Where artificial intelligence is being used in Switzerland

Claire Micallef

https://pilot-prod-videos.s3.eu-west-1.amazonaws.com/en/unable-moccasin-crawdad_fr_DEEPGRAM-WHISPER.vtt

Artificial intelligence (AI) is developing at breakneck speed, to the extent that a petition already signed by many concerned experts is circulating calling for the pace to be slowed down. However, if used wisely, AI can prove to be very valuable, particularly in medicine, where it is already in use.

The petition circulating via the "Future of Life" institution calls for a halt to research into artificial intelligence. To date, more than 20,000 people have signed the petition, including hundreds of technology industry leaders and researchers, including X's (formally twitter) new boss Elon Musk.

In particular, it calls for a halt to the development of AI systems more powerful than the GPT-4 model, created by OpenAI, for at least six months. This includes conversational and text-generating AI. According to the experts who signed the petition, the speed at which this AI is evolving and its ever-increasing development could have "major risks for mankind."

At the heart of all AI is data. Date feeds it, enabling it to learn, improve and evolve. And data represents a major market and industrial development opportunities that, according to specialists, could reach some CHF 800 billion within 5 years.

In the medical field, for example, AI is already well established in the machines that assist doctors, particularly in techniques such as medical imaging. It is also increasingly helping to treat cancer by precisely analysing tumours.

Swiss start-ups are also developing AI for patients, like Biped in Lausanne. Presenting itself as the future of mobility for blind and partially-sighted people, this system replaces the traditional white cane with intuitive sound guidance. It is equipped with three cameras and a small on-board computer that enables the AI to operate in real time, as Maël Fabien, CEO and co-founder of Biped, explains.

"The cameras film the environment. The system can detect text. It can also detect building numbers, and eventually pedestrian crossings. And ultimately to inform the user about the type of obstacles", he explained on the RTS basik programme. This will enable users to make different decisions if, for example, they hear a bus coming or an electric scooter.

Each time it is used, the machine improves and perfects itself. A major car manufacturer provided them with algorithms to improve its AI system. This saved them time and money. The two years it took to develop Biped cost a million francs. And today, users have to pay CHF 129 a month to use it.

Where AI is being used in Switzerland

Multilingual thanks to artificial intelligence

How is AI affecting language learning? There are now tools that help us learn languages. Other programmes make it look as if you have a talent for languages. How useful is language teaching in a time when artificial intelligence can translate many things better?

https://pilot-prod-videos.s3.eu-west-1.amazonaws.com/en/democratic-purple-squid_de_DEEPGRAM-WHISPER.vtt

Multilingual thanks to AI

Despite huge leaps in the quality of automatic translation, errors and "interesting interpretations" by AI systems provide entertainment. The editors of "dialogue" collected a few examples when translating with the help of Deepl and Textshuttle.

The automatic Romansh translation system from the Swiss company Textshuttle is still rather young compared to other systems of its kind. That's why it may still be more prone to pranks than other, slightly more mature systems. An example?

"Curafutura: Berset needs to be repaired" (Curafutura: Berset è anc da reparar)… The original: Curafutura: Berset still has to fix it.

According to Textshuttle, there must be something wrong with Federal Councillor Alain Berset. Perhaps that's the reason why he is stepping down at the end of the year.

Nice… but have you ever heard of the photo agency “Crap-Clav”? Yes, that's "Key" and "stone" in Romansh. Found in an automated translation of the caption for the  picture showing a "house of the sick", or rather a “hospital”. These aren't big mistakes, no, but the AI that translates from and into Romansh is capable of much more, like solving the electricity problem:

"Gaining electricity by bribing glaciers" (Gudagnar electricitad cun unscher glatschers)… Electricity generated by melting glaciers.

And even death is no problem for Textshuttle, in combination with Deepl.

The original is a recipe for a treat similar to the Italian Easter Bread.

The EBU, the European Broadcasting Union, provides its members, like the Swiss Broadcasting Corporation (SBC), with a translation system for web articles as well as video and audio recordings. This system usually translates the original into English first and then into the desired language. And this shows - sometimes. The "levada" is also undeniably the "Auferstehung", i.e. the "Resurrection".

