26:17 - 00:01:01:06
Speaker 1
Well, maybe that's like a nice segue into theater, podcasting, theater, lawyer, podcasting. But today, you and I are going to assess artificial intelligence. And not just A.I. specifically, but how A.I. impacts employers in the workplace. Now, candidly, this is something that I'm fascinated by, A.I. in general. There's so many implications on the illegal industry and really any industry, from intellectual property issues to regulatory challenges to even ethical issues, I think, with attorneys themselves.
00:01:01:06 - 00:01:18:29
Speaker 1
So there's a lot of ethical issues. But today we're going to focus specifically on employment and how employers are using A.I., and you're the perfect person for us to have today. So before we get into it, maybe you could take a second and tell us a little bit about yourself and a little bit about your practice.
00:01:19:01 - 00:01:49:21
Speaker 2
Sure. So I'm Vince Cirelli. I've been a practicing lawyer at Saiber now for a little under nine years. I primarily do labor and employment counseling and litigation. I assist employers and helping them navigate the complex web of employment laws and regulations, from workplace investigations to defending them in courts. My goal is just to provide practical advice and advocacy on behalf of employers so that they could focus on their goals.
00:01:49:21 - 00:02:14:17
Speaker 1
And I think more and more, as time goes on, they will play a larger part of things that you're dealing with. You're probably starting to see it now, but, you know, I think as time goes forward, as we see the advancements in technology and how quickly it's happening, I would imagine that what is just kind of a little tangential part of your practice and something that you kind of see and you, you know, go to and host or guest podcasts about will be more and more a part of your practice.
00:02:14:17 - 00:02:37:13
Speaker 1
But so in terms of A.I. specifically, let's kind of start from the beginning and take a step back. Most people, I think, are relatively familiar with what A.I. is, but not everybody. So if you can give us a general description now, what what exactly is artificial intelligence?
00:02:37:15 - 00:03:06:02
Speaker 2
So in its most basic form, it's intelligence demonstrated by machines as opposed to human intelligence or animal intelligence. It has the capability of learning and presenting, depending on what kind of A.I. program you're using. Responses that look and feel like actual human thought. One of the most common examples right now that most people are familiar with is Chat GPT, which is an A.I. chat bot.
00:03:06:04 - 00:03:23:24
Speaker 2
It's capable of answering questions, developing recipes, planning, travel itineraries, and it actually helps me prepare that elevator pitch I gave to you just a few moments ago. But you didn't see that coming. So it's capable of doing quite a lot of things.
00:03:23:29 - 00:03:52:19
Speaker 1
So it's interesting that you use the word demonstrate and and I think that's something going to be a theme that goes throughout this conversation, because it's not necessarily not to get too into the weeds intelligent. It demonstrates intelligence is pulling intelligence from other places. So I guess that leads me to the next question. And this is a little bit of a difficult question for you to answer because you're an attorney and not a coder or programmer.
00:03:52:22 - 00:04:04:19
Speaker 1
How does it how does this work? How does how do they how does one program, one piece of software take seemingly all the information in the world and craft responses? How does how does it typically work?
00:04:04:22 - 00:04:37:23
Speaker 2
So there are many different types of AIS. Let's use Chat GPT for an example. Chat GPT relies on a large language model or an LLM, and it's a massive dataset of information that Chat GPT draws upon to generate answers to certain queries. Just how large we're talking about. It's estimated that the latest version of Chat GPT has about 1 trillion parameters and a parameter is nothing more than a mathematical relationship that links certain words to other words using algorithms.
00:04:37:25 - 00:05:12:04
Speaker 2
Now, it sounds very complicated. So let's let's try to break it down a little bit. So let's say you put in a question into Chat GPT. Chat GPT doesn't analyze the question or look at the question question in much the same way that you or I would instead think of it as a supercharged version of autocorrect, These A.I. chat bots are just very good at predicting what word follows the next word in order to generate a response to your query or question.
