Happy Nice Mall

Bender is not against the use of language models for question-answer exchange in any case. He has a Google Assistant in his kitchen that he uses to convert units of measure in a recipe. “There are times when it is very convenient to be able to use voice to access information,” he says.

But Shah and Bender did Google’s “What’s the ugliest language in India?” last year. He also gives a more disturbing example that occurs when he answers the question. “The answer is Kannada, a language spoken by about 40 million people in South India.”

there are no easy answers

There is a dilemma here. Direct answers may be appropriate, but they are also often inaccurate, irrelevant, or offensive. They can hide the complexity of the real world, says Benno Stein of the Bauhaus University in Weimar, Germany. In 2020, Stein and colleagues Martin Potthast of the University of Leipzig and Matthias Hagen of Martin Luther University in Halle-Wittenberg, Germany, published a paper. direct answer problems. “The answer to most questions is ‘It depends,’” Matthias says. “It’s hard to find someone calling.”

Stein and colleagues see search technologies as having advanced from organizing and filtering information through techniques such as providing a list of documents matching a search query to making suggestions in the form of a single answer to a question. And they think this is a step too far.

Again, the problem is not the limitations of current technology. We don’t get perfect answers even with perfect technology, says Stein: “We don’t know what a good answer is because the world is complex, but when we see these direct answers we stop thinking about it.”

The Shah agrees. Giving people a single answer can be problematic because the sources of this information and any disagreements between them are confidential, he says: “It really depends on our total trust in these systems.”

Shah and Bender propose a series of solutions to the problems they are anticipating. In general, search technologies must support the various ways people use search engines today, many of which are not provided with direct answers. Shah says people often use search to discover topics they might not have specific questions about. In this case, it would be more helpful to just provide a list of documents.

It should be clear where the information comes from, especially if an AI is drawing bits from multiple sources. Some voice assistants already do this, for example preempting an answer with “What I found on Wikipedia”. The search tools of the future should also have the ability to say, “This is a stupid question,” Shah says. This helps the technology avoid repeating offensive or biased premises in a query.

Stein suggests that AI-based search engines can offer reasons for their answers and present the pros and cons of different perspectives.

However, many of these recommendations simply highlight the dilemma that Stein and colleagues have described. Anything that reduces convenience will be less attractive to most users. “If you don’t click on the second page of Google results, you won’t want to read different arguments,” Stein says.

Google says it’s aware of the many issues these researchers raise and is working hard to improve technology that people find helpful. But Google is the developer of a multi-billion dollar service. Ultimately, it will create the tools that bring the most people.

Stein hopes that not everything will depend on comfort. “Search is very important to us, to society,” he says.