As fun as it is to fantasize about doom scenarios depicted in almost every Hollywood production about robots – that artificial intelligence will become sentient and turn against humanity – we can’t forget that we still have a long way to go.
Where we are now
The three basic areas of use of AI right now are engagement, insight, and automation, meaning that AI is primarily used to optimize processes, analyze data more efficiently, and define what new customer experiences could look like.
However, there are also other application areas. For example, drones use artificial intelligence to analyze their surroundings. Now, they can process images in a way like never before, making it possible for small models to land on an extended hand.
Other examples would be protection from fraud in the insurance and financial industry using AI technology, chatbots and voice recognition in smart phones such as Siri.
Who is pushing forward
Hyperscale cloud providers like Amazon or Google are one of the driving forces behind the rapid advancement of AI.
They have their own projects regarding artificial intelligence, like Alexa, but they also use it to improve their websites and platforms. For example, Youtube is utilizing an AI algorithm to filter out videos with copyrighted material in it.
Startups are also very important for the advancement of artificial intelligence. At a recent Gartner event in Munich, Germany, Alan Priestley, Senior Director and Analyst at Gartner, gave attendants a brief list of promising AI startups:
- Horizon Robotics. Provider of integrated and open embedded artificial intelligence solutions. One of their pronounced goals is to give cars ‘brains‘ to make them smart vehicles.
- Cambricon. Pioneer of deep learning processors. The company is a financial unicorn, meaning that its value is over one billion dollars.
- Esperanto Technologies. Developer of energy-efficient computing solutions based on the open standard RISC-V.
Most AI technology is in its infancy, one or two innovations have hit puberty, but the full potential of it has not been realized yet.
There is a number of reasons for it, one being that AI can’t really reside in ordinary data centers. It has to move to the edge in order to process real world data.
“Right now, AI systems are an approximation“, Alan Priestley explained. “If you give an AI system a hundred thousand pictures of cats, and then show it a different picture of a cat, it will say, ‘I’m 95 percent sure that is a cat‘.“
This becomes a problem if you think about new innovations that use AI, like autonomous vehicles. If an autonomous vehicle is not 100 percent sure that what it is about to inevitably run over (for example, because swaying would be too dangerous) is a cat and not a child, how could you entrust it with these decisions?
Furthermore, the driving decision has to be made at the point where the data is captured, meaning at the endpoint of a network, and not, for example, in the cloud. This still needs a lot of computing power at the moment. Not very practical, considering most autonomous cars will primarily be electric vehicles.
It is fair to say that the future will hold a lot of changes in store for AI technology. One major change, however, will be the processors we use for it.
“Right now, we use mostly GPUs – you know, those things we used to play games on“, said Alan Priestley. “GPUs are intrinsic to AI development now. However, this is going to change over the next couple of years.“
Alan Priestley predicts that the next stage of processors will be ASICs, and the third – and most probable last – stage will be neuromorphic computing.
Neuromorphic computing utilizes biological neural networks as digital or analogue copies on electronic circuits for more energy efficiency, speed, and robustness. This means that it could make a whole different generation of artificial intelligence possible.
Priestley concludes, “All depends on the right software. Software is key to advance AI even further in the future.“