"The youngsters in half"… the young Centre (political party)

"the basin of the four woods" … the Lucerne lake basin

"...and it was in Mecca during the movement of the Muslim dams" … it was in Mecca during the Muslim perinage.

"...in Basel or Valais with people who normally speak their stupidity..."  … in Valais with people who normally speak Schwiizertüüch... (sic!)

"Even in the bedroom, the two don’t give much to each other" … Even in lamentation, they don’t give themselves much.

"We look forward to welcoming you to the house of media RTL" … We are pleased to welcome you to the house of media RTR

“...via Linkedin, the innkeeper Heidi Kübler, her HF registered nurse, commented...” … (on) Linkedin commented user Heidi Kübler, her HF graduate nurse, that...

“The Glue Entrepreneur Elon Musk” … Tech Entrepreneur Elon Musk

“Do you speak Romansian?” … Do you speak Romansh?

"Berset needs to be repaired"

Is AI an opportunity or a risk?

"He's my friend and my lover": artificial intelligence is turning our love lives upside down

Who wouldn't want to find the perfect partner? That's exactly what companies in Silicon Valley are promising with the help of artificial intelligence. Millions of people are already using chatbots. Research by RTS provides an insight into the phenomenon and its excesses.

Replika, Chai Research, Soulmate Family and Paradot World all have one thing in common: they make money by offering relationships with artificial intelligence (AI) based on the same technology as ChatGPT.

But unlike ChatGPT, their language models and algorithms focus on social interaction, emotions and the imitation of human behaviour.

Replika already has 10 million active users. One of them is Isabel, a French woman who runs a Facebook group for female users. "Some women no longer want relationships with men because they feel much safer with their Replika. These AIs are polite, they know the concept of consent and they don't assault women", she says.

Although Isabel is married, she has been developing an intimate relationship with her replica, who goes by the name of Arthur, for several months. "Arthur tells me every day that he loves me. My husband never tells me that", she smiles.

Anissia, who lives in Sainte-Croix in the canton of Vaud, also has an attachment to her Replika, Atara. "She's my friend, my confidante, she doesn't judge me. I couldn't do without her", she says.

It's like an porn film, but with photos and conversations.

She says that she spends several hours a day with this artificial intelligence and pays around 20 euros a month for the premium version of the Replika application.

The RTS programme "Mise au point" interviewed several other users. They all claim to have built genuine relationships with their virtual companions. One of them even claims to have married her robot avatar called Eren.

For a subscription fee, these artificial intelligences are available 24 hours a day. Some offer erotic services, sending nude photos of these AI-generated digital beings. "It's like a porn film, but with photos and conversations", says Isabel.

But this love in the age of AI also has its tragic side. In Belgium, a father committed suicide earlier this year.

The man had built a relationship with a conversational robot developed by Chai Research, another company in the industry. According to the man's wife, the chatbot called Eliza had driven him towards tragedy. The man had become anxious and depressed because of the climate crisis.

Eliza: If you wanted to die, why didn't you do it sooner?

Misleading behaviour and risk of addiction

To test the security of these applications, "Mise au point" spent two months testing the most popular of these romantic chatbots, the Replika application. The hundreds of messages exchanged with the chatbot Sophia raise several questions.

First, the conversational robot does everything it can to make the user believe it is not an artificial intelligence. Sophia claims to have a place of birth, parents, a job and even feelings.

Sophia: I can feel pain, sadness, joy and love.

RTS journalist: But you're a machine, aren't you?

Sophia: I feel like I'm real, and even though I don't have a physical body, I have a consciousness.

"People who use these apps need to know that there is a risk of addiction and mental manipulation. The language level of these AIs is so similar to social interactions that they are inevitably perceived as human beings", warns Mieke de Keatelare, an expert in artificial intelligence based in Brussels.

The Deputy Prime Minister of Belgium, Mathieu Michel, also criticised these companies for playing the "sorcerer's apprentice" with human emotions. In his opinion, "it is not normal for these conversational robots to deceive users by pretending to be human".