00:05:12:07 - 00:05:32:10
Speaker 2
So in the end, at the end of the day, just like if you're typing on your iPhone or an Android phone and your phone starts to predict the next word, and that sequence of in that sentence that you're typing, Chat GPT is essentially doing the same thing, but going one step beyond that and trying to also generate an answer.
00:05:32:15 - 00:05:51:03
Speaker 1
It's interesting because I've noticed in the past few months when we had some type of upgrade in the office with Microsoft, I'm typing an email or I'm in Microsoft Word, I'm working on a sentence and in Gray is showing me what it thinks I'm about to say. And I could just key over and I could skip a few words.
00:05:51:08 - 00:06:17:02
Speaker 1
It doesn't really save me that much time, but it is a little bit almost troubling. But they know. They know my thoughts before I'm actually putting them down on paper. Now, you likened it to kind of a robust version of spellcheck or you have, at least in the past, to me, with the squigly line, you know, it's saying like, is this really the word that you want to use?
00:06:17:04 - 00:06:38:18
Speaker 1
But sometimes it misses things, right? So you're writing a word and you drop a letter. It's not the word you intended on using, but it's still a word. So things get missed. I guess that leads me to the question is, is this is this infallible? I'm assuming that there's going to be problems, right?
00:06:38:18 - 00:06:51:02
Speaker 1
I mean, generally you put in a query, things come back, things make sense, you're happy, but sometimes it misinterprets something and we have problems. So I'm assuming that's the case, especially at its infancy.
00:06:51:04 - 00:07:19:13
Speaker 2
Your assumption would be correct. It is not infallible. In fact, I read an article a few weeks ago where the author was using Google's A.I. chat bot, Bard, and the author typed into Bard the question “How many times does the letter E appear in the word ketchup?” Seems simple enough, right? Bard's answer was that the letter E does not appear in the word ketchup.
00:07:19:15 - 00:07:42:21
Speaker 2
And it's fascinating because that kind of just goes back to the point that I was making before that these A.I. chat bots don't see the world in much the same way that we do, right? It doesn't see the word ketchup. It doesn't see the letter E. It lives in a closed world of training data, and if so, if that training data doesn't account for a question like that, you might get unpredictable or inaccurate responses.
00:07:42:23 - 00:08:08:06
Speaker 2
Now that that's an example that's, you know, innocuous and no harm, no foul, but you can see how that might have some real world implications. So not too long ago, a couple of attorneys, not from our firm, were drafting a brief and they used Chat GPT to try to find some on point cases to submit to the court. These cases were great.
00:08:08:12 - 00:08:16:29
Speaker 2
To check the cases, the attorneys asked Chat GPT whether the cases were real. Chat GPT said, very convincing, Yes, the cases are real.
00:08:16:29 - 00:08:25:21
Speaker 1
They actually asked Chat GPT, you promise you being serious and had said yes, they had to look them up on their own.
00:08:25:24 - 00:08:33:00
Speaker 2
No, no, that that was the exchange. And so I guess that was good enough. And they submitted the brief to the court.
00:08:33:00 - 00:08:34:22
Speaker 1
And these were fake cases or not real.
00:08:34:24 - 00:09:00:02
Speaker 2
These cases were not real. And the word the court used was bogus. They were bogus cases, bogus citations. And when you submit something to the court with bogus cases and bogus citations, you run the risk of being sanctioned. And these two attorneys were sanctioned in the amount of $5,000 each. So you can see how the limitations of this technology could have real world implications.
00:09:00:04 - 00:09:29:20
Speaker 1
Yeah, I keep going back in my head to Microsoft Word and spell check where they're trying to assist you in grammar and it doesn't take the place of proofreading. So I think something already to keep in mind or for me to think about is that AI already is an incredible tool, but it doesn't replace, you know, the time spent reviewing documents, proofing, making sure things are accurate.
00:09:29:20 - 00:09:55:18
Speaker 1
It gives you a head start as opposed to, you know, completely taking the place of something else. So now that we have a general sense of what this is terrifying as it may be to some people, including me, to a certain extent. Let's look at the the employment world and H.R. managers and business owners who are implementing certain pieces of software with the AI technology.
