We live in a time of transition to a new, AI-driven era.
In his book of the same name, Gigaom publisher Byron Reese posits that we now live in The Fourth Age. For Reese, the first age was one of language and fire. The second, architecture and cities. The third, writing and wheels. The fourth age is characterized by robots and artificial intelligence.
Klaus Schwab, the founder and executive chairman of the World Economic Forum, starts the clock later than Reese, but still thinks in terms of four. His book, The Fourth Industrial Revolution, acknowledges the agricultural revolution before laying out four specifically industrial revolutions:
1. Mechanical production (railroads and the steam engine).
2. Mass production (electricity and the assembly line).
3. The computer revolution (semiconductors, mainframe computing, personal computing, the Internet).
4. The Fourth Industrial Revolution (widespread and mobile Internet; small, powerful, and cheap sensors; and artificial intelligence/machine learning).
These authors reflect the sense that artificial intelligence represents a fundamental and historic shift for humanity. Much of this shift will be good. In fact, expect data science will become a sort of 21st-century stethoscope: those who use it may not know exactly how it works, but it will be a fundamental, everyday tool.
However, the rise of machine learning and artificial intelligence is bound to lead to unexpected outcomes. Electricity changed everything and, outside the occasional camping trip, few people today would wish to return to a pre-electrified state. However, electricity also created a plethora of new challenges, from new types of pollution to new ways to wage war — and new reasons to go to war in the first place.
The spread of machine learning in the business world will no doubt have unexpected results of its own. That said, there are few things about the future that I feel pretty sure about, expressed with full understanding that our wise predictions may, in retrospect, appear spectacularly silly.
Sorry, But You Aren’t About to Get AGI
Most artificial intelligence, including most machine learning, is designed to be narrowly targeted to a specific problem. But some dream of a future with artificial general intelligence — AGI — in which AI can do lots of different things. In other words, an AI that is more like the flexible human brain.
However, there is no path in artificial intelligence research to AGI. There are people researching it, of course. OpenAI announced a $1 billion investment from Microsoft “to support us building artificial general intelligence (AGI) with widely distributed economic benefits.” But there is no clear theory to say that once humanity can accomplish X, it will result in AGI.
The idea of AGI is often connected to a fear that these super-smart computers may overshadow or overpower humanity. But Salesforce chief scientist Richard Socher doesn’t buy it. “What some folks like Elon Musk are worried about is this existential threat that AI might pose,” he said on CNBC, “and that is really unfounded because we don’t really have a credible research path right now towards artificial general intelligence that will set its own goals.”
Some have a concept of “imitative AI”: if we can imitate what is physically in the human brain and simulate the number of neurons in it, that will equate to human-level performance. But there’s no indication that just imitating what’s going on in the brain is going to get us to an AGI. Even if you can estimate that, by a given date, we’ll have enough neurons to be able to simulate the human brain, there’s no proof that that will equate to human-level intelligence in software. After all, we can simulate the number of neurons in an ant’s brain today, but computers lack even ant-like independence.
Plus, the best way to create a tool that works for human purposes isn’t always to copy nature. You may have heard the phrase “airplanes don’t flap their wings.” We don’t need to copy the brain in order to make an intelligent, useful tool. Just as humans didn’t use nature’s design to build a flying machine, they won’t need to use nature’s design to build a thinking one. It’s just that for the foreseeable future, those thinking machines won’t think like a generally intelligent human brain.
Are We There Yet?
The authors of the paper “When Will AI Exceed Human Performance? Evidence from AI Experts” surveyed experts who had published at two of 2015’s top machine learning conferences. Their survey asked about “High Level Machine Intelligence”, characterized by a time “when unaided machines can accomplish every task better and more cheaply than human workers.”
Compiling the results, “the aggregate forecast gave a 50% chance of HLMI occurring within 45 years and a 10% chance of it occurring within 9 years.” Interestingly, the survey showed a marked difference in opinion based on geographic region: “Asian respondents expect HLMI in 30 years, whereas North Americans expect it in 74 years.”
To be clear, I’m not saying AGI will never happen. And it’s possible for smart people to disagree about this. Elon Musk has called AI “a fundamental risk to the existence of human civilization.” Facebook’s Mark Zuckerberg says such talk “pretty irresponsible.”
For now, I remain unconvinced that AGI is around the corner. Bold claims about AGI from ten years ago are worth about as much as bold claims from today. In both cases, the evidence of a path from the research has been nil, and nothing in the last ten years has moved the probability any higher.
Seeking out AGI is a misleading paradigm anyway. It’s the wrong target to shoot for, just like “automating everyone’s job” is often the wrong target to shoot for in a machine learning project. It is more productive, especially for today’s business leaders, to focus on the fact that computers are not human, and that they can thus complement what people can do.
Maybe in the distant future humans and computers will merge — whatever that means — but right now that day is far away. We’d prefer to focus on making computers and humans good at what they respectively do best, and the great things they can accomplish when they work together.
Machine Learning Will Continue to Make Impressive Advances
Ok, so you’re not going to have your own personal version of Mr. Data to help you at work. But in just a few years, machine learning will transform large swaths of human labor for the better:
- Any job that requires quick visual inspection will be automated. Humans can spend less time figuring out what something is, and more time figuring out what to do about it.
- Any job that requires rote document analysis will be automated. Human brains will be put to a far better use in analyzing anomalies and determining higher-level strategy based on what’s in the documents.
- Complex pattern analysis will get better and better, enabling advances in areas from medical diagnostics to marketing insights. In the words of authors Peter Keen and Ronald Williams, “[t]he transformative potential of AI comes from its breaking some limit on Pattern-building and application.”
It’s worth expressing one more time: there are plenty of ways to advance the business world just by applying what machine learning can do right now. But the research isn’t going to stop. The cutting edge will continue to advance. And machine learning will totally transform the workplace and the world.
Robbie Allen is a Senior Advisor to Infinia ML, a team of data scientists, engineers, and business experts putting machine learning to work.