Facebook is betting big on chatbots. Should you do it too? Are chatbots the future? The discussion is currently raging, and here at Team Up Start Up we decided to summarize the current debate for you.
AI is the future
The premise of the debate is that artificial intelligence (AI) is the future of computing. There is little controversy around this fact, and all biggest players are currently investing heavily in this space.
How will AI be delivered?
What is not clear is how AI will be delivered to consumers. For example, some of the earliest application of AI are smart scheduling assistants. In this case, AI resides inside your inbox – in the sense that you interact with it via emails. Another likely way we will interact with AI-powered services is through our phone. In this respect, google now already scans your inbox/calendar/feed/searches and suggests itineraries, interesting stories, and so on before you ask. Chats/messages are another potential delivery channel.
How to sound smart in your TED talkRead More
"I have devoted my career to understanding exactly how practice works to create new and expanded capabilities, with a particular focus on those people who have used practice to become among the best in the world at what they do. And after several decades of studying these best of the best—these “expert performers,” to use the technical term—I have found that no matter what field you study, music or sports or chess or something else, the most effective types of practice all follow the same set of general principles.
[people] assume that someone who has been driving for 20 years must be a better driver than someone who has been driving for five, that a doctor who has been practicing medicine for 20 years must be a better doctor than one who has been practicing for five, that a teacher who has been teaching for 20 years must be better than one who has been teaching for five.
But no. Research has shown that, generally speaking, once a person reaches that level of “acceptable” performance and automaticity, the additional years of “practice” don’t lead to improvement. If anything, the doctor or the teacher or the driver who’s been at it for 20 years is likely to be a bit worse than the one who’s been doing it for only five, and the reason is that these automated abilities gradually deteriorate in the absence of deliberate efforts to improve."
... and play Go with it.
...AI is actually in the process of leaving the lab. Google open sourced its AI framework called tensorflow on November 2015. The ecosystem around tensorflow has since grown to the point where easy, step-by-step tutorial on installing and running your own AI model are now available (see, for example, here).
These tutorials are not exactly for everybody. A minimum knowledge of Python and the basic principles of machine learning are required (which, by the way, can also be easily acquired by reading few tutorials or following an online class). Still, it is a huge step in democratizing access to AI.
What will bright entrepreneurs do with it? We already saw few virtual assistants/schedulers. But how about a “candy crush” crushing algorithm? A challenger for “alpha-go”? Or, finally, a towel-folding robot?
One of the most interesting articles I read recently: all you wanted to know about the placebo effect. What I learned:
- The placebo effect works in reverse: if you think you are receiving a poison (when in fact you are not), you may actually die.
- The placebo effect works also in animals.
- The immunitary system associates harmless substances (i.e., sugar pills) with either drugs - in which case it triggers an immunitary response - or poisons - in which case it suppresses its normal immunitary response.
Here is the passage I found most fascinating:
In 1975, a psychologist in New York was studying taste aversion in a group of rats and got an utterly mystifying result.
Robert Ader, working at the University of Rochester, gave his animals saccharin solution to drink. Rats usually love the sweet taste but for this experiment, Ader paired the drink with injections of Cytoxan, which made them feel sick. When he later gave the animals the sweetened water on its own they refused to drink it, just as he expected. So to find out how long the learned aversion would last, he force-fed this harmless drink to them using an eyedropper. But the rats didn’t forget. Instead, one by one, they died.
Though Cytoxan is toxic, Ader’s rats hadn’t received anything close to a fatal dose. Instead, after a series of other experiments, Ader concluded that when the animals received saccharin and the drug together, they hadn’t just associated the sweet taste with feeling sick, they’d also learned the immunosuppression. Eventually, they’d responded to the sweetened water just as they had to the drug.
[...] a similar discovery had already been made in Russia. In the 1920s, researchers at the University of St Petersburg were following up on Pavlov’s work, to see which other physiological responses could be conditioned.
Among them was the immunologist Sergey Metalnikov. Instead of suppressing the immune system, like Ader would, Metalnikov wanted to boost it. In one series of experiments, he repeatedly warmed guinea pigs’ skin at the same time as giving them injections (small doses of bacteria, for example) that triggered an immune response. Then he gave them – and another group of guinea pigs that hadn’t had this conditioning – a normally lethal dose of Vibrio cholerae bacteria, at the same time as warming their skin. The unconditioned animals died within 8 hours, Metalnikov reported, whereas the conditioned ones survived an average of 36 hours, and some of them recovered completely. Their response to a learned cue – the feeling of heat – appeared to have saved their lives.
Cryptography is not exactly my thing. But I really enjoyed this article. Basically, the NSA spent the last 20 years pushing for the adoption of a set of cryptographic standards called Suite B "the first public cryptography standard to include non-classified algorithms certified for encrypting Secret and Top Secret data."
Then, all of a sudden, in August the agency freaked out, and "updated the Suite B website to announce a rapid transition away from Suite B, and to a new set of quantum quantum-resistant algorithms in the coming years."
However, experts agree that there has not been any major breakthrough in quantum computing, and hence the official justification is not believable. So why such a sudden transition away from a standard that they pushed so hard for? Is it possible that this standard had some weaknesses--by design--that now are known not only to the NSA but also to someone else?
"His job is to slice into the brain, the most complex structure we know of in the universe, where everything that makes us human is contained, and the contrast between the extremely sophisticated and the extremely primitive — all of that work with knives, drills and saws — fascinated me deeply. "
Beautiful pictures too. By the way, my own thoughts on Henry Marsh's book "do no harm"