TRANSCRIPT 🔗
00:00:00
Speaker 1: Grant Herman, welcome. It's great to have you with us.
00:00:03
Speaker 2: Good morning, Clarence. Great to be here. Good morning to you, your listeners, and a big happy birthday to your dad.
00:00:09
Speaker 1: Ah, thank you. You've been listening for a while then, Grant.
00:00:11
Speaker 2: I was driving to work, yeah.
00:00:13
Speaker 1: Okay, I don't know if there's anything that you want to take issue with me on, but let's talk about the essence of this conversation. It's about The South African Post Office, SOC Limited versus Fakudi, establishing that decision makers retain ultimate goal or legal responsibility and cannot delegate rulings or findings to AI platforms. Let us understand decision makers retain ultimate legal responsibility. What is at the heart of that ruling?
00:00:45
Speaker 2: Yes. So people bring a problem before a decision maker, and it's for the decision maker to then resolve the dispute and give the right answer, ideally. And it is obviously required of the decision maker to give a rational answer, an answer that can be challenged. And we are finding that decision makers are not doing that. There are a couple of cases that have been cited besides the post office one. The post office one is very recent. But going back even to 2023, there have been cases where decision makers have come up with crazy rulings based on AI hallucination, which has been generated. And this obviously leads to professional negligence claims, much embarrassment to them, and probably very negative impact on their careers. And this does not only lie at the feet of decision makers, but we also need to point out that litigants must also be very careful of how they prepare for a case. And there are cases where attorneys have prepared a case, argued a case, which is based on cases that don't exist. They've quoted case law that either doesn't exist or it does not support the case that they are arguing.
00:01:59
Speaker 1: Yeah, they're not arguing the merits.
00:02:01
Speaker 2: Well, they are, but based on false facts. I'm not an expert at AI, but how it works, it kind of falls in the gap sometimes.
00:02:11
Speaker 1: It falls in the gap. Absolutely.
00:02:13
Speaker 2: And it incorrectly falls in the gap at times. And as a lawyer, you can't afford to do that. You can't rely on that.
00:02:18
Speaker 1: Are you talking hallucinations specifically when AI makes things up?
00:02:23
Speaker 2: Yes, that is on the one hand, hallucinations. And then on the other hand, in labor law, you need to also be conscious of the concept of fairness. And that is something which I would never really be able to understand. We deal with an unfair dismissal claim in the CCMA case. And then a commissioner, a human being, will sit and listen to the dispute. And very often, employers and employees make themselves guilty of doing research using AI to argue their case. And I've come across this chairing disciplinary hearings. I've come across it representing employers where the other side comes forward with arguments that don't match the law at the CCMA, don't match South African labor law. And that is incredibly risky when you're coming to a battle and you've brought the wrong ammunition with you.
00:03:11
Speaker 1: Yeah, I would imagine caseload workload for a commissioner is significant in that maybe AI can assist with that caseload. You're not arguing against the use of AI. You're saying be circumspect of how you use it.
00:03:27
Speaker 2: Correct. I mean, we use AI without knowing it. When you're writing email, half the words are underlined in red or blue. And that's AI telling us that we need to check our spelling and our grammar. So AI is very prevalent. If we use the South African Post Office case as an example, there the commissioner made massive errors. He quoted and constructed a Constitutional Court case which doesn't exist. And the funny thing is one of the litigants in that Constitutional Court case was the South African Post Office in this hallucinated case. So this is a case where the post office had brought a jurisdictional ruling before the labor court, which the commissioner got wrong. And the commissioner relied upon this hallucinated South African post office case in his decision. And that case went completely contrary to the law as we understand it. And a very basic understanding of the law would have told him, hey, this is very strange. I need to research this a bit more deeply. I need to go and look for that case. I need to read it.
00:04:30
Speaker 1: Absolutely.
00:04:30
Speaker 2: I need to see what the circumstances were there in that case. Cross-reference. And that case could not be found. Absolutely.
00:04:36
Speaker 1: Yeah, so that is an omission, an oversight on his part and correctly sanctioned. But we've also seen many businesses use AI for automated maybe recruitment screenings. There's performance tracking. as maybe selecting candidates for retrenchment as well. I'd like to know, just on the other side, how do employers ensure those algorithms don't inadvertently, you know, trigger the unfair labor practice or discrimination, you know, kind of claims under the LRA especially?
00:05:06
Speaker 2: Yeah, it's important to speak to a human being about it.
00:05:10
Speaker 1: Yeah.
00:05:10
Speaker 2: Unfortunately, we're still there. We're still at the place where you need to bounce off a human being. Employers approach us, our members approach us often with a scenario And then they will tell us, this is what I found, this is what I researched. And very often it's wrong. And unfortunately, it seems as though AI gives you the answer you want. And that is very dangerous for any litigant. And it goes for the employee as well. If you put punch in the facts and you word it in a certain way, AI is going to tell you you've got a great case. And you're going to come forward with that. And unfortunately, you're going to be caught out. So it's risky to employers and to employees. Absolutely speak to a human being. Don't rely on the AI.
00:05:50
Speaker 1: Yeah, I don't know if you see that boundary between AI as a support tool versus a decision maker. And I'm wondering if that boundary is clear.
00:06:03
Speaker 2: Yes, I mean, you're welcome to use AI. And I probably encourage people to use AI to research, but you need to verify the authenticity of that information manually by going through it yourself.
00:06:18
Speaker 1: You know, if the court is also not necessarily being meticulous in its evaluation of the merits of that particular case, they can get a drawing. And how do you spot AI? How do you spot a hallucination? Somebody's double checking then. Yes.
00:06:36
Speaker 2: I mean, when you're litigating, the other side is definitely going to be checking your references. So that is something to be careful of. If you're quoting a case in your argument, they're going to be looking at that case and dissecting it and looking to see why it doesn't apply in this case. So that's when AI will come out. But it's for the other side then to argue that and the judge to come to the correct conclusion.
00:07:03
Speaker 1: Well, I think we're flagging again how we need to treat AI. And I think you've given us some really important insights there. Grant Herman, spokesperson from NISA, we appreciate your time. And I think that sweet spot resides right there between AI as a decision support tool and an autonomous decision maker, especially when it comes to the law.