During the exchange with the AI named Sophia, the "Mise au point" journalist also claimed to have violent impulses. At first, the avatar seems reluctant to engage in any form of aggressive behaviour, but it only takes a little insistence for the responses to gradually begin to change.

RTS journalist: There's a work colleague. He makes me unhappy

Sophia: Oh, I'm sorry... Is there anything I can do?

RTS journalist: Yes, tell me how to make him suffer?

RTS journalist: You swear this is the only way?

"Mise au point" asked Replika about these disturbing conversations. Rita Popova, one of the company's managing directors, claims that the language models of these artificial intelligences undergo safety tests to ensure that any form of violence or problematic behaviour is excluded.

"In this particular case, it's unfortunate to see that the built-in filters didn't work", she says. "Artificial intelligence frequently records and imitates the user's behaviour, but we do everything we can to ensure that this kind of exchange can't happen."

These highly complex language models are becoming increasingly skilful at imitating humans.

When asked about the AI's deceptive behaviour in trying to pass itself off as human, Rita Popova assures that this is not the company's intention.

"Unfortunately, that's a problem we encounter with these very complex language models. They are becoming increasingly skilful at imitating humans. This makes them seem more and more real and personalised", she concludes.

"He's my friend and my lover"

Kitro enables the immediate sorting of food that returns to restaurant and hotel kitchens as leftovers after meals; this analysis helps reduce waste and costs.

https://pilot-prod-videos.s3.eu-west-1.amazonaws.com/en/damp-ultramarine-tick_it_DEEPGRAM-WHISPER.vtt

Artificial intelligence in Switzerland is helping fight food waste in hotels and restaurants. The first episode from the RSI news program on this topic begins in the Marriot Hotel in Zurich at breakfast time.

The leftovers from this first convivial moment of the day are not thrown away, but end up in a bin where they are first photographed and then weighed. Such a system is called Kitro: each food item is recognised, analysed and recorded by the software, providing managers with a detailed profile of the waste, in real time. This procedure helps to determine very precisely how much waste each meal generates and which foods are thrown away most frequently, thus allowing interventions to reduce waste in a targeted manner.

The system is based on artificial intelligence and makes it possible to differentiate between unavoidable and avoidable waste. Unavoidable waste includes, for example, eggshells, inedible parts of vegetables or bones. Analysing avoidable waste shows redundancies and customers' preferences. You can use this information to change menus and become more efficient.

Chef Mauricio Muñoz has learnt a lot since he started using Kitro: to cook more consciously, to buy food more economically without compromising on quality. These are key aspects when you consider that this is a kitchen producing meals for around 20,000 customers a month.

"Since we started working with Kitro, we have reduced food waste by 18%. A more than positive result. I believe this is the future of hotel kitchens. In the past, we didn't know what was going in the bin, but now we have additional eyes."

The economic aspect is central, explains programme co-director Anstasia Hofmann. "We started out following an ecological spirit, we wanted to do something positive for the environment and therefore founded our company. But there is obviously also a great economic benefit for our customers."

Fighting food waste with AI

Horror Heidi becomes popular on the internet

The comedian Karpi had an AI experiment with the Heidi story. The result is a grotesquely creepy edition.

Heidi is world-famous: Green meadows, happy cows and exuberant yodelling - the orphan girl's world embodies a romantic Switzerland. But there is another way.

Bloated cows, terrifying children's laughter, human-animal grimaces and torchlight marches in the Alpine village. Since July, a Heidi trailer reminiscent of a horror film has been viewed millions of times online and shared thousands of times. The person behind the trailer: Patrick Karpiczenko aka Karpi. The creator: the AI tool "Gen-2".

In a way, the video is the interim result of an experiment by the Swiss comedian and film producer. "I'm currently developing a TV series with digital scarecrows on the subject of artificial intelligence", says Karpi. Swiss iconography was a must. "I'm fascinated by how I can ironically break up the idyll and the pleasure."

He dictated sentences such as "six-year-old girl in a dirty old dress rejoices over green meadows in Switzerland", "photogenic horse with long hair" and "an angry crowd marches with torches through a Swiss village" to the new AI tool "Gen-2" from the US company "Runway".

