I recently finished teaching a free, eight-week online course on artificial intelligence and cognitive technologies. Over 7,000 students registered for the course. Many of them were active participants in rich discussions throughout the eight weeks. I'd like to tell you what I learned from my students during the course, and invite you to register for the next run of the course, which starts on March 14.
What we covered
Over the eight weeks of the course we covered major categories of cognitive technologies including
- Cognition, including knowledge representation and reasoning and planning
- Learning, including supervised, unsupervised, and reinforcement machine learning
- Perception, including computer vision, speech recognition, and natural language processing; and
- Robotics
Although my production team and I recorded the lectures in advance, there was lots of interaction during the course as students interacted on the discussion boards, completed multiple-choice assessments after each section of the course, and many undertook optional assignments with team members.
I enjoyed engaging with the students on the discussion boards. They posted lots of questions and comments, and even answered each other's questions. Students were fascinated with the ways in which perceptual technologies like speech recognition and computer vision make possible natural interfaces between humans and computers. And they actively debated the implications of cognitive technologies for businesses, workers, and society.
To augment humans, or to replace us?
Of all the topics we covered in the course, one topic provoked more discussion than any other: the impact of cognitive technologies on work and workers. One student saw in the rapid adoption of cognitive technologies a "redefinition of ... the Industrial Age." He predicted that these technologies would displace many jobs. Other students had a more positive outlook on the future of work. They were confident that cognitive technologies will lead companies to invest in new business opportunities, creating new jobs and demand for new skills along the way.
A number of the students expressed concern that we would become too dependent on advanced technologies, allowing our skills to atrophy. Once cited the example of commercial airline pilots whose reliance on autopilot is said to slow their reaction time and reduce their situational awareness, creating greater risks in the event of emergencies.
One student, citing the potential for cognitive technologies to influence medical diagnosis and treatment recommendation, said that American doctors need to "become more like [their] European counterparts and develop cross-disciplinary...skillsets." This way they could take advantage of advanced technology to augment their abilities while doing what humans still do better than computers: thinking broadly. Many students came to see cognitive technologies as an aid rather than a replacement for people.
What businesses should do
In our discussions on the impact on business, students mentioned tangible benefits like lower labor costs, higher profits, greater efficiency, and enhanced business decision making. One felt that any company not experimenting with cognitive technologies is at a "huge risk" of being left behind.
We talked about how organizations should prepare to make use of these technologies. One student asserted that these technologies should not be adopted for their own sake but rather "should be evaluated in terms of solving a business problem." In the course I offered a simple three-part framework for performing this kind of evaluation that I call the Three Vs: viable, valuable, and vital.
Another student wondered about how employees would react to new technology. To gain buy-in, some students suggested, companies should explain how cognitive technology would ultimately benefit employees.
A system of checks and balances
A number of my students expressed concern about how companies will manage information privacy in an era of cognitive technologies. Already, retailers are struggling to balance customers' privacy requirements with their demands for highly personalized shopping experiences. Medical diagnostic and predictive tools supported by cognitive technologies--particularly machine learning--require access to large volumes of sensitive patient data. How do we make sure that our data is protected?
Some students articulated the view that new regulations and government oversight over the use of cognitive technologies are required. One student suggested that deployments of cognitive technologies require specialized quality assurance, possibly provided by an "independent third party, similar to the auditors of today." In my view, it's a bit early for imposing new regulations on the field--it's just changing too rapidly. But it's not too early to think about the issues that regulations might cover and how we address them. Regarding quality assurance, one thing is sure: new technologies fail in new and unexpected ways. Organizations building and deploying these technologies will need to consider how their testing and risk management procedures need to evolve.
Everyday applications of cognitive technologies
There was a lively discussion on the boards about how different industries are using cognitive technologies. There was discussion about how AI is enhancing fraud detection and algorithmic trading in financial services because "humans can't keep up with the speed of transactions." In the field of medicine, students recognized the pros and cons to doctors working in tandem with medical diagnostic and predictive health analytic tools. In the mining industry, AI-driven trucks are starting to appear.
For marketing and e-commerce, one student pointed out that AI has already created new marketing opportunities by making extreme personalization possible. Instead of dividing customers into segments, marketers are able to use advanced algorithms to deliver "the right message, at the right time, on the right device, with the product offer" to an individual shopper. Another student wondered if virtual personal assistants like Cortana or Siri could become the preferred recommendation engine for marketers.
How smart?
Amid the optimistic discussions, many students commented on the limitations of cognitive technologies. One student asserted that this technology will only be as good as the "are only as good as the hypothesis, models and assumptions" given to it by humans. Another student thought that neural networks have made impressive advances (e.g., Gmail offering automated responses to questions and statements in emails), but the technology in general is still very raw and unsophisticated. A student whose mother tongue isn't English explained that while natural language processing works well for English, it may struggle with other languages.
A number of students were captivated by the question of whether computers are going to need to be able to make ethical decisions. What should an autonomous car do, for instance, if faced with a choice of harming a pedestrian or passenger in the car?
Toward the end of the course there was a real mix of view among the students about the ultimate impact of cognitive technologies. Many recognized that these technologies were going to have a profound impact on work and society. But they also recognized that these technologies have clear limits today. They are getting really good at solving narrow classes of problems. But they are still the work of our hands and minds. And because of this, they are flawed.
Taking this knowledge into the future
I think most students came away from the course with a deeper understanding of these technologies and their potential impact on organizations, workers, and society. What's more, they are now better equipped than before to make smart decisions where, where, and how to apply these technologies, in their organizations and in their personal life.
We're running the course again starting on March 14. Why not join the next class, and register for the course today.