Evolving from Narrow Artificial Intelligence to Cognitive Artificial Intelligence

Much has been written about the Rise of AI and it is more to read than you have human time to do it. I have personally written about AI First and the next stage of machine intelligence back in 2015. Today I commit some words about where Artificial Intelligence is today and where it is heading to in the next years. At the end I have a wish list for entrepreneurs, companies and researchers.

Today we are already living in an AI world

Big Data, machine learning, hardware and the Internet have enabled us to create hundred thousands of different Artificial Intelligence applications.

We use AI for automation of processes. We let it trade stocks, drive cars, trucks, control trains, drones and submarines. We let it analyse traffic videos. AI sorts for us our Facebook, Spotify and Netflix content.

We use AI for human machine communication. Either with Google Speech or we communicate directly with Alexa, Cortana and Siri. We even let AI read and write our emails.

Artificial Intelligence is the digitalization of human knowledge work. We trust AI for navigation and research. We even let AI win Go, poker, compose music and create art.

And that all is just the beginning. It is like DOS for the PC era and a Nokia 3210 for the mobile era.

Today we have Narrow Artificial Intelligence

I call these current systems Narrow Artificial Intelligence. They are highly specialized and are often better than humans at the same task. However they need large sets of data and weeks/months/years of training by humans. Additional most models are still highly mathematical (e.g. try to improve gradient descent) and brute forcing trial & error (see error back propagation).

On top the biggest drawback is, that these Narrow Artificial Intelligences can not transfer knowledge.

The field of AI is as old as the field of computers. We started with heuristic systems, then we had expert systems and today we have learning systems. However that is not enough and we need to reach the next stage of machine intelligence.

We need AIs, which can adapt to new situations. We need AIs which can build a memory, gain knowledge and uses its own experiences. We need AIs, which learn faster with less data.

What we need is Cognititve Artificial Intelligence

I call the next generation Cognitive Systems, or Cognitive Artificial Intelligence. These are a combination of machine learning + knowledge technologies. Prof. Hans Uszkoreit briefly touches that topic at the end of his Rise of AI talk.

A Cognitive Artificial Intelligence has perception (sensory input like hearing, listening, reading, seeing, feeling). It can execute action (interpretation, reasoning, planning and communication) and it can react to unplanned situations.

For Cognitive Artificial Intelligence Systems we need learning algorithms, which are faster and need less data (better abstraction). We also need to better understand how our own brain works. Humans are still the only Superintelligence in the room.

My AI wish list

I kindly ask researchers to work more on transfer knowledge. Google’s PathNet leads the direction. Please work more on AI’s memory and experiences. Find ways that AIs can react to unplanned situations without brute forcing their way to optimization.

For entrepreneurs and startups out there: Work on models and the data. You need to have access to plenty of data or own it. Close the feedback look in your AI system. Means, your system should gather, learn and optimize without human help.

Algorithms and frameworks are free. Concentrate on the models, which are compressing your data. Dr. Damian Borth speaks about it at his Rise of AI talk.

For the more mature companies: think AI first. Build internal AI teams, buy AI companies or partner with AI startups.

If you have a service, then sell the solution. Fix your problems and your customer’s problems with AI.

If you sell hardware, then use your hardware as a Trojan Horse for your (software) intelligence.

This is my AI wish list for 2017.