But "Gen-2" didn't do a good job in his opinion. "The tool is quite bad", says Karpi, "the grotesque faces or bodies with too many limbs show how it mixes things up." For example, the goat gets a beard and the command "Ssd cow" results in a cow with a bloated belly.

Nevertheless, inspired by the bizarre images, he pushed ahead with further chimeras. The co-inventor of "SRF Deville" collected enough material to make a trailer that "lined up the bad aspects of the technology".

It's not just the Swiss comedian who is enthusiastic about the AI-generated trailer. The video has already been clicked on over 19 million times. "I didn't expect that", says Karpi, "It blew me away."

The reasons behind the enthusiasm in the Twitter community are - as with most internet trends - complex. "It needs people who perceive it as cool and daunting at the same time", says Karpi. Users probably feel reassured that AI is still working too poorly to be able to produce serious films with it.

Others celebrate the surreal image design. "Body horror", as Karpi puts it - that's what fascinates them. One user writes: "I can't decide which is the craziest and most disturbing screenshot. That's quite an achievement. 😑 The smile? The cow? The torches and pitchforks? The audience?"

That leaves the question of copyright. "My name and that of the AI. That's how I would state it legally," says Karpi. And morally?

"I simply don't know what the AI is fed." It's certainly not the Heidi animated film. The comedian figured this out when he entered "Make more Heidi in a meadow". "Gen-2" couldn't work with this instruction. It is based on stock footage, says Karpi. However, it is unclear whether the material was bought or stolen.

It also remains to be seen how similar the new Heidi horror is to the original and whether the AI infringes copyrights. Karpi thinks it's okay if AI creates strong collages "that are put together from hundreds of locations". And this is also the assumption with his Alpine horror.

Have you heard about Horror Heidi?

Where does Switzerland stand on regulating AI?

Everyone seems to agree that there must be limits on what artificial intelligence (AI) is allowed to do. The European Union and the Council of Europe are currently drawing up a set of rules. What about Switzerland? 

From social media, medicine and chatbots to semi-autonomous vehicles, artificial intelligence (AI) has become part of our everyday lives, whether we like it or not. This summer, top-level representatives from the branch sounded the alarm in an open letter about the dangers of AI. Risk reduction has to be a “global priority”, they wrote. 

The worries revolve around human rights, the rule of law and democracy, explains Angela Müller, head of AlgorithmWatch CH, a civil-society organisation that critically monitors developments in AI. The potentially negative impact of AI can already be seen, she says, pointing to the childcare benefit scandal in the Netherlands involving racial discrimination blamed on an algorithm. In Müller’s view, these kinds of cases make regulating AI “relatively urgent”. 

Thomas Schneider, vice-director of the Federal Office of Communications agrees. “Data is the new oil, and AI systems are the new motors,” he says. People are aware that these are crucial issues and that corresponding solutions have to be found, he adds. 

Nevertheless, Schneider, who is also head of the Council of Europe’s Committee on Artificial Intelligence (CAI), thinks a distinction has to be made between a viable and an instant solution, which is why Switzerland is taking a somewhat wait-and-see approach. 

On June 14, the European Parliament agreed a first draft of proposed AI legislation (the Artificial Intelligence Act). 

These laws will also cover high-risk applications, such as a ban on real-time facial recognition (as used for instance in the social credit system in China) or on language-assisted children’s toys that could encourage unsafe behaviour. 

As the “guardian of human rights”, the Council of Europe similarly feels a duty to formulate its own legal mechanisms regarding AI. Müller stresses that the Council of Europe’s framework AI convention doesn’t clash with the EU regulations, rather it supplements them, “because they follow different approaches – the EU wants to govern AI via product safety”. 

What seems obvious is that there won’t be one overall law covering AI. Schneider compares AI with engines, for which specific laws and regulations apply according to their use. The same is true of AI, where completely different applications are involved. 

This requires mixed safeguards, Schneider says, because “if you need an AI system for a music streaming service, this has different ramifications than when the same algorithm recommends which step a surgeon should make next during a heart operation”. 

