Smart machines era will be the most disruptive in IT history”, according to Gartner and McKinsey.
What does “smart” mean in this context? Compared to old generations of machines, new logistic or agricultural bots, Google cars, interactive humanoids, Apple Siri or IBM Watson have in common their capacities to perceive, decide and act autonomously. That is being “smart”, and constitutes what is usually called Artificial Intelligence (AI).
Why not before? Because building smart machines means facing 3 barriers:
- higher failure rate: within the machine design chain, moving prototypes to industrial (pre)series is a challenge. The more hardware, cpu and OS are proprietary, the more difficult they are to integrate and stabilize. Similarly the adjunction of more actuators such as wheels, arms or any other servomotor requires more months or years of testing and debugging,
- business model adaptation: machines that deliver end user services continuously through the cloud, require their suppliers to put in place the necessary infrastructure, technical staff as well as customer support, and find a way to monetize them. Some companies struggle to adapt,
- Artificial intelligence built for real life: too often nowadays, machine perception capacities such as speech recognition, object recognition or obstacle detection are tuned for smartphone contexts, not those of real life where sound echoes or light changes constantly.
The good news is, that significant progress are being made in 2015:
- machine vendors are adopting more and more standard platforms based on Intel or ARM processors, running Linux or Android. This results in higher reliability, lower cost and larger ecosystems of developers able to write software and applications on top of it. The same dynamic applied in the past to PCs (Windows-Intel convergence) then smartphones (iOS-Android convergence) and paved the way to their exponential industry growth,
- vendors foster their effort to educate the market with recurring prices. In the continuation to digital media such as music or games, machines vendors introduce more devices based on monthly plans. For instance, the new Softbank Pepper launched to public last June is sold at a fix $1600 + $120 monthly for cloud services,
- recent developments in machine learning techniques together with the growing quantity of real life training data from consumers transactions helps increase the performance of perception algorithms steadily, such that they become more and more usable in daily use cases.
So we’re living an exciting time! We should see more machines with smart and useful features all around us, very soon.