{"id":70755,"date":"2023-11-16T23:08:05","date_gmt":"2023-11-17T04:08:05","guid":{"rendered":"https:\/\/world.edu\/?p=70755"},"modified":"2023-11-16T23:08:05","modified_gmt":"2023-11-17T04:08:05","slug":"will-artificial-intelligence-kill-or-save-the-media","status":"publish","type":"post","link":"https:\/\/world.edu\/will-artificial-intelligence-kill-or-save-the-media\/","title":{"rendered":"Will artificial intelligence kill or save the media?"},"content":{"rendered":"

\"\"<\/span><\/p>\n

Information search, content production, translation, detection of hate speech… generative\u00a0<\/span>artificial intelligence<\/span><\/a>\u00a0(AI) promises significant productivity gains in the\u00a0<\/span>media<\/span><\/a>\u00a0world .<\/span><\/p>\n

The media accompany us on a daily basis and are a support for democracy: they have the freedom to show different points of view and ideas, to denounce corruption and discrimination, but also to show social and cultural cohesion.<\/span><\/p>\n

While the public turns to the media for information, culture and entertainment, the media cannot escape the economic concerns and profitability of an industry measured in terms of audience and sales.\u00a0In this context, generative AI brings powerful new tools and will be used more and more.<\/span><\/p>\n

But it is crucial to remember that generative AIs do not have ideas, and that they repeat comments which can be arranged in ways that are as interesting as they are absurd (we then speak of \u201challucinations\u201d of AI systems). .\u00a0These generative AIs do not know what is possible or impossible, true or false, moral or immoral.<\/span><\/p>\n

Thus, the profession of journalist must remain central to investigating and reasoning about complex social and geopolitical situations.\u00a0So how can media outlets leverage AI tools while avoiding their pitfalls?<\/span><\/p>\n

The National Pilot Committee for Digital Ethics (CNPEN) delivered a\u00a0<\/span>general opinion<\/span><\/a>\u00a0in July on the ethical issues of generative AI, which I co-coordinated, to the Minister responsible for the Digital Transition.\u00a0It specifies in particular the risks of these systems.<\/span><\/p>\n

Powerful tools for journalists<\/strong><\/h4>\n

The media can use AI to improve the quality of information, fight against fake news, identify harassment and incitement to hatred, but also because it can help advance knowledge and better understand realities complex, such as sustainable development or even migratory flows.<\/span><\/p>\n

Generative AIs are fabulous tools that can bring out results that we could not obtain without them because they calculate at levels of representation that are not ours, on a gigantic quantity of data and with a speed that a brain doesn’t know how to treat.\u00a0If we know how to equip ourselves with safeguards, these are systems which will save us time searching for information, reading and producing and which will allow us to fight against stereotypes and optimize processes.<\/span><\/p>\n

These tools are not arriving now by chance.\u00a0While we are effectively drowned in a flood of information broadcast continuously by traditional channels or content accessible on the Internet, tools like\u00a0<\/span>ChatGPT<\/span><\/a>\u00a0allow us to consult and produce summaries, programs, poems, etc., from a set of gigantic information inaccessible to a human brain in human time.\u00a0They can therefore be extremely useful for many tasks but also contribute to a flow of unsourced information.\u00a0We must therefore tame them and understand how they work and the risks.<\/span><\/p>\n

Learning generative AI<\/strong><\/h4>\n

The performance of generative AI depends on the self-supervised learning capacity (i.e. without being guided by a human hand, which is a different concept from real-time adaptation) of their internal models, called\u00a0<\/span>\u201cfoundation models\u201d<\/span><\/a>\u00a0, which are trained from enormous data corpora consisting of billions of images, texts or sounds very often in dominant cultures on the Internet, for example GPT3.5 from ChatGPT is mainly powered by data in English.\u00a0The other two types of learning were also used: before its availability at the end of 2022, ChatGPT was optimized using\u00a0<\/span>supervised learning<\/span><\/a>\u00a0and then using\u00a0<\/span>reinforcement learning<\/span><\/a>\u00a0by humans in order to refine the results and eliminate comments. not desirable.<\/span><\/p>\n

This optimization by humans has also been widely criticized.\u00a0How are they trained?\u00a0Who are these\u00a0<\/span>underpaid \u201cclick men\u201d<\/span><\/a>\u00a0\u00a0?\u00a0These \u201cundesirable\u201d comments, moreover, are not decided by an ethics committee or the legislator, but by the company alone.<\/span><\/p>\n

Learning that forgets the sources<\/strong><\/h4>\n

When learning foundation models on texts, the system learns what are called\u00a0<\/span>\u201clexical embedding vectors\u201d<\/span><\/a>\u00a0(of size 512 in GPT 3.5).\u00a0This is the \u201ctransformers\u201d system.\u00a0The training principle of the foundation model is based on the distributional hypothesis proposed by the American linguist John Ruppert Fith in 1957: we can only know the meaning of a word by its frequent occurrences (\u201cYou shall know a word by the company it keeps”).<\/span><\/p>\n

These entities (\u201c\u00a0<\/span>token<\/span><\/em>\u00a0\u201d in English) are on average four characters long in GPT3.5.\u00a0They can only consist of one and one blank.\u00a0They can therefore be parts of words or words with the advantage of being able to agilely combine these entities to recreate words and sentences without any linguistic knowledge (apart from that implicit in the sequence of words), the disadvantage obviously being to be less interpretable.\u00a0Each entity is encoded by a vector which contains information about all the contexts where this entity has been seen thanks to attention mechanisms.\u00a0Thus two entities having the same neighborhood will be considered close by the AI \u200b\u200bsystem.<\/span><\/p>\n

The generative AI system on texts thus learns a production model with mechanisms that have nothing to do with human production located with a body, however it is\u00a0<\/span>capable of imitating it from the texts of the learning<\/span><\/a>\u00a0.\u00a0This operation has the direct consequence of losing the sources from which the identified neighborhoods are extracted, which poses a fundamental problem for the verification of the content produced.\u00a0No verification of the veracity of the statements is easily produced.\u00a0We have to find the sources and when we ask the system to do so, it can invent them!<\/span><\/p>\n

When you provide a prompt to ChatGPT, it will predict the next entity, then the next and so on.\u00a0<\/span>A key parameter is that of \u201ctemperature\u201d which expresses the degree of randomness in the choice of entities.\u00a0<\/span>At a high temperature, the model is more “creative” because it can generate more diverse outputs, while at a low temperature, the model tends to choose the most likely outputs, making the generated text more predictable.\u00a0<\/span>Three temperature options are offered in Microsoft’s Bing conversational tool (GPT4) (more precise, more balanced, more creative).\u00a0<\/span>Often, system hyperparameters are not revealed for cybersecurity or confidentiality reasons as is the case in ChatGPT… but temperature allows different answers to the same question.<\/span><\/p>\n

\u201cHallucinations\u201d and other risks<\/strong><\/h4>\n

It is therefore easy to imagine some of the risks of generative AI for the media.\u00a0Others will certainly appear as they are used.<\/span><\/p>\n

It seems urgent to find how to minimize them while waiting for the promulgation of an\u00a0<\/span>AI Act<\/span><\/em><\/a>\u00a0for the European Union by providing ourselves with\u00a0<\/span>good practice guides<\/span><\/a>\u00a0.\u00a0The CNPEN’s\u00a0<\/span>opinion<\/span><\/a>\u00a0on generative AI and ethical issues includes 10 recommendations for research and 12 for governance.\u00a0Here are some of the risks identified for the media:<\/span><\/p>\n