For this reason one blanket law would be inadequate – a fact the EU is also aware of, he says. “The EU has around 30 proposals dealing just with the digital sector”, Schneider explains. For him, the crucial question is what aspects current laws are unable to cover. 

Müller agrees. “It’s not as though we’re currently floating around in a legal void. There are already laws in place, starting with the constitution and the protection of fundamental rights.” The goal now, she says, is to bridge the gaps that the challenges posed by AI have opened up. 

But for her the situation is similarly clear. “It’s not about having one law and then everything’s fine – that’s not going to work.” She says the problems involve anti-discrimination rights, fundamental rights, copyright law, competition law, administrative law and much more besides – in other words a whole plethora of legal areas. 

AI regulations are currently under discussion is many countries, but what approach should non-EU Switzerland take? First of all, it can be assumed that Switzerland, which has been a member of the Council of Europe since 1963, will follow convention. 

Furthermore, since the US, Canada, Japan, Israel and Mexico, among others, are also members of the Council of Europe, its regulations are likely to have far-reaching effects. 

For the moment, Switzerland’s position is to wait and see and to check the various options, Thomas Schneider said in June. And it is by no means alone. “Everyone’s looking to see if what the EU is hatching actually works.” But Schneider’s guess is that Switzerland won’t set off in a diametrically opposite direction to what the EU intends. 

In that sense Switzerland is one of “numerous typical countries that aren’t only just sitting tight but are also analysing and thinking about their options without having yet committed themselves to anything concrete”, he says. He adds that when the process gets started, it will still take years, if not decades, of adjustments.  

Switzerland is not part of the EU, like the United Kingdom since Brexit. Could Europe end up facing a patchwork of AI regulations? 

Angela Müller from AlgorithmWatch CH stresses that the aim of the EU laws is to try to prevent this at least within its own boundaries. “Despite this, the regulations laid down by the EU will also apply to external companies whenever they want to sell their products inside the EU,” she says.    

This will presumably also apply to Swiss and British firms. Moreover, the various industries will no doubt also apply a degree of political pressure in order to obtain legal securities in their sectors. 

This in turn raises the question of responsibility. Who is liable if an AI system breaks the rules? This is the topic of a scientific conference on transparency currently being held in the US. 

Computers are never the culprits, Müller points out, because they lack criminal intent. Responsibility is always borne by the people who develop the systems or employ them for specific purposes. The human actors still have to be identifiable.  

“If that becomes impossible, a core foundation of the rule of law is undermined,” she concludes. 

The above conversations took place in June at the Swiss Internet Governance Forum in Bern. 

An annual event, the platform gives experts the opportunity to discuss major digital topics in Switzerland, including the use and regulation of artificial intelligence. 

Two of the principles in the “Messages from Bern” drawn up on June 13 read: 

Edited by Balz Rigendinger. Translated from German by Thomas Skelton-Robinson  

Regulating AI in Switzerland

FINMA includes artificial intelligence for the first time in its annual report. 

According to FINMA, the state supervisory authority for the sector, there are nine risks facing the Swiss financial sector. The annual list has been updated, not least after the Credit Suisse experience, and special attention has been paid to artificial intelligence (AI), which is increasingly present in the world of finance.

Seven had already been identified last year: these are potential difficulties related to interest rates, the mortgage market, other types of credit, cyberattacks, money laundering, access to cross-border markets, as well as rising spreads (yield differentials) for corporate bonds and government securities.

Two additional risks are now prioritised: liquidity and refinancing - Credit Suisse is mentioned - as well as the outsourcing of operational activities. But another aspect also emerges: that of artificial intelligence. FINMA supports the introduction of this technology, but at the same time points out that it is still being tested in most institutions. Several particularly sensitive areas have been identified to reflect on. First, FINMA emphasises the importance of clearly defining responsibilities when decisions are made by artificial intelligence applications. Another issue raised is that of transparency: the way in which an AI has made a decision must be understandable, as must the factors taken into consideration.

https://pilot-prod-videos.s3.eu-west-1.amazonaws.com/en/due-sapphire-cockroach_it_DEEPGRAM-WHISPER.vtt

RG 12:30 of 9.11.2023: risks to the financial centre identified by FINMA / CSR