IP Land Dyson has announced that they are opening a research and development office in Singapore focused on AI and robotics. The easy explanation for this move is that Singapore has a well-educated workforce and aggressively targets foreign firms to open offices in Singapore. (Full disclosure: I did some consulting work for Economic Development Board of Singapore.) While that explanation is frequently given for firms opening an Asian office in Singapore, it is incomplete. It begs the question, why not locate your business in China since it’s the largest market in Asia? But western firms remain distrustful of intellectual property protections in China. There is a sense, however unfair, that Chinese companies have stolen intellectual property from western companies in order to advance their economy with the tacit approval or perhaps even the active participation of government authorities. Whatever the qualities of its workforce, Singapore is a tiny country that aggressively protects intellectual property. Why does this matter? It’s impossible not to look at the vast Chinese market and fantasize about breaking in. But China continues to be fraught with difficulty, not least because the intellectual property of foreign companies seems to enjoy few protections. Gradually, as Chinese tech companies have moved from being imitators to innovators, they have begun to get religion about the importance of protecting IP. This is not unusual for a growing economy. Once upon a time, the United States was infamous for stealing the designs of French fashion houses. When you’re looking for a leg up, you take any advantage you can. But China is still perceived as a place where IP is appropriated. The maturity of Chinese technology can be indirectly measured by their commitment to protecting intellectual property rights. In a nutshell: Technology companies remain wary of Chinese IP protections. Read More These gears are made for walking Agility, a company spun out of the robotics lab at Oregon State University recently unveiled Cassie, a bipedal robot with an unusual ability to maintain balance. Bipedal design has always been challenging for robotics. The same lab, built Atrias, a prototype bipedal robot that famously survived a barrage of dodgeballs. Cassie is a material improvement over Atrias both in ability and appearance. The new robot is smaller, more agile, and less unwieldy. While Cassie was built using grant money from the Department of Defence, Cassie’s builders are careful to emphasize her (its?) humanitarian applications, for example as an adjunct to fire fighters. Why does this matter? We are not very far away from sharing the skies with programmatically controlled drones, our streets with self-driving vehicles and our sidewalks, offices and homes with robots like Cassie. While there are efficiencies to be gained and lives to be saved by replacing humans with robots in some situations, it’s an appropriate time to ask ourselves what the social implications will be. We’re talking about a radical transformation of shared spaces. If you live in a city, almost all movement is human controlled. But in a few years, you could be standing on a crowded street corner and be the only human being in a sea of automation. In a nutshell: We’re very close to having robotics fully integrated into the economy. Read More Robot overlords or robot partners Critics of artificial intelligence suggest that a super-intelligent computer is unlikely to subsume its will to less intelligent humans. This is flawed thinking. It implies that intelligence and will are associated. My cat is willful, yet he is remarkably stupid. A better argument against AI is that an independent AI, set to a task (like making shoelaces) might blindly subsume all other considerations to that task. The nightmare scenario is an AI killing people and turning them into shoelaces in slavish pursuit of its prime directive. Google’s AI division DeepMind has been studying the circumstances under which AI will be cooperative or competitive using video games. They have found that, if the game rewards cooperation intelligent AI's will pursue cooperation. But if a game encourages competition, intelligent AI's will behave aggressively. Why does this matter? Most AI engines these days have been given a very narrow scope. They are asked to recognize images or translate text or discern patterns in data. There is little danger to people from such AI’s. The task is discrete. But a broad scope AI, given a task and wider latitude for solving that task, might behave aggressively without limits on acceptable behavior. Just as a kindergarten classroom is set up for collaboration and not Darwinian competition, we need to set up the tasks and parameters for AI engines in ways that encourage cooperative action. In a nutshell: Don’t teach an AI game theory. Read More Learn with less One of the biggest challenges of machine learning is that training one of these systems takes thousands or even millions of well-labeled data inputs. For a computer to recognize a cat, it needs to look at a lot of different cat pictures. But many areas where machine learning might offer the greatest benefit (like medicine) lack these huge data sets for training purposes. But a Boston-based startup called Gamalon has developed a technology that can help machine learning learn with far fewer inputs using Bayesian probability. While this technology has been proven to greatly reduce the number of necessary inputs, the inputs that are shared with the system need to be significant. Meaning that they must be clear examples of a certain state and be well-labeled. The system learns faster, but it cannot learn anything from ambiguous data, whereas classic machine learning can. Why does this matter? Machine learning or AI initiatives are becoming common in the marketing world. Marketers are quick to adopt new technologies, particularly when they seem to offer solutions to intractable problems. But most marketers lack the quantity or the quality of data that would allow them to actually apply machine learning productively. New technologies like Gamalon could eventually offer marketers with a limited data set the possibility of using machine learning. However, this would depend on a detailed hypothesis about the larger, unknown data set since Gamalon depends on significant inputs. We cannot forget that determining which data points are significant will substantially alter our results. In a nutshell: New technologies begin to open up machine learning to smaller data sets. Read More