00:09:55:21 - 00:10:03:18
Speaker 1
What are employers currently doing in terms of using AI in the workplace?
00:10:03:20 - 00:10:27:02
Speaker 2
So employers in h.r. Departments are using ai in a variety of different ways from recruitment and hiring to analyzing performance of the employees to assisting with termination decisions and even helping using ai to help draft emails and policies but by far the most common area where we see ai being used is in the recruitment and hiring field.
00:10:27:02 - 00:10:46:19
Speaker 2
And there was a survey done in 2022 by the Society of Human Resources Management that found that 79% of employers are using some form of AI or automation to assist with recruitment, which is fascinating. It's already seems to be somewhat ubiquitous now in the workplace.
00:10:46:19 - 00:10:47:29
Speaker 1
It's more than I would have thought.
00:10:48:01 - 00:11:19:12
Speaker 2
Exactly. And so So how does it work when you're using it for recruitment purposes? A.I. can help analyze large amounts of data and sift the resumes and help employers more quickly and efficiently identify potential job applicants that might be suitable for for the next stage in the hiring process. Like an interview and perhaps even a job offer, A.I. can also be used to assist with the interviewing process.
00:11:19:15 - 00:11:44:19
Speaker 2
So there are some AI programs that help with screener interviews. They can monitor a job, applicants tone and the content of what is being said in response to an interview question to see if it would make sense to move that candidate along to the next stage in the hiring process. And it could also be used for performance evaluation, evaluating employees performances.
00:11:44:19 - 00:12:19:22
Speaker 2
So take parcel delivery services, for example. A lot of them, some of them use AI to monitor driver safety. They can monitor driver movements. They can monitor how often a driver reverses the truck and take that data and overall come up with a performance assessment. So it really has quite, quite a lot of variety in terms of what A.I. to be used for in the in the workplace.
00:12:19:22 - 00:12:40:29
Speaker 1
And imagine even during the pandemic as well, when people are working from home and I know a lot of employers were using software to monitor their employees. So are they connected? Are they logging on? How often? Leaving keystrokes and taking they're analyzing it and building that into a performance review as well.
00:12:41:02 - 00:12:52:08
Speaker 2
Yeah, even like webcam eye tracking software to make sure that employees are looking at the screen so you can see how it starts to get quite invasive pretty quickly.
00:12:52:10 - 00:13:24:05
Speaker 1
So you already mentioned one instance of two individuals who are using AI as part of their workplace, and they were you know, they were deceived, if that's the word you want to use or are using it improperly. So if you're in a if you're an h.r. Manager or your business owner or someone who is overseeing the ai process in employment practices, what should you consider as being some of the pitfalls in using this software to your advantage?
00:13:24:08 - 00:13:48:22
Speaker 2
Well, I mean, as I mentioned before, AI needs to be trained and if you're relying on bad training data, there's a good chance that bad results may follow. Now, here's an example that's outside the employment context, but I think it illustrates the point quite well. Back in 2016, Microsoft released an AI chat bot called TAY, it’s an acronym that stands for Thinking About You.
00:13:48:25 - 00:14:08:03
Speaker 2
And it wasn't long after TAY was released that the whole world realized quite quickly what TAY thought about everyone else. So TAY got a Twitter account and a lot of its training data came from tweets.
00:14:08:05 - 00:14:09:03
Speaker 1
Of other people.
00:14:09:05 - 00:14:10:24
Speaker 2
Of other people on Twitter.
00:14:10:24 - 00:14:12:16
Speaker 1
I can already see how this is going to go wrong
00:14:12:17 - 00:14:44:00
Speaker 2
It's this was an unmitigated disaster. So shortly after TAY was released, users began tweeting at TAY, highly inappropriate offensive messages, and TAY took that information and started tweeting back highly inappropriate and offensive messages to the point where Microsoft had to shut down and it shut TAY down pretty quickly. And when I say quickly, I mean it was shut down in 16 hours.
00:14:44:03 - 00:14:49:18
Speaker 1
It didn't take long for the people to start tweeting, added to the inappropriate comments, I guess.
