Microsoft Reads Your Resume (plus more from Ford & Sony)

Ford’s object-sound detection, Microsoft’s non-traditional resume finder & more.

Sign up to uncover the latest in emerging technology.

Ford’s object-sound detection, Microsoft’s non-traditional resume finder & more.

1. Microsoft – highlighting non-traditional resumes

One common problem for HR personnel is that they can’t manually sift through the hundreds and thousands of job applications that they receive for each new posting.

In the last few years, a number of tech companies have sought to help HR teams with quickly identifying which applications to inspect deeper and which to ignore.

Most of these systems work by identifying which resumes include keywords that are listed in the job description. However, this introduces many opportunities for bias. For instance, these systems would highlight candidates who use the same vocabulary as the hiring managers, which could be a function of sharing a similar social upbringing or education. Or alternatively, systems that emphasise specific hard skills (e.g. technical skills or via previous employers) might miss out on candidates who have displayed skills such as leadership and teamwork in non-traditional contexts.

Microsoft’s new patent filing looks to create a system for highlighting non-traditional resumes that don’t use the same keywords as what’s included in the job description.

They’re looking to do this by creating a ‘sentiment dictionary’ which includes keywords that belong to at least one category of interest when assessing a resume. In the filing, Microsoft list out the categories as innovation, execution, leadership and teamwork.

Microsoft would then show the frequency of terms from each category that feature in a candidate’s resume. So for instance, Jane Doe’s resume might include 70 words that display “Innovation”, while John Smith’s resume might include 30 words that display “Innovation”.

From here, HR managers could begin to sift through job applications by the frequency of words per category of interest.

Microsoft plans to keep the sentiment dictionary “unbiased” with the help of crowdsourcing and regular updates.

In practice, I think Microsoft’s filing is remarkably unambitious in trying to highlight non-traditional resumes.

As we find ourselves in economies with mass unemployment, the power dynamics of the labour market are extremely skewed. While there is very much a market-need for HR software to sift through job applications, there is a real human at the other end of the application who wants to be considered for a job opportunity. For people who come from underprivileged or non-traditional backgrounds, HR screening software tidily sweeps these applications into a black-box that never needs to be opened by a human. As someone who identifies themselves as coming from one of those types of backgrounds, I know how much human potential is possibly being screened out.

2. Ford – object sound detection for self-driving cars

Ford are looking at how to reduce their dependence on object detection sensors for self-driving cars.

Currently, self-driving cars can come into some issues if objects cannot be reliably identified. For example, an object could be out of the field of view. Or, an object may be difficult to identify if an image is blurry or there is a lot of noisiness in the visual data.

In this filing, Ford describe using audio sensors mounted to the vehicle to detect sound in the vicinity of the car. Based on where the sound is judged to be coming from, and what the source of the sound is estimated to be, the self-driving system can make any necessary adjustments to the vehicle.

For example, the sound of an emergency vehicle might be detected to be behind the car. In turn, Ford’s self-driving system would steer it into another lane to enable the emergency vehicle to pass by without a collision.

Similarly, Ford’s audio system could be trained to listen out for the sound of bicycles, other cars, trains, buses and pedestrians. Depending on the sound and where its location is deemed to be in relation to the car, the Ford vehicle’s steering, braking or acceleration would be adjusted.

With self-driving cars coming under some heat with crashes in the last few years, it’s interesting to see Ford looking to supplement the traditional self-driving object detection sensors with other forms of data such as sound.

3. Sony – users generating alternative movie endings

A much overlooked genre of fiction is fan-fiction.

Fan-fiction is where fans write alternative stories based on the intellectual property of existing stories. For example, have 625,000 user-submitted alternative stories built around the Harry Potter universe.

Sony is looking at doing something similar with movie endings.

Their latest filing would enable viewers to use their creativity, imagination and vision to generate alternate movie endings or whole movie sequences. These creations might then be shared and appreciated by other creators in an online community.

While the specific details are a little vague, Sony describes using a Neural Network that takes user inputs and combines them with materials from the original movie to simulate entirely new scenes.

User inputs could be as simple as a user narrating their desired scene, speaking the text they want the characters to speak, and describing the general setting.

This filing shows that Sony is looking to redefine the relationship between consumption and creation when it comes to movies. In much the same way that TikTok Duets have turned the consumption experience into a potential creation opportunity, or the way that Twitter Quote Tweets turned reading into a process for generating entirely new forms of content (i.e. the dunk), Sony are empowering viewers of movies to be potential creators. In turn, one movie could theoretically be forked into millions of movies with an obsessive community deeply engaging with that movie’s IP (and possibly generating new IP?).

I imagine this filing is more on the ‘hopeful’ end of the spectrum for product features coming soon. That said, it’s fascinating to see another example of how some companies are thinking about the applications for AI simulations. In a previous issue, I looked into a Facebook filing that was looking to use Generative Adversarial Networks to generate personalized ads from scratch.