In this post LACE workplace learning technologies expert and start-upper, Fabrizio Cardinali, takes a look at the revamped interest on Artificial Intelligence, and how new trends such as Machine Learning, Robotic Process Automation and Cognitive Technologies are emerginging which might bring together a greater world to work and learn within, but perhaps with far less (human) workers than smart machines around. A great Disneyland for all of us…but perhaps a Disneyland without Children left to amuse.
On May 5th 2016, TIME magazine summarized the views of the world’s smartest people about the revamped interest in Artificial Intelligence, AI, a controversial topic which for a long time has had the fervent support of science fiction fans but which has never made it far beyond pilots in the real world.
Amongst the supporters were Stephen Hawking, the famed 74-year-old theoretical physicist, Elon Musk, the outspoken 44-year-old entrepreneur and CEO of Tesla Motors, and Bill Gates, the 60-year-old computer software magnate and Microsoft co-founder, all of which praised the coming of computational intelligence to the cloud and at the same time cautioned us that it could potentially be “the last [event in our history], unless we learn how to avoid the risks.”
But it was Nick Bostrom, the 42-year-old director of Oxford’s Future of Humanity Institute, who took a dimmer view of the trend, warning that AI could quickly turn dark and dispose of humans, generating “economic miracles and technological awesomeness” within places of work, study and leisure, but with nobody there to benefit at the end,” like “a Disneyland without children.”
Whatever the outlook, if there is one prediction that we might have overlooked in 2014 at the start of LACE, the European Commission’s initiative to support the study and uptake of learning analytics, it was to add “machines” to the potential beneficiaries of the advancements we were about to promote.
During our first public policy brainstorming event at the EU parliament workshop in April 2015, we were very keen to set time on the agenda to discuss the impact of learning analytics on everyone from children to academics to adults to workforces…but machines were never mentioned…they simply fell off our radar.
Today, we have evidence that a new wave of robotic automation is reaching the workplace. And this is taking place above all through the combined forces of analytics and (machine) learning, leaving us to seriously consider if we forgot to account for something in our list of expected learning analytics outcomes.
New computers now exist that are capable of completing working tasks whilst learning symbiotically from humans, achieving their goalsat a fraction of the time and computing costs required from older robots. Baxter, the latest build of the iRobot home cleaner from inventor Rodney Brook, and Yumi, the latest product released by ABB, the global leader in power and automation technologies, are both commoditized robotic workers you can “employ” for less than half of the cost of a junior FTE. Also, they can work 24/7, don’t need vacation time and, a point worth mentioning to some, don’t fall under any trade union protection laws (yet).
Even beyond these, today we have smart software solutions capable of both learning the repetitive actions of humans and executing them robotically. This trend, called Robotic Process Automation (RPA) or softBOTs, demonstrates that in many applications, digital agents and assistants can not only do the work of humans, but do it faster, better and cheaper.
The vast majority of the 1,896 experts who responded to a study by the Pew Research Center believe that robots and digital agents, which cost approximately one-third of the price of an offshore full-time employee, will displace significant numbers of human workers in the near future, potentially affecting more than 100 million skilled workers by 2025.
Still, someone might argue that we did not fall short in our LACE forecasts, since all this has very little to do with learning analytics, if nothing at all…
Well, I couldn’t disagree more…
I first degreed in Bio-engineering and Artificial Intelligence in 1988 at the Engineering Faculty of Genoa, Italy, one of Europe’s first universities to offer an official qualification in such a new and emerging science. At the time, AI was all about computational logic, with costly Prolog programmers having to elicit reasoning intelligence capabilities using even more costly Lisp machines, all to emulate just a small fraction of the reasoning capability of a human being.
It took me a full thesis dissertation (more or less the equivalent of a year of work) to teach my shared Lisp machine (accessed on a very restricted weekly schedule amongst students wanting to experience the strting flame of computational logics…) how to detect malignant tumours in digital images processed on a deicated Eidobrain MicroVAX system.
Still, that movement was enough to start a new wave in pursuit of “Holy Grail” expert systems, which we then predicted would be able to solve any human-based decision-making task, let alone detect suspicious images in digital tomography and X-ray images. Despite the initial high momentum, this intelligence never made it to massive adoption. The objectives were too ambitious. The computational power was too limited. The costs were too high.
Today, big data feeds and machine learning technologies are democratically available, open source and on the cloud. The computation power of the cheapest smartphone on the market today exceeds that of my large MicroVAX station hundred of thousands times over, not to mention comes dressed up with the sexy voice of Siri, Cortana or Alexa, your personal assistants of choice for the day.
But above all, today’s technologies can easily and cheaply reach and engage thousands of smart start-uppers and entrepreneurs who are looking for the next big thing to get behind following the dot-com and app vein burn-out.
So this time around, my gut feeling is that machine learning will happen at a faster pace than ever before, boosted by the power of the cloud and funded by the fuel of NASDAQ. Cheap and cloud based A.I. will be “the next big thing” in computer science, let alone in learning technologies, we are all looking for.
Faced with such an unprecedented level of acceleration, we must be sure to keep and cultivate human workforces who are still capable of learning better and faster than machines. We must keep our brains free to study further, ensuring that we continue to grow and hone our knowledge and mental capabilities to a level unattainable by even the most advanced machine. This is the only means by which we will be able to preserve the competitive edge that humans have in learning, and in (re)defining our learning objectives.