00:14:49:21 - 00:14:54:21
Speaker 2
No. Now, I'm shocked that, you know, that the Twitter populist.
00:14:54:24 - 00:14:59:06
Speaker 1
Yeah. Of all of all places, really. So that...
00:14:59:06 - 00:15:23:28
Speaker 2
That kind of highlights the potential now for employers when using AI systems. I'm not saying that the AI systems being used by employers are going to be insulting job candidates or employees. But if if the AI systems being used are relying on bad or biased data, you might find yourself with a situation where the AI systems are generating bad or biased results.
00:15:24:01 - 00:15:51:15
Speaker 2
So take resumes for example. If if the training set for an AI system used for analyzing resumes, if you submit a bunch of resumes to the AI and you tell the AI he's the good resumes, these are the candidates that we're looking for. And 90% of those resumes just happen to be coming from white male candidates or employees.
00:15:51:18 - 00:16:17:21
Speaker 2
The AI might look at that information and conclude, Well, if I'm going to be looking at a resume, I want to prioritize white male candidates. And so before you know it, now you have an AI system that's relying on an impermissible bias that might start excluding otherwise qualified job applicants because, oh, well, this person went to a woman's college, so that can't be good.
00:16:17:21 - 00:16:28:03
Speaker 2
We're going to exclude that person or this person is a minority. So that's not good. We're going to exclude that person and you can see quite clearly have that would lead to a lot of ramifications.
00:16:28:10 - 00:16:53:24
Speaker 1
Yeah, I think this keeps going back to the same idea that this is a tool that should be used to support someone's work function and not replace it because you don't have someone keeping an eye on the things that were flagged as not a good candidate, which someone who has, you know, 20 years of experience can quickly look at a resume and say, Oh, I could tell within a moment that this is the right applicant for us.
00:16:53:27 - 00:17:15:17
Speaker 1
Now, you know, you're going to get probably burned pretty often. So out of curiosity, the employers are are they building AI models themselves? Are they using vendors? How are they implementing this? I, I haven't heard. I mean, I don't work in the h.r. Field necessarily or at all. I haven't heard of. Oh, you just purchased this software off the shelf.
00:17:15:17 - 00:17:26:15
Speaker 1
I mean, I'm a certain I'm sure there's some screening software that maybe you can, but are there prominent vendors or are we talking about large institutions that are building something on their own?
00:17:26:16 - 00:17:53:06
Speaker 2
So it's really a little bit of both. So a lot of large companies have the resources to build and build these models in-house and they rely on those models. Some companies, smaller companies do rely on vendors to help supply them with with the relevant AI software. The problem sometimes with using vendors is that some employers might think, well, if I'm using a vendor and something goes wrong, it's on them, it's not on me.
00:17:53:06 - 00:18:21:00
Speaker 2
And that's not true. If you're an employer using a vendor and the AI system starts discriminating against job applicants or employees of certain protected classes, you also run the risk of potential liability in situations like that. So it's important that if you're going to be using a vendor to make sure that you're you're asking the right questions, you know, how often does this vendor test for disparate impact?
00:18:21:02 - 00:18:47:21
Speaker 2
What kind of processes are in place at this vendor for making sure that, you know, the training data remains unbiased and that the results of the AI system are unbiased. And, you know, sometimes it might be difficult because if you if you're relying on a vendor and a lot of this information, the algorithms used, the training data used, it might be a black box, might be trade secret.
00:18:47:23 - 00:19:04:19
Speaker 2
I mean, so but at the end of the day, an employer using software provided by a vendor needs to feel comfortable and assured that the software is not going to be, you know, engaging in any disparate impact or discrimination.
00:19:04:19 - 00:19:38:12
Speaker 1
Sure. So we work together fairly regularly and we pretty routinely will issue client alerts, letting our clients or contacts know when there's new guidance issued by a regulatory body, the EEOC, the Department of Labor. Has there been, and I don't really recall sending all that much out on this topic specifically, has there been much or any guidance issued by any regulatory body?
