Welcome to TechTalks’ AI book reviews, a series of posts that explore the latest literature on AI.
In recent years, advances in neural networks and deep learning have triggered a revolution in artificial intelligence. The AI industry has made huge progress in solving complicated problems such as predicting cancer and driving cars.
But AI also faces various challenges. Today artificial intelligence is in the greatest hype cycle of its history and has become the subject of a mixed jumble of hype, excitement, fear and resentment.
Many of these challenges are at least partly because of a wrong perception toward AI. Most of the effort that goes into AI innovation and engineering focus on the full automation of tasks. Meanwhile, many companies, marketers, investors and media outlets try to portray artificial intelligence as technology that is meant to rival and replace human intelligence.
All of this is creating an erroneous perception of AI, which entails the wrong expectations and fears about the industry. Alex Bates, AI researcher and managing director of NeoCortex Ventures, discusses in his latest book, Augmented Mind: AI, Humans and the Superhuman Revolution, that we need to view AI in a different light. Augmented Mind explains how the real opportunities of AI lie in augmenting the capabilities of humans, also known as “augmented intelligence,” and provides clear examples where the combination of AI and human intelligence becomes much stronger than the sum of its parts.
In an interview with TechTalks, Bates discussed the challenges of the AI industry, augmented intelligence and his book.
The general confusion caused by the AI hype
Bates splits the structural challenges of the AI between outside and inside the industry.
“One problem is that things get overhyped in the media,” Bates says. “On the outside, people get caught up in hype cycles and what tends to happen with journalism is that there’s so much pressure on traditional journalism right now that writers basically have to write a clickbait headline to get people to click on it because that’s how they monetize.”
There are plenty of examples of this. In February, research lab OpenAI revealed its new text-generating AI model, but refrained from releasing it, fearing it would unleash a new fake news crisis. The decision caused panic, and created the best breeding grounds for a large crop of clickbait articles and overblown warnings about fake news bots.
In another case, last year, Facebook shut down an AI model that had deviated from its script and had developed its own way of communicating between agents. The media quickly picked up on this and generated plenty of sensational articles. One Forbes piece, titled “Facebook AI Creates Its Own Language In Creepy Preview Of Our Potential Future,” was viewed more than 500,000 times.
But there are other problems as well, Bates points out. “AI, like intelligence, is not a binary concept. It’s a spectrum, and with any spectrum thing, people almost throw up their hands and say, all of AI is fake and there’s no such thing as AI, because you have a toothbrush that claims to be intelligent,” he says.
As we’ve previously seen in these pages, artificial intelligence is a moving target, and its definition and context shifts in tandem with technological advances, which is why, as Bates explains, AI should be viewed as a spectrum. In this light, even a toothbrush can manifest intelligence, even if it’s extremely low levels of intelligence, Bates says.
“Even rule-based systems can show signs of intelligence. So to me, it’s important to not just give up and say there’s no such thing as AI. I’ve seen people that have this resigned attitude and that’s where I think skepticism goes too far,” Bates says.
The rivalry between AI companies
Adding to the overall confusion of the industry is the tendency for tech companies to “invent” new AI technologies and to set themselves apart from their competitors.
“For people inside the field of AI, one thing I’ve seen is, there’s this tendency to always coin new terms, because no one wants to say I agree with so and so and I’m doing the same approach,” Bates says.
The history of AI is marked with a vast array of different terms, many of which hint at the same technology. “Artificial intelligence” serves as an umbrella term for all these different technologies and names. This helps unify and simplify the communication of AI developments to people who are outside the field. But it also creates confusion, because there are fundamental differences between the problems that each blend of AI can and can’t solve.
When it comes to augmented intelligence, the problem is even more complicated. “For augmented intelligence in particular, we have this huge tower-of-babel fragmentation problem,” Bates says, adding that there are dozens of different terms that basically mean the same thing, such as “human-centered AI,” “ extended intelligence,” “ human-centered machine learning,” “mixed initiative systems…”
“There are probably 30-40 different terms and more come out every year. This is a separate problem. We have one term ‘artificial intelligence,’ but we have no term for this hybrid concept,” Bates says.
The real opportunity of AI is augmenting humans
One of the main themes in Augmented Mind is that the biggest opportunity in AI technologies is augmenting humans. This is one of the underserved areas of AI, Bates explains, because research and development are mainly focused on “pure synthetic or replication intelligence,” where humans are an afterthought.
As Bates points out, most machine learning applications aim to fully automate a task by gathering and labelling data, and then creating an AI model that can use that data to perform the intended task with acceptable precision. In this context, humans mostly serve to curate better labels for the AI system until it can perform its tasks without their help.
