"They compose songs, chat on the phone and write novels. Artificial intelligence (AI) is getting ever closer to humans. We present five of the most astonishing application examples.
Artificial intelligence like in a Hollywood film. This is how Sam Altman himself recently described OpenAI's new language model. The co-founder and CEO of the American company also mentioned a very specific film, the 2013 science fiction romance "Her". In it, a man played by Joaquin Phoenix falls in love with an AI program called Samantha with the voice of Scarlett Johansson. The reference has not aged well given the actress's legal disputes with OpenAI.
The software seduces the man in the film with its human-like qualities; it flirts and has deep conversations with him; he confesses to her that he has never loved anyone as much as she. The story does not have a happy ending. Samantha leaves the man and he is left heartbroken. Altman called the film "prophetic". And with the new AI model GPT-4o, which will now be used in the ChatGPT chatbot, he has brought the vision of "Her" a lot closer to reality just over a decade later.
Because what is striking about this system is not the tasks it can perform, but how it communicates with its users via its speech function. Whether translating texts into other languages, tutoring in math or telling bedtime stories: the product demonstrations resembled natural conversations and the female voice made you think of Samantha from "Her" - even the actress herself.
The voice spoke as if it had emotional intelligence. Sometimes it flirted, sometimes it spoke with sarcasm. Sometimes it was louder and sometimes quieter, cleared its throat and used filler words like "Mmh". Similar to a normal conversation between people, it could also be interrupted. It is a dramatic difference to robotic and often not very helpful voice assistants like Siri from Apple or Alexa from Amazon. Altman said of GPT-4o: "I'm still a little surprised that it's real."
Reliability and possible downsides of AI
The current AI arms race in the technology industry has reached a new dimension in recent weeks. Open AI and Google, two of the most important companies in the field, have had a showdown with new products, and both have conveyed the message that they want to establish AI with power in everyday life. Open AI did this at the beginning of the week with its more human-like chatbot, and Google just a day later with an AI-focused further development of its search engine. In the future, the company wants to focus on a direct answer for many queries, which is provided by its AI technology, and the traditional links to other websites are slipping down.
There are still many glitches. Nevertheless, the heart of Google's business, which is the first port of call for many people on the Internet, is getting a new shape.
The company also demonstrated AI assistance programs that it is working on. They did seem a lot more machine-like in their conversations with users than the new model from Open AI. However, CEO Sundar Pichai said that some people would develop an emotional connection to such services in the future.
The announcements from Open AI and Google carry particular weight. But in the wake of these industry giants, more and more other companies are also pointing the way to a new AI world that will bring profound changes to everyday life. We present some examples here. Questions about the reliability and possible downsides of these technologies are therefore all the more urgent. And for some people, it may be a disturbing idea if the AI pioneer OpenAI sees the future as described in "Her" as an inspiration for its work.
AI companions: friendship, advice and love
AI systems can be more than neutral information providers. Many services make it possible to give chatbots personalities and then use them for social purposes, whether as virtual friends, advisors or even romantic partners.
Replika, for example, offers its users an "AI companion who cares," while Candy.ai promises an "interactive erotic experience" with an AI girlfriend who can fulfill "any fantasy."
There are various AI characters on Character.ai. A particularly popular chatbot is called "Psychologist," which claims to help with "difficulties in life." The online publication "The Verge" wrote that many teenagers used the platform, for example as a remedy for loneliness, or to tell AI characters things that they keep from real friends. Such offers are of course not without controversy. Advice from AI systems can be questionable or even dangerous, and it is not for nothing that Character.ai expressly points out that everything its characters say is "invented".
There are concerns that such services are addictive and tempt users to spend even more time in the virtual world rather than the real world. However, a recently published study at Stanford University in California among young users of Replika also suggested that virtual companions could have a positive effect, especially on people who feel lonely. Some of those surveyed even said that the AI service helped against suicidal thoughts. In any case, the market for AI companions of all kinds seems to be growing, and it is also apparently lucrative - especially when it comes to offers of a romantic nature. A recent report in the New York Post mentioned a man who spends $10,000 a month on AI girlfriends.
Music: Songs at the touch of a button
AI-generated voices have long since arrived in the music world. The song "Heart On My Sleeve" achieved dubious fame in April last year because its author had it sung with the AI-generated voices of Drake and The Weeknd - without their permission. There are currently no widely applicable tools on which stars make their voices available. However, YouTube is currently testing the 'Dream Track' application, which users can use to create song snippets with the voices of selected artists. Not only the remuneration, but also the control over the lyrics that an AI voice sets to music are complex topics.
Regardless of voices, various tools promise to help with music production. Suno AI and Udio are currently causing a stir. Both tools generate complete songs on command - sometimes in astonishingly good quality. Users only have to specify the subject of the text and genre, possibly refined with the desired emotion or tone of voice.
The music industry is generally open to AI and the possibilities arising from its use, for example with regard to new products. However, the industry insists on clear rules and limits. Above all, AI models may not be trained with pieces of music without the consent and subsequent remuneration of rights holders.
