Moore’s Canary Reviews of Intel’s latest flagship CPU’s are out and they are decidedly meh. To those who care about such things, the performance of these chips is not a material improvement on the previous generation of chips. Does this mean that the long-predicted demise of Moore’s Law (predicting that processing power would double every two years) is finally here? After all, there’s only so many transistors you can fit on a chip. And, really, who cares? Intel may not be able to convince consumers to replace their computers after five years but the rest of us should be just fine. In fact, individuals and companies can save money by getting another couple years out of their existing technology. Why does this matter? In the third quarter of 2016, GDP grew about 3%. This growth rate was described by many conservative economists as “anemic” and was even cited as one possible reason for Donald Trump’s unexpected victory. But economists who study growth over the long term (like Thomas Piketty) suggest that annual growth of even 1% is historically anomalous and only possible with massive productivity gains. Productivity growth over the last 30 years has been closely tied with processing power. An argument could be made that U.S. growth over that period of time is due in part to advances predicted by Moore’s Law and that, should Moore’s Law cease to be predictive, productivity gains could slow and GDP growth would stall. Moore’s law in general and Intel in particular could be a canary in the coal mine for larger issues in the economy. Next Steps: Hopefully, quantum computing can yield a commercial product in the next two to five years. Read More Cars that learn to drive Many technology companies announced self-driving car initiatives, only to back off when it became clear how complicated driving really was. It turns out that human drivers are making constant split-second decisions and calculations that allow traffic to proceed quickly with remarkably few accidents (given the huge number of potential accidents.) However, a new generation of researchers are creating systems that can learn from experience. The systems are asked to drive in traffic simulations. Depending on how they perform certain tasks (say, merging into traffic), they are encouraged to reevaluate and reprogram their own system. They are rewarded for positive outcomes, while negative outcomes are discarded. Ideally, this will allow self-driving cars to constantly improve once they are on the road and dealing with unpredictable human drivers. Why does this matter? For a time, self-driving cars will necessarily share the road with human drivers. This is the wild card. If the roads were entirely populated by self-driving vehicles, the vehicles would move in concert. Add in a human driver and you add in an agent of chaos. So having self-driving cars that can anticipate human randomness is an important safety precaution. The danger is that the driving skills learned by self-driving cars through interactions with human drivers might only be a local maximum. Meaning that the self-driving cars cannot be purely efficient since they must efficiently deal with human randomness. It’s even possible that the traffic would initially get slightly worse as human drivers adjusted to self-driving cars and then the cars adjusted to those adjustments. Sub-optimal behavior may be encouraged at both ends. Next Steps: I believe self-driving cars will be a commonplace technology within five years. As a frequent rider of New York City taxis, I am fully supportive. Read More New Revenue Models Recently, Ev Williams announced that the blogging platform Medium would be departing from an ad-supported revenue model. This isn’t particularly surprising if you’re familiar with digital advertising and media. There just isn’t the money in digital advertising to support an infinite number of digital content providers. Some people have suggested that Medium should consider looking at the music streaming service Spotify for their new model. Readers would pay a flat rate for access and income would flow to the content creators based on the number of readers they attracted. This micro-payment model for online content has been tried in the past without much success. But venture capitalist and technology gadfly Fred Wilson has suggested another model. In a recent post, he suggested that the blockchain-based social network Steem might be a better model. Steem is both a social network and a cryptocurrency. Members create content on the social network and then, through a system of incentives and credits, they are rewarded by the community with the cryptocurrency. I’ve read their white paper. It’s fascinating, although it’s not at all clear to me if the system could work in practice. Why does this matter? The digital advertising system is filled with fraud and the value it creates for advertisers (even absent the fraud) is in some doubt. This has led many technology companies to start to look for new revenue models, be it subscription-based or through referral agreements. Steem represents an interesting new approach. Social networks succeed only with high levels of user engagement and Steem is attempting to lock in users through revenue sharing. I wish it didn’t give off the faint odor of a pyramid scheme, but you can’t have everything. Regardless, it will be interesting to see how revenue models grow more complex and varied as new technology startups emerge. Next Steps: First, it’s always worthwhile to read what Fred Wilson has to say. Second, it’s worth clicking through to read the Steem white paper. Read More Uber shares data We are told that data is valuable. If it is well-organized and labelled, data can free us from making business decisions by gut instinct. For a long time, people in technology believed that Uber’s traffic data was hugely valuable. It was taken on faith that, while the ride-sharing service would eventually make money on its own, the data that was entering Uber’s servers was the secret treasure trove. After all, Uber had minute by minute knowledge of traffic patterns in every city in the United States. Now, Uber has announced that they will be providing this data free of charge to municipalities and researchers. Why does this matter? Maybe I’m cynical, but I don’t think that a company like Uber has decided to open source valuable data out of pure generosity. Instead, I think that they couldn’t find a way to monetize the data. While the data was well-organized and labelled, there just weren’t customers for it. Existing traffic data published for free by state and municipal governments gave businesses enough information to select retail locations and local city governments don’t have the funds or inclination to invest in traffic data. The lesson seems to be that data has a value – the value people are willing to pay for it. If there are no customers for your data, you’re not selling it no matter how interesting it might be. There’s a temptation in this era of cheap cloud storage and machine learning to imagine that customer data contains hidden value. Try to imagine the use case before you invest in the infrastructure. Next Steps: Don’t assume that your customer data represents a possible revenue stream. Read More