1
00:00:03,280 --> 00:00:11,480
Venolan Naidoo:
Hi everyone, I'm Venolan Naidoo and this is the next podcast 
episode on AI and the world of work.

2
00:00:12,360 --> 00:00:17,280
This episode tackles when AI goes wrong at work, 
discipline,

3
00:00:17,400 --> 00:00:22,000
data and defensibility. In the previous episode, 
if you recall,

4
00:00:22,000 --> 00:00:27,760
I spoke about AI workforce literacy or AI workforce fluency 
as I like to call it,

5
00:00:27,760 --> 00:00:30,320
as a board management level risk.

6
00:00:30,840 --> 00:00:40,240
In this episode, I want to focus on what happens when 
organisations don't address it and how AI issues surface as

7
00:00:40,240 --> 00:00:47,960
real workplace disputes. Because most organisations won't 
realise they have an AI problem until it lands on someone's

8
00:00:47,960 --> 00:00:54,880
desk as a disciplinary issue. A data breach, 
an AI output relied upon that went wrong because it was not

9
00:00:54,880 --> 00:00:57,440
properly checked or a legal challenge.

10
00:00:58,800 --> 00:01:01,040
Discipline and procedural fairness.

11
00:01:01,360 --> 00:01:04,920
The stuff that us Labour lawyers love to deal with.

12
00:01:05,320 --> 00:01:14,520
One of the most difficult questions organisations will face 
is this can disciplinary outcomes be defended when AI played

13
00:01:14,520 --> 00:01:22,120
a role? We're already seeing situations where AI influences 
performance assessments.

14
00:01:22,240 --> 00:01:30,760
We see it where AI assists in investigations, 
and we see it where employees rely on AI without clear

15
00:01:31,120 --> 00:01:35,440
guidance. If an employee is disciplined in those 
circumstances,

16
00:01:35,880 --> 00:01:43,040
the question becomes whether the process was fair, 
especially if the organisation never set clear rules around

17
00:01:43,080 --> 00:01:49,200
AI use. AI complicates accountability, 
and decision makers,

18
00:01:49,520 --> 00:01:56,040
especially a presiding officer in a hearing, 
will look closely at whether the employer created an

19
00:01:56,040 --> 00:02:02,560
environment where AI use was properly governed and if it was 
underlyingly responsible.

20
00:02:03,560 --> 00:02:11,950
Data confidentiality and POPIA risk is something else that 
many organisations do not realise is also important.

21
00:02:12,910 --> 00:02:15,790
It is a major risk with data protection.

22
00:02:16,310 --> 00:02:21,230
Employees using AI may unintentionally upload confidential 
information.

23
00:02:21,750 --> 00:02:27,270
They may expose personal data or trigger a cross-border data 
transfer,

24
00:02:27,430 --> 00:02:33,590
which has its own rules. When that happens, 
many organisations focus heavily on international AI

25
00:02:33,630 --> 00:02:38,590
frameworks while overlooking South African specific 
workforce risks.

26
00:02:38,630 --> 00:02:42,670
Under POPIA people, organisations don't necessarily know 
this,

27
00:02:42,670 --> 00:02:49,430
but POPIA applies to the use of AI systems, 
especially in workplace environments involving individuals.

28
00:02:50,030 --> 00:02:55,750
That gap creates exposure, particularly when employees were 
never properly guarded on,

29
00:02:55,750 --> 00:03:03,470
and it isn't acceptable. Sector specific realities, 
AI doesn't affect every sector in the same way.

30
00:03:03,750 --> 00:03:07,910
For example, industries like insurance, 
financial services,

31
00:03:08,070 --> 00:03:12,630
Professional services. Ai already influences outcomes there,

32
00:03:13,030 --> 00:03:18,550
from claims processing, assessments, 
recommendations and decision making.

33
00:03:19,070 --> 00:03:24,950
But the common thread is this AI is already being used in 
ways that affect people and outcomes.

34
00:03:25,270 --> 00:03:31,710
Without AI workforce literacy, 
these sector specific risks multiply exponentially.

35
00:03:32,190 --> 00:03:35,070
From training to diagnosable risk.

36
00:03:35,670 --> 00:03:39,590
And this isn't about AI training, 
per se.

37
00:03:40,110 --> 00:03:43,470
Organisations don't pay lawyers to teach people how to use 
tools.

38
00:03:44,150 --> 00:03:46,670
They pay for answers to hard questions.

39
00:03:47,390 --> 00:03:51,990
Are our AI influence decisions defensible from a legal 
standpoint?

40
00:03:53,430 --> 00:03:56,070
Are they workforce risk gaps that exist?

41
00:03:56,150 --> 00:04:02,230
Have we identified those and are we exposed from an 
employment or data protection perspective?

42
00:04:02,990 --> 00:04:06,350
These are the hard questions that organisations need to 
address.

43
00:04:06,790 --> 00:04:11,340
And that's why AI literacy needs to be treated as a 
diagnosable legal risk,

44
00:04:11,540 --> 00:04:15,260
not only an education exercise, 
which is, of course, important.

45
00:04:16,340 --> 00:04:24,260
This would innately require AI risk management exercises 
being conducted to identify the nature and level of risk.

46
00:04:24,780 --> 00:04:33,220
And then with this very critical information to begin with, 
education edifying a workforce for understanding the nuances

47
00:04:33,220 --> 00:04:37,620
of AI use and ultimately for better adaptation.

48
00:04:37,660 --> 00:04:44,500
Yes, it's about adaptation. In conclusion, 
AI won't wait for organisations to catch up.

49
00:04:45,260 --> 00:04:54,300
The choice is whether you address workforce issues with 
readiness proactiveness or manage the fallout

50
00:04:54,340 --> 00:05:01,820
reactively. In my view, proactive is the prevailing position 
to pursue,

51
00:05:02,140 --> 00:05:09,900
and I look forward to catching up on the next episode where 
we'll be talking on another AI topic and the world of work.

52
00:05:10,220 --> 00:05:10,860
See you then.