Artificial intelligence has come a long way, but it needs to meet safety criteria
Washington Post, March 7, 2022
Despite the fact that smart speakers have become widespread by now and can be found in almost every second home, their behavior is still far from ideal.
However, despite widespread use, voice assistants often do not understand their owners, misinterpret what is happening, incorrectly determine the context of situations and, in general, provoke a lot of sting from users.
Interacting with smart voice assistants requires emotional labor and cognitive effort. It creates the kind of work we don’t even know what to call it,”- admits Erica Hall, co-founder of consulting studio Mule Design Studio. Her company advises companies on best practices for creating conversational interfaces.
Some voice assistants not only do not simplify, but also complicate the work if they are asked to do even something insignificant – for example, shopping with Alexa can end up reading the characteristics of each product for a long time, which in the end is faster to make purchases on your own.
Everything is really not so bad when it comes to complicating some household chores – but there were also cases when assistants not only interfered – their advice could simply do harm, as, for example, in the recent case with Alexa, who advised the little girl to insert a metallic coin into an outlet.
Read more about the difficulties of interaction with voice assistants in the article.
Inside Big Data, March 8, 2022
Shayak Sen, Co-founder and CTO, TruEra talks about how data scientists who put a lot of effort into their work cannot guarantee the quality of their models in production – and this is a serious problem in the industry.
Nowadays, many people are talking about the fact that machine learning models do not work properly, which leads to serious consequences. Just remember the situation with Zillow when, as a result of an AI fails, the company had to lose its iBuying business, and its shares fell by 30% Zillow is not alone. Amazon and Apple have also had similar negative experiences.
In such situations, some people think that the problem is that the specialists did a bad job, which led to a poor-quality result, but this is not always the case. The model may seem effective in the lab, but in the real world its performance may differ, the performance of the model may degrade over time, or the actions of attackers may cause negative changes.
Shayak began building industrial-grade machine learning models for algorithmic trading 10 years ago, when he thought the systems needed to be made understandable, privacy-friendly, and fair. It is his research that is the basis of a number of innovative breakthroughs in the field of explainable AI. Read more about the article with reflections on the topic of AI that can and cannot be trusted.
Taipei Times, March 13, 2022
Creators of sexual content involving anyone without their permission could face up to three years in prison.
The Executive Yuan approved a bill aimed at controlling the creation of “deepfake” pornography and measures to combat it. According to the new amendments to the law, the production and distribution of fake images and videos for the purpose of obtaining officially will be considered a crime with a penalty of up to seven years in prison.
It is important that the law includes, among other things, the use of machine learning to create fake
materials.
While there have been many instances of fake porn creation in the past, new fixes have been proposed since the arrest of the Taiwanese YouTuber. The man was suspected of creating and selling deepfake pornographic videos featuring more than 100 public figures.
Stay up to date with what is happening! Get a first look at news, noteworthy research and worst attacks on AI delivered right in your inbox.
As smart systems become increasingly common in the realm of national security, risk issues require more attention The benefits of peripheral vision for machines MIT News, March 2, 2022 There ...
