Thud The sound of our world is over-stuffed with clues and cues that hold potential economic value. The squeal of a car’s brake pads hints at the need for replacement or increased liability for an insurance company. The slow, labored chug of an overtaxed computer fan portends lost productivity. John Mannes, in an article on Techcrunch called “The Sound of Impending Failure” talks about the potential and difficulties of applying machine learning to sound. The largest challenge in applying machine learning to sound is that sound data is almost entirely unlabelled. There are vast, labeled databases of image and voice recognition data. Sounds tend to be interpreted through “common knowledge.” For example a malfunctioning automobile is recognizable because the sound it makes is atypical. But computers are notoriously bad at common knowledge. Actionable information for a computer needs to be both detailed and explicit. In addition, sound is difficult to isolate. In sound design, we create a sense of place by overlapping discrete sounds to create an undifferentiated sonic panorama. But going in the opposite direction, from the general to the specific, is difficult. Sensory input is singular. It is our reasoning capacity that suggests which sounds can be traced to which source. Why does this matter? Machine learning depends on labeled data. A computer recognizes the difference between kitty cats and puppy dogs only after it has been supplied with a vast quantity of images accurately labeled “kitty cat” and “puppy dog” on which to learn. Google has underwritten at tremendous cost an effort to create a similar labeled database of human voices, so their machine learning systems can learn to associate the spoken words “kitty cat” with the concept “cat”. A machine learning system confronted with unlabeled data in a new format will be helpless. Sound (beyond human voices) has played almost no role in the digital revolution of the last 30 years. So the comparative amount of data to train machine learning systems doesn’t exist. This becomes an issue for technologies like self-driving vehicles. While a mesh network might create mutual awareness between different vehicles on the road, sound is a much more efficient and universal indication of object, velocity and proximity. In a nutshell: Except for human voices, there is no database of sound appropriate to the training of machine learning. Read More Net Neutrality: not with a bang but a whimper The appointment of Ajit Pai as FCC Chairman seems to indicate that net neutrality is on life-support. Pai has been public and vocal in his opposition to net neutrality – the policy that prevents ISP’s from creating an internet fastlane for certain, favored content. The previous vote in support of net neutrality was a party line vote in which the three Democratic commissioners voted in support and the two Republican commissioners voted against. Now, the balance has shifted and a Republican-led FCC can choose to overturn the regulation or simply choose not to enforce it. The path of least resistance is to just grant exemptions to any ISP that requests them. This prevents a big, public fight with technology companies (since the policy is officially unchanged), but renders the regulation toothless. Why does this matter? The way that the internet is delivered to consumers in the United States has always been a weird compromise. Because the large cable and cell phone providers were the only companies who had the infrastructure in place to provide broadband then these pre-internet behemoths became the default ISP’s. However, they were always a strange fit for the open culture of the internet. They simply couldn’t understand why they weren’t allowed to charge both consumers and content providers, as they do with cable television. Contrast this with Google, which promotes and facilitates all web traffic based on the assumption that they inevitably profit from increased traffic. Now we also have an administration that is also a strange fit for the open culture of the internet. One wishes they would simply over-rule net neutrality and have an open argument. But Pai is probably too smart to blunder into opposition. In a nutshell: Net neutrality will become a toothless and unenforced regulation and consumers will suffer. Read More The best bootstrap Bootstrapping success is a tired cliche of entrepreneurship. The startup, short on funds but long on ambition, hitches its unique advantages to inexorable market conditions and levers up all metrics simultaneously – users beget vendors beget investors beget users beget... well you follow the rest. In my experience, most of these bootstraps are simple cases of information imbalance. Or, if you are of a moralistic disposition, lies of omission. Everyone is subtly misled about the success of the platform so that they will participate. Which is why I found the case of WeChat so fascinating. WeChat had the same challenges many technology startups face. They needed to convince users and vendors to accept their platform for mobile payments. To do this, they tapped into the Chinese cultural practice of giving cash in red envelopes and gave their users a virtual way of doing this. The practice caught on and combined with savvy situational marketing, the service became almost universally accepted. No lies necessary. Why does this matter? The United States is ground zero for the startup economy. That’s a good and a bad thing. It’s good because it encourages people to take risks and create new value. It’s bad because it creates a culture where sloppy or incomplete thinking about go-to-market and monetization strategies are common. Very few American startups have a clear vision for growth beyond “get a lot of buzz” and “turn that buzz into money (somehow).” Meanwhile, Chinese technology companies are criticized as being me-too plays that benefit from government protection. WeChat may have benefited from seeing similar companies experience success in other markets, but it is undeniable that they intelligently built their user base using cultural insights. In a nutshell: Hype is not a strategy. Read More