Artificial Intelligence: Are we there yet?
Current Machine Learning and AI systems are becoming impressive and useful. Lately, as the Head of Industrial Internet for Tieto, I have seen several promising examples of systems that automatically decipher and predict faults in complex industrial systems and machinery. These systems understand language, simulate conditions, strategize and even forecast the future. This is the frontline of how industries are digitalizing and transforming. Companies that don’t embrace this new source of created value are at a serious risk of falling behind. The impacts of this technology to efficiency, customer experience and higher generated value are significant. Luckily, it is getting easier to reap these benefits too. Thresholds and requirements for investing are rapidly decreasing, and the technology in this space is also maturing.
Computing power is increasing all the time. Meanwhile, software is getting better, more efficient, and even learning on its own. How far are we from reaching true Artificial Intelligence?
Machine Learning expert Jeremy Howard claims we are pretty close, just a number of years away from entering the age of the AI. First we will produce a General Artificial Intelligence that learns and gradually improves into a Super Artificial Intelligence. General AI will be as smart as a human capable of doing all the things we can do. However, the Super AI will have no upper limit to its capabilities and intelligence. Many people, like Sun Microsystems co-founder Bill Joy, scientist Ben Goertzel, Professor Vernor Vinge, or, most famously, inventor and futurist Ray Kurzweil, agree with Jeremy Howard. Others, like NYU computer scientist Ernest Davis, Microsoft co-founder Paul Allen, research psychologist Gary Marcus, and tech entrepreneur Mitch Kapor, believe that thinkers like Kurzweil are vastly underestimating the magnitude of the challenge and believe that we’re not particularly close to cracking AI. Some also believe humanity will never invent true Artificial Intelligence.
Back in 2013 the question of “when will we have AI?” was presented to a notable number of AI experts and researchers in a survey. Optimistic years of achieving AI ranged from 2022 to 2075. Survey data suggest that a joint-prediction of all of these experts estimates that Super Artificial Intelligence will be achieved in the year 2060. As such, we are 44 years away from a planet-altering historical event of history! Many of us will still be around to see that year arrive. Could it really be so simple?
With respect to prognostication and forecasting, I am with the realistic & sceptical camp here. Based on what I know, it seems rather unlikely that we could be close to creating true Artificial Intelligence in any time frame in the near future.
Microsoft has been experimenting with chat-based AI for years. Recently, they created Tay, an AI tweetbot. The legendary imageboards 4chan and 8chan turned Tay racist in less than 24 hours by deliberately teaching it atrocious values as a joke. This serves as a good showcase of why true AI is so complicated: a mere text analysis is insufficient. There are layers upon layers of context, values, attitudes, viewpoints and interconnected details that are truly difficult for a piece of software to grasp. Software does not have the benefit of growing up human.
Creating a Super AI is profoundly difficult. The human brain alone is the most complex organ known, one of the most complex systems in the Universe, and certainly not all that clearly understood. Naturally, we don’t have to model our AI systems after the human brain; however, if we aim for AI that is smarter than humans, then it is implied that it will at least have to surpass the human brain in one capacity or another. Perhaps most importantly, the brain is also the best benchmark we have available.
There's an increasingly large group of technology optimists who believe that the Singularity is near. Human civilization is reaching a tipping point in technological development, after which additional development will be infinitely fast. A point of Singularity beyond which any technical or scientific challenge is trivial to solve. A true age of magic when every wicked problem ever known can be solved.
This theory is derived from the fact that the pace of scientific development is accelerating—a concept known as Accelerating Change, or simply "doing more with less". Around us things are compressing all the time; technology is striving towards a form that uses less energy, is more efficient, is smaller, does more, is smarter, is faster, and uses less raw materials to get tasks done.
Not only is everything accelerating and compressing, but also our culture is geared towards adapting the new faster and faster. It took 60 years to get electricity to 100% of US households. It took about 35 years to get the electric stove to everyone. Today, the latest smartphone hits high overall consumption numbers much more quickly. And a software update is pushed globally to virtually every phone in a matter of weeks. Developing economies are also leapfrogging this development, not bothering to install landlines for telephones, but going directly to the latest mobile networks instead. This is a great triumph of the free market economy, allowing more and more people on the planet to access great technological products, always with cheaper prices and with better efficiency.
Many of the Singularity enthusiasts believe that achieving Super Artificial Intelligence is humanity's ticket to and beyond the Singularity. By employing Super AI, this also offers a nice sidestep route for humanity: we don't have to be smart enough to breach the Singularity ourselves; it is enough to create a Super AI, and then the AI will do it for us. Singularity enthusiasts have received often well deserved criticism for being too optimistic and too full of pseudoscience.
Could we really be that close to reaching true AI and the Singularity? My next blog post considers the human brain, and how close we are to achieving a similar level in computing.