Innovation Shifts East A few years ago, IBM owned artificial intelligence. Using neural networks and machine learning, they produced Deep Blue, a system capable of crushing both chess masters and Jeopardy with aplomb. IBM learned from their brief flirtation with producing commodity servers and PC’s to stick in the high end of the business market, providing bespoke technology and consulting to large businesses. Watson represented the next generation of these services. But a perception began to bubble up in technology circles that they were over-hyping Watson and falling behind their competitors on the west coast in terms of technological innovation. At the same time, IBM has jumped ahead of the pack in quantum computing – technology which uses the multiple positional states of quantum particles to move past the old binary model of classic computing. Now IBM has opened a $240 million research center at MIT to research the combination of AI, quantum computing and innovative chip design. I have been hearing rumors going back years now that IBM was sitting on some extremely innovative new chip designs and I have enjoyed playing with their IBM Quantum Experience. It is heartening to see them combine these promising areas of research at an elite technical university outside of Silicon Valley. Why does this matter? The limitation of machine learning (also known as AI) is that it requires large, well-labeled data sets in order to work at its best. If you wanted to design a system to recognize cats in photographs (because cats!), you would first have to train that system on a vast database of images labeled “cat” or “not cat.” Machine learning doesn’t actually learn to recognize cats, it learns to associate the label “cat” with certain patterns in images. But most data in the world isn’t labeled. So the new holy grail of machine learning is getting it to recognize patterns in unlabeled data. To do that properly, you would need a system that was capable of learning indiscriminately; not just “cat” vs. “not cat” but cat, dog, fence, sunset, window, orange, etc. Machine learning is theoretically capable of learning in this way, but it would require a massive jump in computing power. Quantum computing has always been an intriguing combination with machine learning because it can provide that jump in power. In addition, there is a theory that the human brain actually processes data using quantum states. After all, people don’t evaluate objects in the world according to binary categories of “cat” and “not cat.” Even people who really, really like cats. In a nutshell: IBM and MIT are doing something cool together using quantum computing, innovative chip design and AI. Read More Privacy or Competition? Apple's new operating system for Mac and iPhone is rolling out soon and the advertising industry is having a fit about one of the new features in Safari. Apple is establishing a 24 hour limit on ad targeting cookies, insisting that the move is important in protecting consumer privacy. This is hardly surprising. Consumers are exhausted by the onslaught of online advertising. Ad blocking and cookie blocking technologies are become popular with ever-expanding numbers of users. Mozilla and Google’s Chrome browser have integrated enhanced blocking capabilities in order to protect their market share by keeping pace with consumer desires. But a consortium of advertising industry groups have spoken out against Apple’s changes since they will do serious damage to the digital ad ecosystem by making it difficult to retarget messages to consumers who have shown interest in products. They rightly point out that, although consumers don’t like ads, they very much enjoy all the free content that advertising makes possible. Take away the advertising and paywalls will be the inevitable result. Apple is holding firm for now, but I have to wonder about their sudden, strident commitment to consumer privacy. Why does this matter? Apple’s largest competitors need cookies to work much more than they do. Google and Amazon especially rely on cookies to power their business models. Google makes their money on the adwords/analytics mega-platform that powers much of digital media. Amazon attracts buyers through retargeting based on previous searches. Apple wouldn’t be particularly sad to see Google and Amazon lose a lot of money because cookies stopped working. Apple’s walled gardens, like Facebook’s, don’t depend on guessing at the behavior of web users. Apple has never particularly cared about consumer privacy. In fact, they insist upon tracking every action within their own platforms through your Apple ID. But they’re happy to rest on the moral high ground if it damages the bottom line of their competitors. In a nutshell: Big tech continues to do things in big tech’s interest, not yours. Read More On Exactitude in Mapping In the classic Borges short (short) story, the production of a 1:1 map is meant to illustrate the absurdity of trying to achieve perfect knowledge through representation. But the latest a16z Podcast reminds us that yesterday’s absurdity is today’s killer business plan. Map collector David Rumsey and Wei Luo, the COO of DeepMap – who build maps for autonomous vehicles – talk about how new mapping technology and new uses for maps are transforming our understanding of what maps need to be and what scale of accuracy is acceptable. Currently, maps aren’t being constantly updated. While data is collected by GPS and driving technologies, the maps themselves on which those systems run were created by a survey fleet prior to launch. Although drivers may pick up inaccuracies in those maps, nothing can be changed until the survey updates their map. However, autonomous vehicles introduce new demands and possibilities for digital maps. First, autonomous vehicles require accuracy levels down to the centimeter. Many of the mapping conventions that exist assume a human user. Humans are capable of reasoning, robots are not. Robots require that maps define things like lanes on the road, curb distance and height, and speed limits. Maps for robotic users also don’t visually resemble the maps we know. A map, after all is just data on the structure of physical space. Visualization of maps is a convenience for human users. But a robot reads the data as raw numbers. The deep maps produced for autonomous vehicles would look like a very detailed spreadsheet to you or me. The benefit of autonomous vehicles for mapping is that they are also continually collecting physical data on their environment via cameras and lidar. That means that the deep maps are built to be dynamic – constantly updating based on real time information from the field. This can generate maps of remarkable and timely accuracy; maps that are accurate across four dimensions. The more autonomous vehicles that get on the road, the more accurate maps will become. 1:1 scale is just the beginning. Why does this matter? On the one hand, an infinitely perfectible map is an intriguing notion. But I do think we should pause to consider the creeping digitization of human capabilities. It does not take much to imagine a future in which autonomous vehicles handle all transportation and people are left with little spatial sense for where they are at any given time. If this sounds melodramatic, consider the other capabilities we have handed over to technology. Our knowledge for time, weather and phone numbers are fully dependent on digital assistance. Our memories for facts and events are increasingly outsourced. If we sacrifice our spatial sense, what is next? Certainly, recognizing faces and associating those faces with names would be an appealing bit of functionality I personally would love to access. What about correspondence? While I like to believe that each of my emails to clients is a precious gem of unique wisdom, a machine learning algorithm would probably be able to predict 90% of my responses accurately. Digital technology is dangerous to us precisely because it is so darn convenient. Someday soon your car will know your neighborhood much better than you will. That sounds fine, until you get lost at the end of your driveway. In a nutshell: Deep maps provide unprecedented depth and can dynamically update themselves. Read More