AI updates in brief

 Judges versus AI. European lawmakers are working to reduce the power of AI and  AI graduates medical but can it work with its human colleagues?

Judges vs AI

European lawmakers have proposed a new charter that will force engineers to build kill switches into their robots. The kill switch is one of the ways the lawmakers are making sure engineers build ethical and safe machines. 

AI graduates college but can it work with its human colleagues?

When the Duke hospital installed a machine learning software to tell nurses if someone was at risk of developing sepsis, it left doctors confused. One of the doctors said, “it wouldn’t give good answers because it is based on algorithm”

AI powered robot built helps in the fight against COVID-19

A robot called Breezy One is spraying airports with disinfectant. The AI powered robot rolls out after most passengers must have left the terminal and sprays the disinfectant over surfaces in the airport.

YouTube announces the use of ai to age-restrict inappropriate videos

YouTube will now make use of artificial intelligence to age-restrict videos that are not suitable for children. Currently, the company uses human reviewers to identify videos that should not be watched by underaged viewers.  Soon, YouTube will be making that decision via machine learning. Although some videos may be wrongly flagged by the system’s AI, this will not be an issue as the uploaders can contact YouTube if they feel their videos were wrongly flagged as inappropriate for viewers under the age of 18. This move will help the video hosting site efficiently and more conveniently regulate the kind of content of the platform.

Nvidia launches imaginaire: new library for image and video synthesis, and other gan related tasks

Nvidia researchers have released a new universal Pytorch library called Imaginaire. The new library features an optimized implementation of various kinds of Generative Adversarial Network (GAN) images and videos. The library currently features three kinds of model implementations including:

  • Unsupervised Image-to-image translation
  • Supervised Image-to-image translation
  • Video-to-video translation

The library includes several state-of-the-art algorithms from each of the three domains, including: UNIT and MUNIT from the unsupervised image-to-image translation, SPADE from the supervised image-to-image translation, and vid2vid, wc-vid2vid from the video-to-video translation, etc.

Princeton university researchers have developed a new tool to identify bias in computer vision

The computer vision applications of AI have been under serious threat due to the issue of bias in AI systems. Earlier this year, Amazon, Facebook and Microsoft stopped the commercialization of their facial recognition software and banned the police from using it. This came after a lot of black people, women-of-colour, and other minorities were wrongfully convicted and arrested for a crime they didn’t commit because of the bias in AI systems. Now, researchers at Princeton University are addressing this significant issue with Computer Vision software, through the development of a new tool known as REVISE(REvealing VIsual biaSEs), which leverages statistical methods to recognize potential biases in a dataset. The tool can help prevent the biases of AI against geographically-underrepresented people, racial minorities and gender minorities within a given territory. 

AI can detect Covid-19 in the lungs through computed tomography

A new study by the University of Central Florida has shown that Artificial Intelligence(AI) can detect Covid-19 pneumonia in Computed Tomography (CT) scans almost as accurately as a physician. In the study, AI algorithms were trained on the scans of 1,280 patients, to identify the Covid-19 pneumonia. When tested on a new dataset of 1337 patients, the AI system was able to detect the Covid-19 pneumonia with an accuracy of 90.8%. Among the test cases, the AI was able to correctly identify 84% of the positive Covid-19 cases, and 93% of the negative cases. 

AI may help us create new Shakespeares: AI learns from Shakespeare’s sonnet to develop unique poetry

To celebrate the National Poetry Day, data scientists at Zyro embarked on an experiment to create AI-written original poetry. The AI was trained on about 150 Shakespearean sonnets for weeks, to mimic and replicate Shakespeare’s style of writing. Here is the first Shakespearean piece created by Zyro’s AI:

Thy brow is gold,

And I see thy glassy face bright,

Who by thy wrinkled brow hath much seen.

And in thy glazed eye I saw a part

Of life that could bear records

Of thy friend, for to age thyself must

Despise vain debate, and to lack

That argument with store, though yet barren.

Thy blackened brow cannot show me this,

Instance of apparel may blot out

As black a point as this, though thy life doth

Heap upon life and lives upon life.

For thus had this friend age’d,

And old age hath no garments to tell it.

The performance of the AI was impressive, although it was unable to perfectly grasp the Bard’s writing signature.

Goodbye to hold music: Google’s new ‘Hold For Me’ feature helps users know when to get back on a call after waiting on hold.

For a lot of people, calling a customer care line or any administrative office usually comes with long periods of waiting and listening to cliché music, before a human representative eventually picks up. Google has released a new feature to change this. With Google’s new feature, Hold For Me, users can drop their phones and engage in other activities while they are placed on hold, without putting the phone on loudspeaker to monitor the call before a human representative gets on the call. With the feature, the Google Assistant will call for you and notify you with a vibration, sound, and a screen notification when a human representative gets on the line. The Hold For Me feature is currently available on Google’s Pixel 5 and Pixel 4a phones in the United States.

 AI helps in beer creation in Switzerland

AI is here to transform the future. With time, every limitation on what is possible with AI will fade off. Swiss microbrewery, Deeper, from MNBrew, in conjunction with Lucerne University of Applied Sciences and Arts (HSLU) and the software company, Jaywalker Digital, has created an AI-developed beer in Switzerland. Although the beer was still physically produced by humans, the entire recipe, which determines the outcome of the beer, was solely developed by an AI, known as Brauer AI.  For this project, the Indian Pale Ale Beer was chosen. The AI algorithm analyzed a database of 157,000 international beer recipes. With the aid of neural networks, the AI was able to understand the patterns in the recipes and it was able to select the malt type and their proportions, and the suitable hops and their cooking times. According to the researchers, Brauer AI can also provide suggestions for a beer that has really never been seen before. 

Microsoft releases new tool to detect deepfakes

AI is getting smarter by the day and it is becoming increasingly difficult to identify computer-manipulated images/videos created via deep learning and AI, known as deepfakes. 

Microsoft has recently announced a new tool, known as Microsoft Video Authenticator to identify deepfakes and prevent misinformation, particularly ahead of the 2020 US election. Microsoft’s new tool can analyze videos or images and determine a confidence score to identify if they are artificially manipulated. 

Learn more on AI here

Decentralize Newsletter

From Crypto and Blockchain to AI, Fintech and Web 3.0 delivered twice in a week (Mondays and Fridays)

Thank you for signing up for Decentralize Brief newsletter.

Decentralize Newsletter

From Crypto and Blockchain to AI, Fintech and Web 3.0 delivered twice in a week (Mondays and Fridays)

Thank you for signing up for Decentralize Brief newsletter.

Decentralize Newsletter

From Crypto and Blockchain to AI, Fintech and Web 3.0 delivered twice in a week (Mondays and Fridays)

Thank you for signing up for Decentralize Brief newsletter.

Be the first to comment

Leave a Reply

Your email address will not be published.