The success of artificial intelligence is linked to the ability to increase, not just to automate

0

Artificial intelligence is just a tool, but what a tool it is. It can lift our world into an era of enlightenment and productivity, or plunge us into a dark abyss. To help achieve the former, and not the latter, it must be handled with great care and forethought. This is where tech leaders and practitioners need to step in and help lead the way, by encouraging the use of AI to increase and amplify human capabilities.

cn-tower-toronto-photo-by-joe-mckendrck.jpg

Photo: Joe McKendrick

These are some of the observations from Stanford University’s recently released report, the next installment in its 100-year study of artificial intelligence, a very long-term effort to track and monitor AI as it unfolds. as it progresses over the next century. The report, first launched in 2016, was prepared by a standing committee that includes a panel of 17 experts, and urges using AI as a tool to increase and amplify human skills. “All stakeholders should be involved in the design of AI assistants to produce a human-AI team that outperforms one or the other on its own. Human users must understand the AI ​​system and its limitations in order to trust it and use it appropriately, and AI system designers must understand the context in which the system will be used. “

AI has the greatest potential when it augments human capabilities, and this is where it can be most productive, say the authors of the report. “Whether it’s finding patterns of chemical interactions that lead to the discovery of a new drug or helping public advocates identify the most appropriate strategies to pursue, there are many ways in which the AI can increase the capabilities of people. An AI system might be better at synthesizing available data and making decisions in well-characterized parts of a problem, while a human can better understand the implications of the data – say if the missing data fields are in fact a problem. signal important, unmeasured information for a subgroup represented in the data – – work with goals that are difficult to fully quantify and identify creative actions beyond what the AI ​​can be programmed to account for. ”

Complete autonomy “is not the end goal of AI systems,” say the co-authors. There must be “clear lines of communication between human and automated decision makers.” Ultimately, the domain’s success will be measured by how well it has empowered everyone, not how effectively machines devalue the very people that we are. try to help. ”

The report examines the key areas where AI is growing and making a difference in work and life:

Discovery: “New developments in interpretable AI and AI visualization allow humans to more deeply inspect AI programs and use them to explicitly organize information in a way that allows a human expert to put the pieces together and draw ideas from them, ”the report notes. .

Decision making: AI helps summarize data that is too complex for a person to easily absorb. “Synthesis is now being used or actively considered in areas where large amounts of text need to be read and analyzed, be it media monitoring, financial research, search engine optimization, or research. ” analyze contracts, patents or legal documents. advancements in the generation of very realistic (but currently unreliable or accurate) text, such as GPT-3, may also make these interactions more natural. ”

IA as an assistant: “We are already starting to see AI programs capable of processing and translating text from a photograph, allowing travelers to read signage and menus. Improved translation tools will facilitate human interactions between cultures. amounts of time can become accessible to more people by allowing them to seek expertise specific to a task and context. ”

Language processing: Technological advances in language processing have been supported by neural network language models, including ELMo, GPT, mT5, and BERT, which “learn how words are used in context – including elements of grammar. , meaning and basic facts about the world – to sift through the patterns in the natural text. The ease of these models with language already supports applications such as machine translation, text classification, speech recognition, writing aids, and chatbots. Future applications could include improving human-AI interactions in various languages ​​and situations. ”

Computer vision and image processing: “Many image processing approaches use deep learning for recognition, classification, conversion, and other tasks. Training time for image processing has been significantly reduced. The programs run on ImageNet , a massive standardized collection of over 14 million photographs used to train and test identification programs, are completing their work 100 times faster than just three years ago. ” The report’s authors warn, however, that such technology could be subject to abuse.

Robotics: “The past five years have seen steady advancements in intelligent robotics driven by machine learning, powerful computational and communication capabilities, and increased availability of sophisticated sensor systems. While these systems are not fully able to take advantage of all advances in AI, primarily due to the physical constraints of the environments, highly agile and dynamic robotic systems are now available for home and industrial use. ”

Mobility: “The optimistic predictions of five years ago for the rapid progress of fully autonomous driving have not come true. The reasons can be complicated, but the need for exceptional levels of security in complex physical environments makes the problem harder and more expensive to solve. Self-driving car design requires the integration of a range of technologies including sensor fusion, AI planning and decision making, vehicle dynamics prediction, on-the-fly rerouting, communication between vehicles, etc.

Recommendation systems: The AI ​​technologies powering recommendation systems have changed dramatically over the past five years, the report says. “A change is the almost universal incorporation of deep neural networks to better predict user responses to recommendations. There has also been an increased use of sophisticated machine learning techniques to analyze the content of recommended items, rather than using just metadata and user clicks or clicks. consumer behavior. ”

The report’s authors caution that “the use of increasingly sophisticated machine-learned models to recommend products, services and content has raised significant concerns about issues of equity, diversity, polarization and the emergence of filter bubbles, where the system recommender suggests. Although these problems require more than just technical solutions, increasing attention is being paid to technologies that can at least partially solve these problems. ”


Source link

Leave A Reply

Your email address will not be published.