Everyone agrees that AI has made significant progress during the last few years. And yet, there are some major roadblocks that prevent us from smiling when we're thinking about the future of artificial intelligence.
Take Microsoft's chat bot, for example. Make no mistake: it's a hugely complex piece of software, and yet it was unable to properly filter the information that was fed to it, so it ended up making racist remarks on Twitter. Things got out of control so bad that Microsoft had to shut it down.
The example above highlights the first AI hurdle: poor unsupervised learning. Most AI systems learn best by interacting with people who provide the needed information. That is called "supervised learning", but it isn't feasible when it comes to huge amounts of data.
Supervised learning isn't scalable, but things should change for the better in the near future. As computers become faster, AI systems will learn to process more data in real time, and this will reduce the number of errors.
What about AI creativity? Google is doing something with its AI-based music and art generator, but let's be honest: the end results could be much better. Computers lack creativity, and this may continue to be a problem for many decades.
Finally, most people don't trust AI-based systems. Everyone wants to drive an intelligent car, which is able to avoid collisions, right? But this will only happen until your self-driving car crashes into another one.
However, as artificial intelligence is integrated into more and more practical projects, people will accept it in greater and greater numbers.
IBM predicts that by 2020, 85% of customer interactions will be handled by an AI-based system.
Google acquires DeepMind, a UK-based AI startup, for a reported $525 million.
Init.ai has discontinued its services. The six member team will start working for Apple, helping the company improve Siri.
Mitsubishi Motors Corporation will fit its vehicles with a personal AI assistant, which should make driving safer and easier.
Discover the untapped opportunities of applied AI. Groundbreaking business applications, compelling use cases, rock-solid technical skills, tear-downs of successful AI projects and dissections of failures. The conference will take place at the Hilton Midtown, a spectacular location just steps from Central Park.