“You can do that and get some degree of success. But you should also look for underserved areas. If you approach AI differently, if you look at the people in the work process and how we can augment them, what are their strengths and weaknesses, what are their current challenges, and how we can build the AI tech around that perspective, I think that’s the big opportunity,” Bates says.
Bates points out that AI technology should free people from the menial, low-level tasks that take up a large part of their time, and give them work environments where they can be more creative.
“When you study the human cognitive limitations, we have these amazing gifts. But we also have these known limitations. Our working memory is about seven chunks. So when we’re setting a goal to come up with a creative solution to a problem, we’re using these seven chunks of working memory. It’s extremely limited. And then on top of that, we have distractions popping up in our environment,” Bates says.
With augmented intelligence, humans will be able to enhance the capacity of the brain by leveraging the computing power of AI systems. This will help them come up with solutions that weren’t possible when their entire cognitive power was focused on solving problems that can be deferred to an AI assistant. As a testament, Bates points out to the many scientists who had epiphanies and eureka moments when they were in uncommon situations, like taking a bath or wandering in a forest. When the mind’s cognitive load is reduced, it can start thinking in new ways that were previously impossible.
In Augmented Mind, Bates lists many fields where the collaboration between AI and humans can yield tremendous results. Some of these areas include mathematics, finance, health care, arts and music, and games like chess.
Bates names these combinations of humans and AI “centaurs.” Centaurs are mythological creatures that are constituted of the upper body of a human with the lower body of a horse. This composition combines the intelligence and dexterity of the former with the strength and the speed of latter.
After a fashion, this is the perfect way of describing the combination of human and AI. Humans have abstract thinking and general problem-solving capabilities that no computer has been able to replicate yet. On the other hand, AI technologies can leverage sheer computing power and search through massive amounts of data that outmatches the most brilliant humans.
“In the chess centaur example, a term that was coined by Kasparov, the centaur team started to dominate. One way to think about that is that we have super-human chess today. So we have augmented ourselves with AI, and the team component is greater than the sum of its parts. That team has achieved a higher level of intelligence,” Bates says.
Bates acknowledges that the chess centaur example is a bit contested, and that some people rightly believe that AI will eventually dominate all games without the need for human assistance.
“I agree that in the circumscribed realm of games, team AI is going to dominate. Our human gifts of embodied cognition and intuition aren’t quite as relevant in these ideal, simplified, circumscribed gaming realms, and I think pure synthetic AI will dominate those realms,” Bates says.
But when you get out of games, into the extremely messy and nonlinear real world that we live in, creating an AI system that can factor in all the different complexities and interactions is virtually impossible.
“This is where our human intuition sense does amazingly well. That’s the key area where the centaurs are going to reign. There are lot of the examples in the book, things like health care, where it’s not a gaming environment, it’s the real world. And that’s where the real opportunity is,” Bates says.
Is AI unemployment a real threat?
On the sidelines of our discussion, I asked Bates about his view on the ongoing debate on whether AI will drive humans into mass unemployment. In the past years, as AI technologies have become increasingly efficient at solving complicated tasks, there’s fear that one day, automation will take away all jobs from humans. Many different AI experts and tech thought leaders have weighed in on the topic and are exploring different solutions such as reskilling programs, basic income and robot taxes.
But to Bates, the history of humanity shows that those fears are a bit overblown. “Humans and homo sapiens were tool users and tool creators, so we’re always going to be creating tools and it’s going to be newer and better and faster ways to do things. And people adapt. We’re extremely good at adapting, that’s part of what really underlies our human intelligence,” Bates says.
A famous example is the farming industry. In the 1800s, more than 90 percent of the American population were farmers. Today farming accounts for less than 2 percent of jobs, even though the population has grown and farming yield has multiplied. Technological advances have helped automate many of the farming jobs that previously required human labor. But the people who are no longer engaged in farming haven’t been driven into unemployment. As the industrial revolution destroyed many jobs, it created many others.
AI is another one of those revolutions that will create a shift in the way we do things.
“AI is a tool. Certain things will go away, but people will be very good at adapting,” Bates says. “It’s actually going to be very empowering—I call it the augmented age—but this next chapter where AI starts to augment and improve our lives and make us do less of the soul-sucking, low-level work that most of us waste time on, and focus more on the creative and fulfilling things that we’re passionate about. There will probably be a couple of rocky transitions, and it’s going to take some experimentation with things like basic income. But in general, I think we’re going to move beyond the 40-hour work week. I think it’s really an exciting time.”