Skip to player
Skip to main content
Skip to footer
Search
Connect
Watch fullscreen
Like
Bookmark
Share
Add to Playlist
Report
The Law Report's Damien Carrick speaks to ABC about the inaccuracies of ChatGPT
ABC NEWS (Australia)
Follow
10/15/2024
Experts say AI bots create falsehoods by using word prediction. These AI platforms can generate wild inaccuracies – known as 'hallucinations'.
Category
📺
TV
Transcript
Display full video transcript
00:00
Well, Martin Bernklau is a journalist who lives in the city of Tübingen, which is in
00:07
the south of Germany, and for decades he's been reporting on the local courts.
00:12
About a year ago, he pivoted his journalism into arts reporting, and he was curious to
00:18
know how his arts reviews were tracking.
00:21
Now his Microsoft system suggested that he use Copilot to do the search, which is the
00:30
AI platform attached to Bing, which is the Microsoft search platform.
00:36
So he starts dialoguing with Copilot, and he gets a huge shock, because Copilot tells
00:44
him that Martin Bernklau is a convicted child abuser, has been convicted of defrauding the
00:50
elderly, has escaped from a psychiatric institution, and is guilty of many, many other crimes.
00:59
Now what appears to have happened is that these are all crimes and court cases that
01:04
over the years Martin Bernklau had reported on, and what had happened was that when the
01:11
AI platform scraped the internet, it came to total wrong conclusions or hallucinations
01:17
about who Martin Bernklau is.
01:20
And just another detail, not only did it say this is who he is, it also gave his telephone
01:26
number, his address, and a route map to his house.
01:29
My goodness, so what did he do when he discovered these hallucinations?
01:34
Well he took a couple of steps.
01:36
He approached the prosecutor in his local town.
01:39
I don't think that step led to anything happening.
01:43
He also approached the German Office of Data Protection, and that office contacted Microsoft's
01:51
European HQ in Dublin, and they said they would look at it.
01:56
But subsequent to that contact, similar searches resulted in the same kind of slanders or inaccuracies,
02:04
according to Martin.
02:06
But Martin tells me just in recent weeks, when you type his name into this platform,
02:12
nothing comes up at all.
02:13
And it would appear that what has happened is, according to Martin, is that it can't
02:18
deal with the inaccuracies, so it's just kind of scraped him from the system totally and
02:23
won't respond to any queries about him.
02:25
So have there been legal cases arising from these types of AI hallucinations, Damian?
02:30
Well Martin certainly is considering it, but when he talked to a lawyer in Germany, the
02:37
lawyer said, look, we don't know how long it would take to resolve, whether you'd get
02:43
a good outcome, and it would be very expensive.
02:46
Have there been other cases?
02:47
Well here in Australia, just earlier this year, what was going to be the big probably
02:54
world test case was dropped.
02:56
That was being brought by the mayor of the Hepburn Shire Council in regional Victoria.
03:02
His name is Brian Hood.
03:05
And he dropped what was believed to be the first test case in the world because it was
03:10
too expensive.
03:11
Now what had happened to Brian Hood was he typed his name into ChatGPT, and it described
03:16
him as a convicted criminal, a fraudster.
03:19
In fact, Brian Hood is the polar opposite.
03:22
He is a very highly respected whistleblower who came forward when he realised that there
03:27
was something wildly amiss at a subsidiary of the Reserve Bank of Australia.
03:32
And he's widely respected for having brought that criminal behaviour to light.
03:38
So he was very concerned, but he decided that because of the, largely because of the cost
03:43
associated with uncertain litigation, he decided to drop that test case.
03:48
Are there any legal actions happening elsewhere in the world?
03:52
Well, the one that is going ahead is perhaps, not surprisingly, in the more litigious United
03:58
States.
03:59
And that involves a radio and podcast talk show called Mark Walters.
04:04
And he's suing OpenAI, the company that owns ChatGPT.
04:08
Now Walters hosts a program called Armed American Radio, which discusses and promotes gun ownership
04:16
in the USA.
04:18
And ChatGPT had hallucinated that he was actually being sued by another organisation or an organisation
04:25
called the Second Amendment Foundation, which is a US-based organisation that also supports
04:30
gun rights.
04:32
Somehow ChatGPT had hallucinated that he was being sued for defrauding this organisation,
04:38
when in fact he'd had nothing to do with it, he'd never worked for it.
04:42
And this was just a total hallucination.
04:45
ChatGPT has tried to dismiss this action, but it is going through the courts.
04:49
So it will be very interesting to see what happens.
04:52
And it may prove to be the test case that we're all waiting for around the world.
Recommended
2:06
|
Up next
AI Created This Video: The rise of ChatGPT
Australian Community Media
2/8/2023
10:21
What Is ChatGPT?
WatchMojo
2/9/2023
1:35
AI chatbots will soon be hard to tell apart from humans: 'The chatbots will get trained to handle less common problems'
Bang Tech News
9/27/2024
2:01
AI pioneer warns of its dangers
Brut India
5/4/2023
2:30
ChatGPT: Is it possible to detect AI-generated text?
euronews (in English)
1/19/2023
0:49
A conversation with OpenAI's ChatbotGPT
Australian Community Media
12/12/2022
0:46
Google launches AI Chatbot 'Bard' to rival ChatGPT
Australian Community Media
2/8/2023
2:37
What is ChatGPT, and should we be afraid of AI chatbots?
Tech Insider
3/31/2023
4:24
ChatGPT Users Grapple with Uncertainty As Watershed Lawsuit Threatens AI Future
Entrepreneur
1/9/2024
0:40
Las Vegas Cybertruck Explosion Suspect Used ChatGPT to Plan Attack, Authorities Say
Benzinga
1/8/2025
52:53
The future of AI might look a lot like Twitter | The Vergecast
The Verge
10/2/2024
2:39
Audio deepfake scams: Criminals are using AI to sound like family and people are falling for it
euronews (in English)
3/25/2023
0:53
ChatGPT firm develops AI model for creative writing: 'I have been really struck'
Bang Tech News
3/15/2025
0:49
Sam Altman Warns Users Not To Blindly Trust ChatGPT Despite Its Rising Fame
Benzinga
6/30/2025
1:13
Google launching their own AI chatbot to rival popular ChatGPT
Bang Tech News
2/7/2023
1:26
AI To Talk to Animals? Some Say It’s Very Possible
Amaze Lab
7/1/2025
8:31
ChatGPT: When AI upends the real world, mixing 'real elements with crazy, disgusting hallucinations'
FRANCE 24 English
3/26/2025
1:09
AI video clone Saffron Howden speaks multiple languages
Australian Community Media
1/12/2024
1:18
'Godfather of AI' warns of potential dangers of artificial intelligence
The Manila Times
6/30/2023
1:13
AWANI Tonight: AIFA: Govt launches AI chatbot to combat fake news
AWANI
1/28/2025
1:39
Suffer from social anxiety? These ChatGPT glasses use AI to do the small talking for you
euronews (in English)
6/12/2023
2:40
The silent AI takeover of white-collar jobs
India Today
5/2/2025
1:36
Apple ‘working on stripped down AI chatbot to rival OpenAI’s ChatGPT’
Bang Tech News
6 days ago
8:59
What Would Happen If AI Cloned and Replaced You? | Unveiled
Unveiled
6/23/2023
1:30
King: AI must not threaten protection of people or democracy
Royal Family Channel
11/1/2023