But be aware: IQ marks will not be the most important KPI in the human vs. machine learning competition. As Charles Darwin observed a long time ago, “It is not the strongest of the species that will survive, nor the most intelligent. It is the one that is most adaptable to change.”
KPMG has classified the self-learning evolution of software into 3 classes (See Figure 1).
Fig 1. RPA (Robotic Process Automation) Tools Classification by KPMG – “Bots in the back office. The coming wave of digital workers” (KPMG, 2016)
The reasoning capabilities of computers and robots are going to adapt very fast. So we must adapt faster.
Class 1 (Robotic Process Automation) exists today, and is starting to be embedded into platforms usually dealing with (human) workforce support and improvement.
There is increasing evidence that learning solutions originally meant to empower (human) workforces with the support of machines will gradually integrate machine learning capabilities to teach themselves how to robotize repetitive tasks and substitute human workforces and possibly reduced the time lost on tedious tasks in favour of better talent allocation.
For example, the Skillaware™platform , a workforce performance support solution I founded a couple of years ago and originally meant to train (human) workforces during the adoption of new software platforms and processes, not only inclides automatic (i.e. robotic) generation of task guidance agents and training documentation, but in its just released version 3 has recently added an advanced rule-based reasoning engine to its capability. Skillaware is now capable to possibly automate routine tasks normally performed by operators in call centers, ticketing services and/or CRM-based salesforces, possibly improving process outcome with reduced (human) FTE allocation. This addon has recently granted Skillaware a listing by Gartner, the leading IT analyst, as one of the world 4 cool vendors in 2016 for CRM systems training and support (http://blogs.gartner.com/tad-travis/2016/04/08/gartner-announces-the-2016-cool-vendors-in-crm-for-sales/)
This is a wave many technologies are starting to ride with newly emerging tools fully dedicated to workforce substitituion rather than training (e.g. BluePrysm™, AutomationAnywhere, etc.) . All use similar first generation technologies to that adopted by Skillaware (which indeed was made to support human workforces than to make them obsolete due to machine automation) but push the robotic level even further.
These tools offer a basic learning intelligence based on screen scraping (i.e. a technology coming from testing and disability support, enabling software to capture data entry from humans interacting on computer screens) combined with rule-based reasoning and workflow automation. New solutions are experimenting with accessing big data series with cognitive technologies, such as Microsoft’s Cortana™ and IBM’s Watson™, which are capable of processing unstructured data and taking intelligent decisions on behalf of workers based on that knowledge. This represents a steady step into Class 2 (Cognitive Technologies) tools and architectures.
Finally, Class 3 (Cognitive Automation), the real advent of machine intelligence, will be based on accessing a diverse set of cognitive automation technologies and algos, accessible on a cloud-distributed and multi-service level, combining a variety of technologies such as language parsing, big data analytics and adaptive alteration.
These plug-ins will significantly the learning technologies of machines, perhaps towards the singular breaking point of creating a machine capable of learning alone or even teaching other computers to do the same work at a shared workplace…something humans have never been really good at.
Today, we are still at the beginning of Class 2, and a stable, low-cost group of Class 3 technologies is most likely to be available only far beyond 5 years from now.
Figure 2. Mori, M (2012) . “The uncanny valley “. IEEE Robotics & Automation Magazine.
In essence, as robotics professor Masahiro Mori put it in 1970, we are still at the beginning of our walk towards the uncanny valley or “Bukimi no Tani Genshō (不気味の谷現象).” (see Figure 2)
Mori’s original hypothesis stated that as the appearance of a robot is made more human, some observers’ emotional response to the robot will become increasingly positive and empathic, until a point is reached beyond which the response quickly becomes that of strong revulsion, creating a downturn curve similar to that of a deep valley. However, as the robot’s appearance continues to become less distinguishable from that of a human being, the emotional response becomes positive once again and approaches human-to-human empathy levels (See Figure 2).
So today, at the end of my 2.5 years engagement in promoting learning analytics at the workplace within our EU LACE initiative for learning analytics promotion, I think we must indeed engage machines in our learning goals, and think ourselves at the beginning of such journey, distant from the fears and pitfalls of Mori’s uncanny valley. But the steadiness and pace of our journey relies deeply on our capability to maintain a competitive “learning edge” on the machines, balancing our capability to teach machines and possibly learn with them.
To ensure this, we must think of robots as working companions, making them more and more efficient in chosen tasks, meanwhile leveraging them to gain the freedom to put our time, talent and skills towards smarter, more qualifying and more valuable learning pursuits. In this way, we can retain our superior intelligence, and keep it comfortably out of reach of the singular “rise of the machines.”
At the same time, we must remain acutely aware that the capability of machines to learn will happen at a very fast and cheap pace, and smarter intelligence may not suffice to avoid a complete machine take-over, at least at the workplace…
So, for the time being, we can rest and enjoy the benefit of computational help in our workplaces. Our freed time can possibly be used trying to understand how humans can learn better using machines whilst we move forward.
But…we must also keep a cautious eye out on who will end up better adapting to the demands of today’s workplace to win the learning race and become the superior workforce of the future.
Will it be us or the machines?
 “2014 Future of the Internet” (Pew Research Center, August 6, 2014)
 “Bots in the back office. The coming wave of digital workers” (KPMG, 2016)