'Human Computation' Could Save the World Without the Risks of AI

Forget artificial intelligence: The key to solving the world's most complex problems could be human-machine collaboration.

That's the rallying cry of researchers who penned an editorial in the journal Science championing "human computation"—systems that combine the talents of computers and humans. The authors claim these systems could ultimately tackle issues such as climate change and geopolitical conflict, all without the existential risks posed by true AI and the technological singularity.

Authors Pietro Michelucci and Janis Dickinson imagine a system that would provide a technical framework for ideas to be shared, analysed, and revised until the best bubble to the top; Michelucci envisages it as a "dynamic Wikipedia." The idea would be to develop our understanding of real-world issues online, and test potential solutions in this computational space, then applying new knowledge back in the real world so as to actually effect some change.

"Imagine something like the game SimCity, but a thousand times more detailed, and then link in real-time sensors attached to the internet," said Michelucci. "The more faithful that model of the real world becomes, the more accurate it would be for testing out solutions and predicting outcomes."

Michelucci is the director of the US-based Human Computation Institute. Dickinson, as the director of citizen science at the Cornell ornithology lab, is involved with an existing human computation project to gather information on conservation efforts called YardMap.

Human computation is rooted in the fact that computers and humans bring different things to the table: computers can quickly analyse huge amounts of data, but humans win when it comes to proposing new ideas. In their article, Michelucci and Dickinson use the existing Wikipedia model as a well-known current example of human computation: Many humans work together with a computer platform to produce a single output with an accuracy that's perhaps surprising given the frequent disagreements inherent to the collaborative nature of the work.

But this system of simple crowdsourcing, or the kind of "microtasking" in citizen science projects that asks you to, say, count penguins in photographs of the Antarctic, won't be enough to take on more nebulous problems like climate change. That's where the "dynamic Wikipedia" comes in: it goes beyond a database of documents to act as a kind of simulation of real-world scenarios.

To give a practical example of how it could be used, Michelucci suggested that "an engineer could go in and explain how an engine works, and then actually have a working engine simulated in this new dynamic Wikipedia. That would allow people to come in and pour new fuel combinations that they describe chemically into this simulated engine and see what the output of that is."

Like Wikipedia, the system would aim to attract contributions from millions of individuals with their own specialist knowledge, from schoolchildren to working scientists to retirees. But rather than just pool their ideas, the systems proposed by Michelucci and Dickinson would combine human cognitive power with computer algorithms to create a feedback loop that promote constant evaluation and revision of ideas.

The advantage of human computational systems is that we already have the fundamental technical infrastructure that could potentially support these kind of projects. Michelucci said he got into the field after seeing the "incremental" progress made in AI.

"In a way, human computation is like cheating at artificial intelligence," he said. "It's sometimes jokingly referred to as 'artificial artificial intelligence,' because what we effectively do is take an AI algorithm and say: this is the hard part that we can't do with computers, so let's farm this part out to a human. It's really like saying we can create the sort of artificial intelligence we imagine for the future today, just by building humans into the system."

Given the strong role of humans in the system, there's also less risk of the machines wielding the kind of humanity-threatening power Elon Musk and Stephen Hawking fear, should they eventually reach that level of intelligence. For Michelucci—who says his motto is "always build in a kill switch"—the key is to keep the overall goals in the control of humans.

"I think human computation provides an alternative vision of the future, one in which there's increasing integration between humans and machines," he said.

It's not foolproof, of course, because not trusting machines with power means trusting humans with power. In their article, the researchers note that human computation systems could be misused by malicious actors. "It is important that nefarious uses, such as disinformation engineering, in which human computation systems are designed to incite panic, steal information, or manipulate behavior, are not overlooked," they write.

Michelucci now wants to see human computation systems develop further beyond the traditional crowdsourced citizen science model to get to a point where they can really tackle those global issues. We can't blame the bots: If we mess it up, it's our own undoing in every respect.