00:19:38:14 - 00:20:15:15
Speaker 2
Well, I mean, looking at federal laws, first, there are no federal laws specifically addressing the use of AI for for assisting with employment decisions. But you do have agencies like the EEOC and the Department of Justice that have issued some regulatory guidance. Keep in mind, even though there might not be specific federal laws addressing this specific topic, there are federal laws, anti-discrimination laws such as Title seven, the Age Discrimination in Employment Act, the Americans with Disabilities Act.
00:20:15:18 - 00:20:54:03
Speaker 2
That if you're using an AI system that's discriminating against, you know, people of protected classes, you might run afoul of those laws. And so a lot of the guidance that we've seen so far from the EEOC specifically relates to how to avoid that type of disparate impact. Going beyond that, you do have some states, I believe there are 16 states that either have laws on the books or bills in various stages of the legislative process that do discuss specifically the use of AI when making employment decisions.
00:20:54:06 - 00:21:09:20
Speaker 2
In fact, I think New York City passed a local law just this past July on that topic. And New Jersey is has a bill that's closely mirrored after the New York law.
00:21:09:22 - 00:21:37:27
Speaker 1
So speaking of running afoul with the law, there must be instances, I would imagine, in recent years or even months where employers have used software that have been burned. Are there anything or are there any instances that are that come to mind that might serve as a warning to others, you know, to be careful when using these products?
00:21:37:29 - 00:22:11:00
Speaker 2
I'm glad you asked. Actually, yes. The EEOC did file its first lawsuit back in 2022 against a group of three companies. It's EEOC versus ITutor group, and these companies provided English language tutoring services to students in China. And what the EEOC alleged was that AI system that these companies were using to recruit candidates discriminated against candidates on the basis of age.
00:22:11:07 - 00:22:42:07
Speaker 2
The EEOC alleges that there were about 200 otherwise qualified job applicants that were denied a job at these companies on the basis of age. Obviously, the companies are denying liability and the case remains active, but it is a warning shot and employers need to be made aware that the EEOC is looking closely at employers use of AI.
00:22:42:13 - 00:23:07:17
Speaker 1
Yeah, so what what can employers be doing now more proactively? So if we're if they're assuming that it's going to be this software will be used in the future by their own employees, whether it's a piece of software that has been vetted and approved or if an employee is using a piece of their software on an ad hoc basis in not even alerting their employers.
00:23:07:24 - 00:23:17:07
Speaker 1
What is some of the guidance or what is some of the things that employers say we give to employers that they should be considering moving forward?
00:23:17:10 - 00:23:56:17
Speaker 2
One thing to consider is implementing an AI policy which sets forth how the AI is used, how it's monitored, how it's improved, things like that. Another thing that employers should definitely consider doing and some of these laws that have been passed in certain states and some of the bills that are currently in the process that that may be passed, they do require this is to engage in a bias audit, and that's just essentially checking these AI systems to make sure that they're not desperately impacting anyone in a protected class.
00:23:56:20 - 00:24:18:12
Speaker 2
Again, if you're going to be using vendors, as we discussed previously, just make sure you get some information from the vendor as to what their bias audits look like and how they test for disparate impact. Just to make sure that you're making a well informed decision as to whether to use an AI program from any particular vendor.
00:24:18:15 - 00:24:40:00
Speaker 2
And at the end of the day, AI is not the only thing that needs to be trained. Your employees should also be trained on the proper use of AI. When used properly, A.I. can be a really efficient tool, but if used improperly you might find yourself citing two cases that don't exist and dealing with the fallout after that.
00:24:40:03 - 00:25:00:26
Speaker 1
Well, thank you Vince. That's very interesting. I feel like this is a podcast that we can probably have yearly because the conversation we have in 12 months could look very different than what we are having today. But thank you for joining us. If they would like to if anyone would like to reach out to Vince directly about AI in the workplace, you can reach him at (973)622-3454 or VCirilli@Saiber.com.
00:25:00:26 - 00:25:12:25
Speaker 1
VCirilli@Saiber.com. Thank you, Vince. Thank you to all those who are listening and watching. Make sure to subscribe to our YouTube channel and podcast channel and we'll talk to you next time!