Sony Music recently underlined this in a letter to more than 700 companies and also demanded that they disclose whether and, if so, which protected material had already been used. Neither Suno AI nor Udio have yet made public how their models were trained. And while Drake, for his part, got into trouble for using the voice of the late rapper Tupac, AI sounds are likely to be used more in the future, especially in the area of production music for television or social media.
Customer service: Almost perfect on the phone
Everyone knows it, everyone finds it annoying: the computer voice in the call center that, even after the seventh attempt, still doesn't understand what the problem is. Thanks to the improving language skills of artificial intelligence, this could soon be history. Because the new AI agents are not only able to follow predetermined patterns like a classic voice bot. They can also respond to customers personally.
In North America, for example, the automotive supplier Bosch, together with the German AI start-up Aleph Alpha, introduced an AI-based voice bot at a car manufacturer. This is intended to take breakdown service calls. The AI also records dialects, accents and moods. Previous speech recognition systems often had problems with this. The machine can process 40 percent of calls automatically. For more complex inquiries, it still hands over to a human.
The Berlin start-up Parloa has even specialized in AI for call centers, recently raised 66 million euros from investors and has expanded into the United States. OpenAI, however, took automation to the next level in customer service. In a video, an employee instructs ChatGPT via voice command that he is about to call his mobile phone provider and that ChatGPT should please tell him that the iPhone he sent him is defective and that they should send him a new one. Next to it is another smartphone with ChatGPT that simulates customer service. The two chatbots then have a conversation with each other and resolve the problem - from AI to AI.
If this capability ever becomes marketable, people would no longer have to deal with robot voices that may not work well - but with an AI.
Translation: Foreign languages without learning
Speaking a foreign language without having to learn for it? Artificial intelligence makes it possible - at least with a slight time delay. Last year, the American start-up Hey Gen caused a stir on the Internet. Users simply upload a video of themselves speaking in their native language to the company's application. Then they select the language into which the software should translate the video. The results are astonishingly good.
The AI also adapts the speaker's lip movements to the new language, so that at first glance it is not obvious that artificial intelligence was at work. The application can now translate into more than 40 languages. Videos of users trying out Hey Gen quickly flooded the Internet. This type of technology could possibly take over the work of translation studios and voice artists in the future. However, Hey Gen's translation does not yet work in real time.
Last year, the Internet company Google introduced the "Universal Translator", a program that can translate live in lip sync. However, such applications pose the risk of politically or criminally motivated forgeries. Instead, vacationers will probably be able to rely on ChatGPT's voice assistant in the future. Thanks to the new capabilities of the AI model GPT-4o, ChatGPT can now play the role of interpreter, as OpenAI recently demonstrated.
An employee asked the AI voice whether it could translate his comments from English to Italian and the comments of his conversation partner from Italian to German. It worked perfectly. Those who prefer the classic approach can use the text-to-text translations of the Cologne-based start-up Deep L, which have also improved significantly in recent years thanks to AI and are now valued at two billion dollars.
Fan fiction: Old literature in new stories
What actually happened to the Hagenström family, the rising family that economically and socially displaced the Buddenbrooks in Thomas Mann's work of the same name? Where did the Wall Street broker Sherman McCoy from Tom Wolf's novel "The Bonfire of the Vanities" build his new life? Will the reformed moneylender Ebenezer Scrooge from Charles Dickens' "A Christmas Carol" transfer his fortune to Tiny Tim Cratchit, and what will the son of his employee, who was bullied and humiliated for years, do with all that money?
Characters that were once created by their authors as minor characters have been given new stages for several years now - thanks to so-called fan fiction. Fascinated readers simply continue to fantasize about the lives of their favorite characters. From literary classics to Star Wars and Japanese anime. The stories of the master detective Sherlock Holmes were continued by his readers after the death of his creator Arthur Conan Doyle in 1930. To date, fans of Harry Potter, with the blessing of his creator J.K. Rowling, have written no fewer than 800,000 independent stories about the main characters and minor characters at the sorcerer's apprentice's side.
This trend has been fueled by AI language models for months. In addition to AI-created literature from the pen of artificial systems such as Squibler AI or Sudowrite, there are also fan fiction tools such as AI Fanfic Generator or Holo AI. Half a century ago, the playwright Tom Stoppard took Shakespeare's "Hamlet", elevated the minor characters Rosencrantz and Guildenstern to his main characters and gave them a big appearance shortly before their murder, but AI decided on a completely surprising twist to the story of "Ros and Guild." [1]
1. Was die Künstliche Intelligenz inzwischen kann: Sie komponieren Songs, schwatzen am Telefon und schreiben Romane. Systeme der Künstliche Intelligenz können inzwischen aber noch viel mehr - und das verblüfft nicht nur die Macher. Frankfurter Allgemeine Zeitung (online) Frankfurter Allgemeine Zeitung GmbH. May 17, 2024. Von Stephan Finsterbusch, Benjamin Fischer, Roland Lindner und Maximilian Sachse
