Vogon Today

Selected News from the Galaxy

StartMag

Everyone is crazy about artificial intelligence, which has one problem: it learns from us

Everyone is crazy about artificial intelligence, which has one problem: it learns from us

Mario Marchi's post

And suddenly everyone realized that artificial intelligence is among us. She learned a lot, even more we taught her without our knowledge. We will have to teach her a lot, perhaps asking ourselves what we expect from her.

THE CHATGPT CASE

As often happens, the great awareness was followed by the "pop" declination of the technology, ChatGPT , accompanied by a social campaign that rebounded all over the world within a few days. After all, an artificial intelligence that literally talks to the user, responds amicably to his questions, providing news and data like no search engine has ever been able to do, moreover packaged in already elaborated and complex texts, could only have success.

This is the Chatbot, an AI system that communicates, making available a wealth of continuously updated data, partly by the programmer, but above all through real self-learning, precisely on the basis of the questions that are asked.

The result is that millions of users have connected to ChatGPT, often crashing it and already giving rise to some problems in real life, given that in the US the school system is taking measures to stem the use of Chabot by students, to have tasks, themes and research produced.

In short, a beautiful revolution, of which – however – we notice decidedly late.

There are dozens of applications similar to ChatGPT that have been available for some time: they do everything from writing poems on commission, to automatically producing posts for social networks, up to composing music to order. Some are even downloadable to your smartphone.

THE SOCIAL IMPACT OF ARTIFICIAL INTELLIGENCE

As far as the social impact is concerned, then, much more has already happened and much more than the tasks commissioned by crafty students.

In 2017, The Washington Post newspaper launched an automated news service called Heliograf, which uses artificial intelligence to generate stories about sporting events, elections, and other topics.

In 2018, the financial newspaper Forbes used artificial intelligence to generate a series of articles on market trends. Articles were written by the AI ​​in minutes, based on market data and relevant news events.

A year earlier, an AI developed by Stanford University had even written a science fiction book called "Shelley", which was published on Amazon.

Staying in the literature, in 2020, an IDEA developed by the company Primer Labs wrote a book of poems called “From Darkness to Light”.

So far the curious uses, but decidedly more serious drawbacks have not been lacking.

For example, recruiting firm HireVue was accused of discriminating against female candidates because its AI system was trained on historical data showing a higher likelihood of success for male candidates and elaborated searches and selections Consequently.

The United States government itself, in 2019, had to take a step back in the use of very effective artificial intelligence systems for assessing the risk of recidivism of prisoners, since it had been demonstrated that these systems discriminated against ethnic minorities, again due to of the elaboration of historical data and current news.

In short, today we are passionate about chatting with artificial intelligence, but we must know that it is not always our friend. It depends on us, on what and how much we teach them. In this sense, popular use does not promise very well: in social networks there are countless suggestions for using other Chatbots that do not have the "ethical filters" of ChatGPT.

In fact, the OpenAI programmers have instructed their creature not to provide answers to violent or morally censurable requests. But apparently there are already those who would like an artificial intelligence capable of giving the worst of itself. And about us.


This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/innovazione/tutti-pazzi-per-intelligenza-artificiale-che-ha-un-problema-impara-da-noi/ on Tue, 17 Jan 2023 10:36:57 +0000.