The success of Artificial Intelligence is tied to the ability to expand, not just 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 it can plunge us into a dark pit. To achieve the former, rather than the latter, it must be handled with great care and foresight. This is where technology leaders and practitioners need to pave the way and promote the use of AI to expand and strengthen human capabilities.

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

Photo: Joe McKendrick

These are some of the observations from Stanford University’s recently published report, the next issue of theirs Centennial Study on Artificial Intelligence, an extremely long-term effort to track and monitor the advancement of AI in the coming century. The report, which was first published in 2016, was drawn up by a standing committee made up of a panel of 17 experts and called for the use of AI as a tool to expand and enhance human skills. “Everyone involved needs to be involved in the design of AI assistants in order to create a human-AI team that surpasses both alone. Human users need to understand the AI ​​system and its limitations in order to trust it and use it appropriately, and AI system designers need to understand the context “in which the system will be used.”

AI has its greatest potential when it expands human capabilities, and this is where it can be most productive, the report’s authors argue. “Whether it’s finding patterns in chemical interactions that lead to new drug discovery, or helping defenders find the most appropriate strategies, there are many ways AI can improve people’s skills. An AI system might be better at synthesizing available data and making decisions on well-characterized parts of a problem, while allowing a human to better understand the implications of the data – say, whether missing data fields are actually a signal for important, unmeasured ones Information for a subset represented in the data is – – to work with difficult-to-quantify goals and identify creative actions beyond what the AI ​​may consider. ”

Complete autonomy “is not the ultimate goal of AI systems,” according to the co-authors. There must be “clear lines of communication between human and automated decision-makers. Ultimately, the success of the field is measured by how it has empowered all people, not by how efficient machines are devaluing the very people we are.” try to help.”

The report examines key areas where AI is evolving and changing work and life:

Discovery: “New developments in interpretable AI and the visualization of AI make it much easier for humans to examine AI programs more closely and to use them to organize information explicitly in such a way that a human expert can put the pieces together and gain knowledge,” says it in the report.

Make a decision: AI helps summarize data that is too complex for a person to easily absorb. “Summary is now being used or actively considered in areas where large amounts of text need to be read and analyzed – whether it be tracking news media, doing financial research, performing search engine optimization, or analyzing contracts, patents or legal advances with very realistic (but currently not reliable or accurate) text generation like GPT-3, these interactions can also make these interactions more natural. ”

AI as an assistant: “We’re already seeing AI programs that can process text from a photo and translate it so travelers can read signs and menus. Improved translation tools will make human interaction between cultures easier. Time can be made accessible to more people by searching for task and context-specific expertise. ”

Speech processing: Advances in language processing technology have been aided by neural network language models, including ELMo, GPT, mT5, and BERTwho “learn how to use words in context – including elements of grammar, meaning and basic facts about the world – by searching through the patterns in naturally occurring text. The language capability of these models already supports applications such as machine translation, text classification, speech recognition, Typing aids and chatbots. Future applications could include improving human-AI interaction in different languages ​​and situations. ”

Computer vision and image processing: “Many machine vision approaches use deep learning for recognition, classification, conversion, and other tasks. The training time for image processing has been significantly reduced. Programs running on ImageNet, a massive standardized collection of over 14 million photos used to train and test visual images. Identification programs do their jobs 100 times faster than they did three years ago. ”The report’s authors caution, however, that this technology could be misused.

Robotics: “Over the past five years, intelligent robotics has made steady progress through machine learning, powerful computing and communication skills, and the increased availability of sophisticated sensor systems. Although these systems cannot fully exploit all advances in AI, mainly 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 from five years ago about rapid advances in fully autonomous driving have not materialized. The reasons may be complicated, but the need for exceptional safety in complex physical environments makes solving the problem more difficult and expensive. ”The design of self-driving cars requires the integration of a number of technologies including sensor fusion, AI planning and decision making, vehicle dynamics prediction, flying diversion, communication between vehicles and more. ”

Recommendation systems: The AI ​​technologies that power recommendation systems have changed significantly in the past five years, the report said. “One shift is the almost universal integration of deep neural networks in order to be able to better predict user reactions to recommendations. Increasingly sophisticated machine learning techniques have also been used to analyze the content of recommended items, rather than just using metadata and user clicks or 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 fairness, diversity, polarization, and the emergence of filter bubbles, where the recommender, while these issues demand more.” than just technical solutions, attention is increasingly being paid to technologies that can at least partially solve such problems. ”


Share.

Leave A Reply