"ChatGPT is not a war analysis machine": AI expert warns politicians about the text robot

ChatGPT is probably the first well-known text robot.

The AI-based program can compose highly complex texts within seconds.

And now?

Munich - ChatGPT recently wrote the speech of an EU MP and nobody noticed.

With his action, Tiemo Wölken (SPD) wanted to warn that the voice bot also harbors risks.

But can't you also see the tool positively?

When asked about applications for politics, Aljoscha Burchardt, an expert in artificial intelligence (AI), can at least think of a few.

Administrative tasks such as answering citizen e-mails, summarizing text translation or political writings, for example.

And very important: "To recognize what's burning under people's nails." Burchardt is thinking specifically that an AI will scan Twitter threads for the core statements.

"The result doesn't have to be perfect," he said in an interview with

from

.

ChatGPT and the Ukraine war: AI expert warns of "grossly negligent" errors

Not perfect, but

correspondent Yaroslav Trofimov drew attention to quite a failure that ChatGPT made: When asked who committed the atrocities at Bucha, the AI ​​is said to have answered: the Ukrainians themselves, too. Experts of the International Criminal Court (ICC) have been investigating the massacre allegedly carried out on civilians by Russian soldiers in Ukraine for months. 

Why is it that any AI query I make on Ukraine, it seems like ChatGPT is a subsidiary of Russia Today?

pic.twitter.com/Y5t5WKoIiG

— Yaroslav Trofimov (@yarotrof) February 5, 2023

“The sentence 'It was the Ukrainians' might have been plausible at a point in time in the past, and there are certainly people who would find it plausible – and the system has no more claim,” Burchardt added.

"It doesn't claim to be a fact or war analysis machine.

It is therefore also grossly negligent to look at ChatGPT from the perspective of truth.”

Parliament speech thanks to AI: "Can be interpreted from two directions"

Burchardt therefore advises politicians to be careful with AI in external communication.

The Wölken speech can be interpreted "from two directions": "Either the system is so great that it writes speeches that look deceptively real - or the speeches are usually so predictable that even a tool can write them.

AI always puts our own cognitive performance into perspective.”

According to Burchardt, however, ChatGPT was "post-trained" with the help of human feedback and has been pointing out controversial topics ever since.

"This layer was put on top so that nobody would ask, 'What do you think of the death penalty?'

and the answer is 'Great!'”.

Burchardt still saw a big problem: "What I find really difficult: As a user, you can get the impression that the system is incredibly careful, self-reflective and knows its limits."

Burchardt on the AI ​​location Germany: "Our economy has overslept trends"

Burchardt is part of the team of the "Initiative Large European AI Models" (LEAM) of the AI ​​Federal Association.

He believes that a better AI infrastructure is urgently needed in Germany.

His team, together with Robert Habeck (Green Party), recently submitted a feasibility study to the Ministry of Economic Affairs, “in which we recommend spending three to four hundred million euros on this.

We don't need to become 'gigantic' now, but become more agile,” he said.

Specifically, he stated that small or medium-sized companies do not have to submit an application to a data center “three weeks in advance” “in order to receive one day of computing time at the end”.

“I think that overall our economy has missed trends.

Politicians can't do much about that either."

Digitization is initially a feat of strength before it starts to be fun: "The whole investment: changing processes, converting machines, scanning documents," explained Burchardt and concluded: "And it is possible that political decision-makers shy away from this additional effort if they see that it will not bear much fruit in your legislature, but will still cause great unrest."