First things first
On a relatively slow news day, here are the three things that really piqued my interest:
- Herd of AI startups is milking the Internet of Cows (Nvidia): Yes, this is definitely a piece of marketing content from Nvidia, but it also demonstrates what I think is the best way to tackle vertical artificial intelligence applications -- locally and by industry experts. It seemed like the first wave of "data science for agriculture" startups, or however you want to define them, had a Silicon Valley bent, but the companies profiled here are from cities (small ones, even) where I have to assume they have a local connection to the dairy industry. I'm generalizing, but the difference in geography and industry knowledge is the difference between searching for a problem that needs your solution, and searching for a solution to your problem.
- What you need to know about California’s new data privacy law (Harvard Business Review): Honestly, this new law got way less attention than I thought it would, and way less attention than the GDPR. Perhaps that's because the California law appears to have lesser penalties (GDPR's max fines are steep) or because California is just one state out of 50. But it's also a huge economy in its own right, and home to a sizable percentage of tech companies that matter. I would also bet that other states, and eventually the federal government, will follow in California's footsteps in the not-too-distant future. So maybe start paying attention now.
- Troubling trends in machine learning scholarship (Arxiv): A really interesting paper that lays out issues not with machine learning and AI, but rather with the academic publishing process. It covers everything from overpacking terminology to misaligned incentives to include more math, and also an area that resonates with me: giving into the urge to please reporters, investors and other external entities. But a lot of this is on the journalists and others, too. They need to do a better job unpacking what terms like "human-level" actually mean, and also distinguishing between results in the lab versus anything currently applicable in production.