Things tagged ai:

Looking to Listen: Audio-Visual Speech Separation

Inbar Mosseri and Oran Lang at Google Research:

People are remarkably good at focusing their attention on a particular person in a noisy environment, mentally “muting” all other voices and sounds. Known as the cocktail party effect, this capability comes natural to us humans. However, automatic speech separation — separating an audio signal into its individual speech sources — while a well-studied problem, remains a significant challenge for computers.

In “Looking to Listen at the Cocktail Party”, to appear in SIGGRAPH 2018 this summer, we present a deep learning audio-visual model for isolating a single speech signal from a mixture of sounds such as other voices and background noise. In this work, we are able to computationally produce videos in which speech of specific people is enhanced while all other sounds are suppressed. Our method works on ordinary videos with a single audio track, and all that is required from the user is to select the face of the person in the video they want to hear, or to have such a person be selected algorithmically based on context. We believe this capability can have a wide range of applications, from speech enhancement and recognition in videos, through video conferencing, to improved hearing aids, especially in situations where there are multiple people speaking.


Yeah whatever, a bunch of mumbo jumbo right? Well just watch this:



François Chollet on Twitter

François Chollet, an AI expert at google on Twitter:

The problem with Facebook is not just the loss of your privacy and the fact that it can be used as a totalitarian panopticon. The more worrying issue, in my opinion, is its use of digital information consumption as a psychological control vector. Time for a thread



Economics of AI

On 13-14 September, 2017, we held our inaugural conference in Toronto to set the research agenda on The Economics of AI.

Paper presentations are good, but holy shit the comments are amazing.

Check the conference site for the papers and slides.



Cognition Builders: Extreme Parenting For The One Percent

Kim Brooks in The Cut:

The company was called Cognition Builders, and Harris explained that they would send people to a family for a period of weeks to observe everyone’s behavior and to figure out how parents could get better control over their kids. The people they sent were called “family architects.” They’d move in with a family for months at a time, immersing themselves in their routines and rituals. The family architects were the foot soldiers in the Cognition Builders team, but the most critical part of the company’s strategy involved the installation of a series of Nest Cams with microphones all around the house, which enabled round-the-clock observation and interaction in real time. At the end of each day, the architects would send the parents extensive emails and texts summarizing what they’d seen, which they’d use to develop a system of rules for the family to implement at home. Over time, the role of the family architects would evolve from observing to enforcing the rules.

One line that stood out to me was a throwaway from the writer about being a combo of life-coaching with Amazon Echo. But, think for a minute what life is like when the AI’s can parent like this. Robo-nannies will change the world.



A Model of Technological Unemployment

There is a bit of a backlash against UBI in the economics community at the moment, which I think is unfortunately due to a lack of awareness of potential changes to the labor market coming. Here is a paper by Daniel Susskind talking about why mainstream economics may have it wrong on the labor issues:

The economic literature that explores the consequences of technological change on the labour market tends to support an optimistic view about the threat of automation. In the more recent ‘task-based’ literature, this optimism has typically relied on the existence of firm limits to the capabilities of new systems and machines. Yet these limits have often turned out to be misplaced. In this paper, rather than try to identify new limits, I build a new model based on a very simple claim – that new technologies will be able to perform more types of tasks in the future. I call this ‘task encroachment’. To explore this process, I use a new distinction between two types of capital – ‘complementing’ capital, or ’c-capital’, and ‘substituting’ capital, or ‘s-capital’. As the quantity and productivity of s-capital increases, it erodes the set of tasks in which labour is complemented by c-capital. In a static version of the model, this process drives down relative wages and the labour share of income. In a dynamic model, as s-capital is accumulated, labour is driven out the economy and wages decline to zero. In the limit, labour is fully immiserated and ‘technological unemployment’ follows.

Personally I see huge changes coming in the near to medium term (5-40 years), such by the end of that about 40% of currently employed population will have no jobs available to them. (If minimum wage laws hold, that would be literately no jobs, if minimum wage laws fall, then it means jobs that pay only sustenance level). That will cause major changes to the economy of course. Goods will be produced at much lower cost, but how will people who don’t have jobs buy them? This is where UBI begins to make sense to me.

Then in medium to long term we have the potential of AI singularity, in which case, who knows. Even if no singularity, the automation encroachment on refuge labor will continue …

Via MR.