We are experiencing the biggest disruption in business history since the introduction of computing and the internet. AI is broadly defined as machines that can perform tasks that normally require human intelligence, such as visual perception, speech recognition, and complex decision-making. This definition, however, is not yet complete; the term will further evolve as we push the boundaries of cognitive computing and consultants race to invent new buzzwords. Meanwhile, CEOs and their executive advisors are thinking and talking about how they are using AI to create new competitive advantages. In just the last 10 days, two CEOs of two large global companies were telling me their companies’ future is all about AI. When I politely asked why and how, I was met with silence.
Machines have become dramatically smarter and more complex. Cognitive computing algorithms have enabled machines to learn abstract concepts efficiently, without human supervision. They can now learn to read X-rays, maps, and facial expressions, and understand voice, tone, and context. With almost two quintillion bytes of data being created every day, we are powering up more empirical and real-time patterns. We are pushing machines to do tasks which managers or professionals should be doing.
Many people—including the media—are quickly jumping to conclusions about how we will be getting “intelligence in a box,” whether these devices are helping patients, doctors, law enforcement professionals, or white-collar workers across a wide range of industries. The “box” idea, however, is not the right metaphor—far from it. It won’t be a box, but a stack of technologies connected to thousands of sensors facilitating a system that learns about something and creates its own rules. This “black box” idea is wrong because it’s very dangerous to leave such things to a “box,” with very little understanding of what it’s actually learning and the rules it’s creating.
The long-standing debate about AI systems as complements or substitutes for human labor is not relevant. It will become part of our intelligence system, affecting enterprises and individuals, managers and consumers. One thing is for sure: many jobs will disappear and many new ones will be invented and hopefully soon enough.
The biggest challenges for AI include:
Maintaining a Degree of Independence: How can we ensure that a given AI system respects aspects of costs while pursuing its mission? When does it check and confirm a situation with a human? How many “no’s” does it need in order to accept the command is indeed a “no”?
Seeing The Forests: How can we ensure that an AI system will understand that certain decisions have long-term negative implications, even if it’s the fastest way to pursuing its immediate goals? This applies to everything from healthcare to environmental and social side effects.
Multi-Stakeholder Risk Management: How do we ensure that an AI system doesn’t take unnecessary risk when pursuing its mission and the risk assessment is reflecting the level of risk/reward that’s implicated by the human?
Managing Ethical Boundaries: How do we ensure the machine is learning the ethical considerations that we have as human beings, and how do we ensure they are not misinterpreting our past ethically driven actions?
Rules Extraction and Management: How do we ensure we can edit or suppress certain rules that the machine is creating, and how do we validate them so that the machine learns in the direction that we want them to? These rules need to be explicit and cannot be hidden deep within the black box.
Solving Data-Hungriness: Without data, there is no learning; it’s like a rocket with no fuel to burn. Most AI experiments today either run on not having big enough data sets, or the wrong dark data. Even if this dark data becomes visible, it doesn’t mean it’s usable. The problem is not only solving the data problem, but also designing our deep learning systems to be more data efficient and able to work with less data.
I believe senior managers are far from obsolete. As machine learning progresses at a rapid pace, C-suite executives will be called on to create innovative new organizational design that allows people to co-decide in strategic matters and co-manage enterprises. The battle is not only a technical one; it is not about using various forms of AI or machine learning to handle some abstractions, deductions, inferences, and analogies such as these Machine Learning examples: DeepStack, Libratus or Deep Symbolic Reinforcement Learning.
Given that it is estimated that there will be around a million job opportunities in the US alone in advanced computing, it is not a demand we can fill easily. The race is on. A few companies will pick up 40-50% of the top talent, but most enterprises will not be able to attract any talent at all. They will simply have to sit back and watch what’s coming next.
Read the original post here.