Application testing company Applause recently launched a new artificial intelligence (AI) solution that promises to help solve algorithmic biases while providing the huge amounts of data needed for AI training.

Applause has built a huge global testing community for its application testing solutions, which are trusted by brands such as Google, Uber, PayPal. The company is now using this relatively unique position to help overcome some of the biggest obstacles to AI development.

At the end of November, Kristin Simonini, vice president of products at Applause, gave an interview before the keynote speech at the North American AI Expo, discussing the company's latest solutions and their significance to the industry.

Simononi explained: "Our customers always need us to provide additional support in the field of data collection to support their AI development, train their systems, and then test the functions, while the second half is more in line with what they traditionally do to us. Expect. "

Applause works primarily with companies in the voice field, but they are also increasingly expanding into areas such as collecting and labeling images and running documents through OCR (Optimized Character Recognition).

In the most commonly used areas of AI today, this breadth of existing experience puts Applause and its testers in a great position to provide really useful feedback on areas where improvements can be made.

Specifically, Applause's new solution spans five unique types of AI activities:

Voice: The source speaks to train voice-enabled devices, and tests these devices to ensure that they understand and respond accurately;

OCR: Provide documents and corresponding texts to train algorithms to recognize text, and compare the accuracy of printed documents and recognized texts;

Image recognition: deliver photos of predefined objects and locations, and ensure that pictures and objects are identified correctly;

Biometrics: Obtain biometric inputs, such as faces and fingerprints, and test whether these inputs produce an easy-to-use and practically effective experience;

Chatbots: Give sample questions and different intentions for chatbots to answer and interact with chatbots to ensure they can understand and respond as accurately as humans.

"We have a well-prepared global community that can bring together, on a large scale, any information an organization might be looking for, and do so in a combination of breadth and depth that makes the introduction very different Data to train AI systems becomes possible. "

Simononi provided some examples, including voice utterances, specific documents, and images that met set standards (such as "street corners" or "cats"). These data types were provided by Appleause's global testers. The lack of such a diverse data set is one of the biggest obstacles facing us today, and one that Applause hopes to help overcome.

Heavy responsibility

Everyone involved in developing emerging technologies has a significant responsibility. AI is particularly sensitive because everyone knows that it will have a huge impact on most societies in the world, but no one can really predict how it will affect it.

How many jobs will AI replace? Will it be used for killing robots? Does it decide whether to launch a missile? To what extent will facial recognition be applied to society as a whole? These are important questions, and no one can give a completely positive answer, but the films surrounding "1984" and "Terminator" definitely affect the public's thinking.

One of the main issues with AI is bias. Work done by institutions such as the Algorithm Justice League has revealed that the effectiveness of facial recognition algorithms depends on the huge differences between each person's race and gender. For example, IBM's facial recognition algorithm has an accuracy rate of 99.7% when used on men with light skin, but only 65.3% for women with dark skin.

Simononi highlighted another study she recently read in which the algorithm recognized white males with speech accuracy rates exceeding 90%. However, for African-American women, this percentage is just close to 30%.

Addressing this difference is critical, not only to prevent things like inadvertently automating racial characterization or giving certain parts of society an advantage over others, but also to allow AI to reach its full potential.

Although there are many concerns, as long as it is developed in a responsible manner, AI has tremendous power. AI can improve efficiency, reduce the impact on the environment, allow people to spend more time with their loved ones, and fundamentally improve the lives of people with disabilities.

A company's failure to take responsibility for its own development will lead to over-regulation, which in turn will lead to less innovation. When asked if he believes that robust testing will reduce the possibility of over-regulation, "in some cases people may try to regulate, but if you can really prove that an effort has been made To reach a high level of accuracy and depth, then I think this possibility will be reduced. "

Human testing is still essential

Applause is not the only company working to reduce bias in algorithms. For example, IBM has a tool called Fairness 360, which is essentially an AI system that scans other algorithms for signs of prejudice.

Asked why Applause believes that human trials are still essential, Simoniini commented: "Humans are The interactive aspect of the app is unpredictable. We haven't seen signs of doing it effectively without human factors. "

A major challenge often encountered in speech recognition is the languages spoken and their regional dialects. Many U.S. speech recognition systems are even problematic in recognizing accents in southwest England.

Simoniini added another consideration about slang and the need for voice services to keep pace with the changing vocabulary. She explains: "When teens like something hot or cool, they like to use the word 'Fire'. We can bring these devices into the home and really try to understand some of these nuances."

Simononi then further explained the challenges in understanding the background of these nuances. In her "Fire" example, she obviously needed to understand when it was interpreted literally and when someone was praising something cool. "How do you distinguish between emergencies such as fires? The tone, tone, and other things about how to use the same voice command will be different," Simononi said.

AI application and service growth

Applause has built its own business in the traditional application testing space. Considering the expected growth in AI applications and services, Simoni was asked if he believed his AI testing solution would become as large as it is today, or perhaps even larger.

In response, she responded: "We did talk about this. You know how fast this will grow? I don't want to talk about voice all the time, but if you look at the statistics, the growth and adoption of mobile The growth of the voice market is happening at a much faster rate. I think it will take an increasing share of our business, but I don't think it will definitely replace anything because these channels (such as mobile and desktop Applications) will still exist and complement each other. "

Simononi also said: "The angle we choose to talk about is actually the intersection between humans and AI, and why we don't believe it will become a substitute, but how it works and complements each other. Basically, from the test perspective, human-centered demand is still very high "(from: artificialintelligence-news author: Ryan Daws compilation: NEW YORK intelligent participation: small).

Source: Netease Intelligence, translated by Google Translate

Statement: this information is reprinted from authoritative news media. Reprinted for the purpose of transmitting more information and academic exchange, it is not used for commercial purposes, and does not mean to agree with its views or confirm its description. The content of this article is for reference only. If you violate the rights and interests of a third party, please contact us and we will deal with it as soon as possible.