Artificial Intelligence is on Everyone's Lips, But Are We Talking About the Same Thing?

In 2017, AI attracted $12 billion in venture capital investment. We are only just beginning to explore the benefits of AI applications. Amazon recently opened a real grocery store, replacing cashiers and checkout lines with computer vision, sensors and deep learning. Between investments, media, and innovation, "artificial intelligence" has become a buzzword. But is there really such a thing?

At the World Economic Forum, Dr. Kai-Fu Lee, a Taiwanese venture investor and founding chairman of Google China, said: "I think it's attractive for entrepreneurs to market their company as an AI company; it's attractive for venture investors to say 'I'm an AI investor'." He then predicted that some of these AI bubbles could burst by the end of 2019, specifically referring to "start-ups that fool risk investors by creating a story that isn't true because they don't know any better".

But Dr. Lee firmly believes that AI will continue to evolve and put many people out of work. So for better or worse, what's the difference between a legitimate AI and a made-up story?

If you look at the few news stories that have ostensibly been made about AI, you realize how vastly different people's definitions of AI are, and that there is a blurred line between imitations of human intelligence and applications of machine learning.

In an attempt to find a consensus, I spoke to experts in the field of AI, but this question alone raised new questions. For example, when is it important to stick to the primary definition of a term, and when does this fidelity become unnecessary rigor? The answer is not clear, and fashion often trumps small details. What's more, there is now a vested interest in this fashion - a $12 billion vested interest, to be precise.

This conversation is also meaningful because world-renowned thought leaders are openly discussing the dangers of AI. Facebook CEO Mark Zuckerberg suggests that the naysayers who are "spouting these doomsday scenarios" are being pessimistic and irresponsible. Influential businessman and OpenAI co-founder Elon Musk countered on Twitter that Zuckerberg's understanding of the issue is limited. In February, Elon Musk engaged in a similar debate with Harvard professor Steven Pinker. Musk claimed on Twitter that Pinker did not understand the difference between limited AI and general AI.

Given the fears around this technology, it is important that the public clearly understands the differences between the various levels of AI and can realistically assess the potential threats and benefits.

As Smart as a Human?

Erik Cambria, an expert in NLP (Neuro Linguistic Programming), told me: "No one is doing artificial intelligence today, but everyone says they are because it's a cool, sexy word that's on everyone's lips. A few years ago, the same was true for the term 'big data'."

Cambria said that AI as a term originally referred to the imitation of human intelligence. "And today there is nothing that even comes close to the intelligence of the dumbest human in the world. I mean, nobody is really doing artificial intelligence yet, because we don't even know how the human brain works," he said.

He added that the term "artificial intelligence" is often used for the powerful tools used in data classification. While these tools are impressive, they are a class apart from human cognition. In addition, Cambria has noticed that people claim that neural networks are part of the new wave of AI. This struck him as strange because this technology has been around for fifty years.

However, technologists no longer have to make inferences about a feature on their own. They now have access to a larger computer capacity. This is all very nice, but it's not honest to claim that machines can mimic the complexity of cognitive processes.

"Companies are just trying to create behaviors that look like intelligence; it's a mirroring of intelligence. These are expert systems, maybe very good in one area, but very stupid in other areas," he said.

This mimicry of intelligence has awakened the public imagination. Domain-specific systems have added value to a wide range of industries. However, these benefits have not cleared up the confusion.

Assisted, Augmented or Autonomous

When it comes to scientific integrity, the issue of correct definition is not to be underestimated. In his famous 1974 commencement address at the California Institute of Technology, Richard Feynman said: "The first principle is not to fool yourself, and the easiest person to fool is yourself." In the same speech, Feynman also said: "When you speak as a scientist, you must not deceive the layman." He said that scientists should go out of their way to show that they could be wrong. "If you present yourself as a scientist, you need to tell the layman what you do and respect their decision if they don't want to support you."

In the case of artificial intelligence, this may mean that professional scientists have to make it clear that they are developing extremely powerful, controversial, controversial, profitable, even dangerous tools that have nothing to do with intelligence as we know and understand it.

The term "artificial intelligence" may have become over-hyped and confusing, but there are some efforts to provide clarity. A recent PwC report distinguished between "augmented intelligence," "augmented intelligence" and "autonomous intelligence." Assisted intelligence is seen in the GPS navigation programs that are common in cars today. Augmented intelligence "enables people and organizations to do things they could not otherwise do." Autonomous intelligence is "machines that can take action on their own", such as autonomous cars.

Roman Yampolskiy, an AI researcher, has written a book entitled "Artificial Super Intelligence: A Futuristic Approach". I asked him whether these broad and different meanings pose a challenge for lawmakers dealing with AI legislation.

Yampolskiy explained: "Intelligence (whether artificial or natural) is a continuum, and so are the potential problems associated with such technologies. To avoid confusion, we usually call general artificial intelligence that will one day acquire the full range of human abilities. Beyond that is superintelligence. What we have today, and what is often used in business, is limited (narrow) artificial intelligence. Regulating the field of technology by law is as difficult as regulating any other field. The problem here is not the terminology, but the complexity of such systems, even at current levels."

When I asked whether people should be afraid of AI systems, Dr. Yampolskiy commented: "As skills are continuous, so are the problems associated with each skill level." He said that they are already aware of accidents related to AI products, and that as the technology advances further, its impact may transcend privacy issues or technological unemployment. These concerns about the real-world impact of AI probably take precedence over its dictionary meaning. But it is also about honesty and deception.

Is This Buzzword Already Obsolete?

Finally, I put my questions to a company that actually markets an "Artificial Intelligence Virtual Assistant." Carl Landers, Conversica's chief marketing officer, confirmed that there are myriad explanations of what AI is and is not.

He said: "My definition of AI is a technological innovation that helps solve a business problem. I'm not really interested in theoretically debating the question, 'Can we build machines that think like humans?' That can be a nice conversation, but I'm trying to solve a real business problem."

I asked him if he thought AI was just a catchy phrase to advertise and attract customers. According to Landers, this was certainly true up until three years ago, but the effect has already started to fade. Now many companies claim to have AI in their products, so it's no longer a distinguishing feature. However, there is still a certain intention behind these words. Landers hopes to send the message that what was previously impossible is now possible. "There's something new here that you haven't seen before, you haven't heard before," he says.

According to Brian Decker, founder of Encom Lab, machine learning algorithms only work to the extent of their existing programming, not out of an intrinsic drive to understand better. For this reason, he sees AI as a purely semantic debate.

Decker said: "A marketing executive might claim that a photodiode-controlled outdoor lamp has artificial intelligence just because it 'knows when it's getting dark', but a good hardware engineer will tell you that not a single bit in the entire history of computing has changed unless the logic of its current programming dictates so."

While it is important for everyone to agree on features and underlying meaning, AI products override these debates by creating immediate value for people. And ultimately, people value value more than semantic distinctions. In a Quartz interview, Kai-Fu Lee said that algorithmic trading systems have delivered 8 times the return of private banking investments. And he said, "I don't trade with people anymore."

ByDavid Pring-Mill

This articleoriginally appeared on Singularity Hub, a publication ofSingularity University.