For my last two pieces, I talked about how technology is going to make a lot of jobs obsolete in the coming decades. The analysis I presented was largely economic, and while I mentioned some of the major technological advances that will be responsible, I didn’t go into too much detail. I figured that I would complete this series with a look at where artificial intelligence is today. The list of recent public advances in AI is pretty staggering: Google’s self-driving cars have logged over 700,000 miles of automated, accident-free driving on Californian roadways; IBM’s Watson supercomputer has bested the world’s finest <em>Jeopardy! </em>players; and last June, a computer programmed to talk like a 13-year-old Ukrainian immigrant convinced 10 out of 30 judges that it was human. All of these incredible feats are possible because of a field of computer science called machine learning. Computer programs, at their most basic, are a set of instructions that cover all of the scenarios the programmer writes them to. For example, <em>Angry Birds</em> uses a physics engine to control what happens when objects in the game move and collide. The rules that the engine defines will never change, no matter how many times you play a level. If the scenarios you are trying to write rules for get sufficiently complicated, then traditional programming becomes impractical. Google’s cars aren’t programmed for every possible event; instead, they used a set of complex algorithms that learn how to drive. If a Google car encounters a situation it doesn’t know how to handle, it will analyze the data in the context of all of the scenarios it does know, come up with the best solution it can, and add that event to its existing set. Each time it does this, it gets smarter. Similarly, Watson prepared for its<em> Jeopardy!</em> game by reading Wikipedia and teaching itself trivia using complex natural language processing techniques. In addition to self-driving cars and robots that can play <em>Jeopardy!</em>, there are a host of other areas where computers are performing incredible feats. Deep learning, a form of machine learning that attempts to mimic human neural networks, has been used to identify molecules that can be used for new drugs, understand and categorize images without using any metadata and can help pathologists diagnose cancer. Astoundingly, deep learning can do all of these things with minimal human interaction by simply observing large sets of data. The team that identified molecules for drug discovery did so without a single member having a background in chemistry or biology — they were all mathematicians and computer scientists. One of the largest areas of research in AI is developing computers that can understand the semantic meaning of text and then make decisions based on that information. Watson is capable of incredible levels of understanding, but it had to read millions of pages of text in order to get there; in other words, it’s a really fast reader, but it takes very little from each page. It’s worth noting here that on top of winning <em>Jeopardy!</em>, Watson is being used for linguistic and semantic interpretation in a diverse range of fields including the financial sector, health care and cooking. In all cases, it helps experts make better decisions by reading and analyzing huge amounts of data. If you need any more convincing that AI is going to become a seriously big deal over the next few years, you need look no further than the flow of capital into the market. Google has been buying up robotics and AI firms since the start of last year and building a team that consists of “less than 50 per cent but certainly more than 5 per cent” of the world’s leading machine learning experts, according to Peter Norvig, Google’s research director. Silicon Valley investors have poured money into a fresh round of startups focused on building new technologies using AI. Stephen Purpura, founder of big data analyst company Context Relevant, estimates that there are about 170 new ventures attempting to capitalize on the trend. Rapidly increasing investment levels will inevitably lead to faster progress, which will mean we’ll be seeing smarter machines doing more complicated tasks. The better and more reliable these algorithms and systems get, the more places they are going to be used and the bigger the workload they’re going to be able to take on. Self-driving cars may well add truck, bus, and taxi drivers to the list of technologically obsolete professions. Computers like Watson may be primarily augmenting the work of experts right now, but research indicates that when machines like it become more ubiquitous, we’ll see them start to take on job titles like office clerk, telemarketer, and lab tech.