[ad_1]
AI has entered a new era, that of deployment, in many sectors of activity. Throughout 2022 and early this year, new large-scale AI models were released every month. The Stanford Institute for Artificial Intelligence has published its annual report which details the forces present, the major trends and the challenges ahead.
Models, such as ChatGPT, Stable Diffusion, Whisper and DALL-E 2, are capable of performing an increasingly wide range of tasks, from text manipulation and analysis to image generation, through voice recognition of unprecedented quality.
The proportion of companies having adopted l’IA in 2022 has more than doubled since 2017, even if it has plateaued in recent years between 50% and 60%.
According to the results of the annual McKinsey survey. This adoption seems positive as companies that have integrated it into their operations are seeing tangible benefits in terms of costs and revenue.
For the Stanford Institute for Intelligence, “AI will continue to improve and, as such, become an integral part of our lives”. But in the latest edition of its AI Index, this interdisciplinary group of experts from academia and industry emphasizes the need to be vigilant.
Industry and AI
“Given the increased presence of this technology and its potential for massive disruption, we should all start thinking more critically about how we want AI to be developed and how we want it to be developed. be used ».
Another reason to be vigilant, AI is increasingly “defined as cutting-edge technology by the actions of a small group of private sector actors, rather than by a broader range of societal actors. »
After this preamble, this index paints a picture of the current situation of AI, in order to highlight what could await us in the future. This report addresses different themes and its spectrum of analysis has been expanded from 25 countries in 2022 to 127 this year.
First observation, the industry is increasingly integrating AI. In 2022, 32 machine learning models have been produced by industry, compared to just three in academia.
AI capabilities are primarily used for robotic process automation, computer vision, text understanding and virtual agents or chatbots. Customer segmentation (19%) and support analytics are starting to incorporate AI to be more effective early by reducing costs.
Marginal progression
Publications in the areas of pattern recognition and machine learning have experienced significant growth in recent years. Since 2015, the number of articles in this area has doubled, while the number of articles on learning has almost quadrupled. This is followed by work on computer vision, algorithms and data mining.
The Institute tempers this proliferation of research by noting that although AI is improving, this progress remains marginal. “Traditionally, AI systems have performed well in narrow tasks, but have struggled with broader tasks”we read in his report.
These experts are also concerned about the serious environmental repercussions that AI could have. According to scientific research (conducted by Alexandra Sasha Luccioni, Sylvain Viguier, Anne-Laure Ligozat), all of the processes of BLOOM (a linguistic model of 176 billion parameters) emitted 50.5 tonnes of carbon.
However, new reinforcement learning models like BCOOLER show that AI systems can be used to optimize energy use.
Another point of concern is the ability of AI systems to create synthetic images which are sometimes impossible to distinguish from real images, the famous deepfakes.
[ad_2]
Source link